Posts by CheesyTheClown

Intel CTO and AI boss quits to join OpenAI after just six months in the job

What an amazingly bad idea by OpenAI

As a person who regularly investigates accelerators for AI, Intel is the most opaque company in the AI industry.

So here's how it works.

1) Student at the university is studying AI

2) Student is given a project and investigates the tools available to them to implement it.

3) They settle on PyTorch or another toolkit for their project

4) They start coding on their laptop or gaming PC

5) They request resources from the university HPC department

6) They complete their project

7) They enter the industry

8) They prefer to use the tools they studied and develop systems professionally as they did in the university

This has been the story of all Intel failures over the past several years.

Intel released piles of new technologies. Some of them far better than their competitors and able to accelerate some tasks by 1000x or more over using general purpose computing. These technologies were absolutely amazing. And they ALL FAILED.

The reason is simple.

University students, hobbiests, researchers, etc... don't have access to Intel's hardware and technologies.

But wait! Intel lets you download the software! That's good right?

Well, the software is useless without the hardware.

Here's where AMD killed Intel.

Whether you're coding for a laptop or a super computer, coding for AMD CPUs is the same. Of course, it's not nearly as amazing as using specialized CPU cores or fancy encryption engines and whatnot. But, the students have access to the technology they need wherever they are. In fact, coding for an AMD super computer can be easily achieved from a Windows WSL2, Mac or Linux session on an old laptop running an Intel chip. Coding against an Intel super computer will not be nearly as optimal because Intel forsaked high performance HPC cores for specialized cores that no one codes for.

Here's where NVidia absolutely destroyed Intel.

Every kid these days has access to a CUDA capable GPU either in their laptop or gaming rig. Coding for an HPC loaded with NVidia H200's (I'm using one with 300 of them right now), is precisely the same as coding for my laptop which has an old RTX4060 in it. The laptop also has Intel Iris XE graphics. But, if I want to code for Gaudi, I can't use the Iris XE for that.

So, that brings us to developer support.

Intel has absolutely no interest in working with the little guy. In a perfect world, what intel would do is sell a student kit containing a ARC B580 12GB for a deep discount and ensure that all Gaudi development tools worked with the card as a first class citizen.

Let's say this, I'm about to spend about $200-$300 on a used NVidia card. On that card, I'll focus my AI development using CUDA and NVidia development tools.

If I could buy a brand new ARC B580 with similar performance to the used NVidia card and I could run the Gaudi development tools on it, I would consider that instead.

Another option, I just looked at Intel's developer support for accessing Gaudi in the cloud.

1) It's offered by IBM/Redhat. There is no way in hell I'm going to build on that platform. I'm not a bank. Only banks and major financial institutions buy IBM stuff. That's not for the small-fry.

2) I couldn't afford it anyway, I followed the link and apparently, it's like $43000 a month. That's not a made up number, it's what the website for Gaudi on the IBM cloud says.

In fact, here's a secret, when I have access to NVidia pretty much everywhere and I know that NVidia works and is proven and I can get support from just about everywhere, I'll drop money on NVidia. But Intel is unproven. At the HPC conferences when you ask Intel about Gaudi and are even willing to accept less performance to see if building systems on it could have potential, the sales people either ignore you or treat you like you're pathetic if you want to even look at their pathetic solution.

See some of us only have $10 million or so to spend on AI accelerators per year. With that budget, you have to wait 24 months for delivery from NVidia today.

But if Intel has a stockpile and capacity to build more, giving them $10 million today can get you cards tomorrow. They apparently are only as fast as the card NVidia can deliver today as well. They don't match the cards you order today and receive in two years. But in two years, I'll have another $10 million to spend and would prefer to not wait two years for whatever the new stuff from NVidia is them.

Then there's the other HUGE, MASSIVE, IRREDEAMABLE failure from Intel.

If you ask for information on Gaudi, rather than asking "What's your address? We'll send you an presales team!!!", they tell you "Talk to our partner Arrow".

Seriously!!!!! WTF?!?!?!?!

It took NVidia 3 days to send a team to me in Norway to have a discussion. A sales person and two engineers where there in an instant.

The funniest thing was, I almost begged Intel to talk with me. They insisted I had to talk to a company I have no relationship to. Not only that, I can't order from Arrow. We have framework agreements which last years and Arrow isn't one of them. I would be asking Arrow to invest time and money into making a sale for Intel even if they had no possible opportunity to make money from it. Do you think Arrow would want to talk with me based on those conditions? Not only that, but Arrow certainly doesn't keep a team of skilled Gaudi engineers in Scandinavia.

So, when Intel says that they are having problems selling Gaudi, let's make it clear.

Their problem isn't finding customers. Their customers are happy to try and find them. Their problem is telling the customers to piss off.

So, Katti might be one of the worst procurements in the history of humanity. He has absolutely no idea about technology. He has no clue how technology works. He basically took one of Intel's most promising projects in decades and destroyed it.

Intel, if you're reading this (I'm sure at least one person there is), get your shit together. Start seeding the developers. Get the students up and running. Get the tech in the hands of the people who are interested in trying it. I have no idea what it costs you to actually make a Gaudi 3 card, but make a pile of 5000 of them and send them out as Christmas gifts to absolutely anyone you think should try them.

BTW, I have a Intel Xeon Phi sitting on display on a shelf in my office. This was one of those gifts from Intel... they showed up at my door and said "Here, try it" and my colleagues and I did. We then build a super computer on them. Then just as all the students who used that HPC and the Xeon Phi graduated, you pulled the damn plug and they had to learn NVidia instead.

Senate report says AI will take 97M US jobs in the next 10 years, but those numbers come from ChatGPT

Need a job? replace a human

If you live in a society where AI and robots are replacing humans left and right, the obvious job to make money is to hear people say “This job can’t be done by a robot or AI” and then build a robot or AI to do it.

We’re heading that way. I’ve been actively identifying things like that. For example, network cabling. This requires a tremendous amount of agility that is almost certainly beyond the means of robots for the time being. So, the trick is to change network cabling to be robot friendly.

I also see a lot, so making robot arms and training them to perform for sewing tasks is obvious. Five years ago, it would have been impossible. Now, using two Hugging Face $100 robot arms, most tasks are trivial.

Slow down on building power plants for all those new AI datacenters, report warns

Peek power soon

AI is growing massively. It’s a race to see who can be the first… at everything. It’s coming at the cost that we’re throwing unlimited processing power to training and much more than that to intuition.

The telephone I’m typing this on is an iPhone 15 and based on the Top 500 supercomputer list, it would have been the most powerful computer on earth in 1999 or 2000. For a similar computer in 2000, it would have required special buildings, massive cooling systems and nearly an entire power plant to operate. Today, it requires a tiny battery to run for a day.

As we optimize code and build specialized semiconductors for processing and intuition, we should expect a minimum of 6 orders of magnitude reduction in power consumption. Assuming our AI grows 3 orders of magnitude in scale, we should still require at most 1/1000th the power used for AI today.

IBM Cloud login breaks for second time this week and Big Blue isn't saying why

Re: Ah, the Cloud

Uhm…

You mean, so long as we have to deal with the people running the businesses we’ll have to do the business stuff?

I agree wholeheartedly. IT would be so much better if we did t have to facilitate stupid things like running companies. We should find a way to get rid of the operations aspects of business. We did that in the dot-com boom and it was really fun!

PCs are basically dumb terminals in 2025. Linux is perfectly suitable for the desktop. People generally won’t complain. The core issue is mostly that the majority of Linux friendly laptops look like shit. And pen support on Linux is horrible. And my god, Linux seems like it will never get any decent graphics tools.

VMware revives its free ESXi hypervisor in an utterly obscure way

Not a good option for much

I don’t know if people remember, but virtualization was about cost savings. The whole point was, you could replace a bunch of basically cheap servers which cost a lot to manage with a beefier server that was cheaper to manage.

Here’s the catch. We used to have servers that cost between $500 and $2000 and we would run 10-20 of them. You had Active Directory on a few… after all, AD was a database that stored a few megabytes of data. Two dns servers, a database typically a few KB to a few MB. We had three SQL servers. A few web servers and such.

These days, to achieve full performance, you can easily run everything but the database servers on $300 Latte Panda single board computers with a few SSDs.

Oh, and except for Windows Update and maybe a file server, 100Mb/s networking is good enough and splurging for gigabit is a no-brainer.

To run a similar system virtualized system requires a LOT more. In fact, even using free virtualization options, the prices are 10-50 times more expensive. It’s a bit more work to go bare metal, but the hours involved costs peanuts by comparison.

These days, since Active Directory and Exchange have moved online, I use strictly $150 single board computers running Ubuntu and Linux containers to run everything. Paying for developers is expensive, but our hardware and software costs are next to nothing. Also, since we script everything using cloud-init and Ansible, we have can fully redeploy our entire enterprise IT in minutes in a disaster recovery scenario.

And if you’re wondering, the bank I worked at in the early 1990’s handled over a million customers a few hundred bank branches, a few thousand ATMs, paper check scanners, etc… on a mainframe roughly as powerful as a Nintendo DS.

When you invest in building your IT systems properly, which means using web browsers as dumb terminals and servers as online transaction processing systems, then hardware is dirt cheap and extremely easy to maintain.

But, I do maintain an old cluster of servers. These days, 40 cores, 256GB of DDR4 ECC, 40Gb/E costs about $500 on eBay. I use Cisco blade servers because I think they’re good value for the money. But those machines are generally only there for compiling C and C++ legacy systems. We have automated powering them off when not needed.

TSMC blew whistle on suspected verboten exports to Huawei – that may cost it $1B+

Re: Question...

There is a huge amount of equipment in the fab process. The lithographic step is just one part of it.

Unlike China who is rapidly working to remove all dependence on outside technology for advanced semiconductors, TSMC, Intel, Global Foundries and others depend on silicon ingots from one place, air purification systems from another and so on.

The reason no one country can produce the technology outside of China is mainly because of patents. However, the Chinese are exempt from patents because if people won't license tech to them and accept fair payment, then they can just make it themselves and happily not pay the companies who are refusing their money. Remember, you can't pay someone who refuses to take your money.

This will cause a massive backlog. And if the US ever lifts export restrictions to China, it will be impossible to ever recoup the lost money from western companies refusing to be paid. And since the majority of the costs involved in semiconductors is patents, China will produce competitive products at substantially reduced costs. Not a single company in the codependent environment of semiconductors will be able to financially compete against China as this progresses.

ASML makes hay while Sun shines, but Trump could rain on its parade

ASML can sell to China or sell to no one in a few years.

So long as ASML sells to China, China will rush less to compete with them. One way or another, China will get the technology. For ASML, the best way is that China buys it from them. The other way which would put ASML out of business would be if China makes their own equipment because they couldn't buy ASML. Then China sells their equipment to everyone else at prices ASML can't compete with.

The result would be that the whole world would be dependent on China and China could withhold their technology from the west.

If a single EUV lithography system costs $100 million and to run an effective fab, you would have 10 or more of them, and China comes to me offering competitive technology for $50 million instead, government rules or not, I'd buy Chinese.

If Deepseek R1 should have shown us anything, China can catch up. If you try to stop them, it will make them work harder and faster.

Altman reportedly asks Biden to back a slew of multi-gigawatt-scale AI datacenters

Energy waste in this scale

The law of conservation has a lot to say about this.

Energy is extracted either from fuel or collected from natural resources such as wind and solar. It is the transmitted at a loss over high tension cabling across sometimes long distances where energy is lost. Of course lost means that through radiation (not nuclear) and other methods, it’s released into the environment as heat.

Then it’s fed into cooling systems which don’t actually cool, but instead move heat from hot places to cooler places. The heat being moved is radiated by semiconductors and of course resistors.

Energy cannot be created or destroyed. This is a law, not an idea.

If we produce gigawatts or terawatts of electrical energy, it is released into the environment and more importantly the lower atmosphere.

There really isn’t such a thing as green energy, especially at this scale. There is only greener energy.

We need to realize that 10 years of additional R&D is a much better investment than more energy sources. For example, if we build quantum dot based semiconductors and move from electronics to photonics in semiconductors, we can fit super computers that fill data centers now into telephones. While I LOVE OpenAI products, even though I’m alread 50 years old, I can wait 10 more years for a better solution.

OpenAI should also be fined for not increasing efficiency enough. Their algorithms are possible a dozen orders of magnitude off of where they could be. Also, AI GPUs from NVidia are insanely wasteful. We don’t need GPGPU circuits to run graph based neural networks.For a simple hack we should be building NPU cores that closely mirror FPGAs. Make a huge sea of low precision FP cores connected by and gates and flash cells which connect neighboring cores based on models.

We have such a long way to go and telling OpenAI that they have to improve their algorithms is a much better solution than giving them more power.

The definition of an AI PC is now even muddier, helping no-one – not even AIs

Re: TOPS

It’s a pretty arbitrary and irrelevant measure though.

For inference, being able to rapidly traverse a neural network is what is important. So a special purpose LLM oriented ASIC with let’s say 32 tiers of register oriented memory (think d-latch 4-bit registers) but extremely wide with a massively wide multiplex between each tier might have a pathetic theoretical TOPS, but would destroy every alternative general purpose inference engine as it would allow FPGA style representation of a neural network. This would make it probably the most powerful inference engine on earth, but would not meet minimum requirements for any “AI PC”.

We need a measure of “inference operations per second” with a depth and width variable.

US government reportedly ponders crimping China's use of RISC-V

This is cute

So, first of all, RISC-V is shit. The vast majority of the capabilities of a CPU come from a unified ecosystem. Instruction sets are just a compatibility layer, but if you want a high performance system, the compilers, operating systems bus architectures, memory architectures, and the CPU cores have to be designed almost as a single component. This is why code compiled for a specific AMD CPU performs substantially better than code compiled just for x64. It’s also one of the key reasons Apple ARM chips perform so well compared to other ARM chips, Apple tunes the tools to the CPU.

RISC-V is sexy because you can design a CPU to the spec and get a generic set of tools which work on it. But code will never be nearly as efficient as an across the board design.

If someone were to ask, what would it take to build an alternative to RISC-V, the answer would be “buy a CPU architecture book, design an instruction set, the build a binutils and gcc (or llvm) for the architecture and port an OS. This is shockingly easy to do. You don’t need the US to be involved and honestly, the only reason most projects like this fail is because they lack momentum.

China could very easily make their own ISA. If you want to cut them off; the big hit would be attempting to limit their access to FPGAs which oils take substantially longer to recover from than even semiconductor limitations.

Still no love for JPEG XL: Browser maker love-in snubs next-gen image format

Cool... but

JPEG is a very simple format with the exception of the headers (more accurately the trailers) which are a nightmare. It's a simple early DCT based encoding based on 8x8 macroblocks. Implementing a cleanroom JPEG decoder is simple and easy to do cleanly.

JPEG's primary form of loss comes in quantization and "zig-zag truncation" which is when the image is transformed from the spatial domain to the frequency domain, generally a lossless or less-loss operation allows simply dumping high frequency detail from the macroblocks and then representing frequencies in steps rather than values of a generally guassian distribution.

The "compression" aspect in the sense that after we're done just arbitrarily dumping data is based on quite old and inefficient entropy codings.

Most other DCT compressions could easily implement the feature you're talking about and achieve similar results. As a matter of fact, it would require no changes to the standards and with only a little work could run some mathematical magic to merge 8x8 macroblocks into larger macroblocks with negligible loss. The only thing required to do what you're suggesting is to implement an abortive decoder which simply decodes purely the frequency domain (therefore pre-iDCT) and then pass that data to the compressor of any other standard. The one requirement is that the decoders for those codecs make use of the standard (thanks to JPEG) DCT coefficients for the process.

Here's the kicker... JPEG-XL is a monstrosity of a code nightmare. To implement all the fantastic features which JPEG-XL's baseline requires, there's a tremendous amount of code complexity. A hand written, sandboxed, safe implementation of such a decoder could be very time consuming. Where a good cleanroom JPEG decoder would require a few days to create, JPEG-XL implementation could take a year. Also, without extensive mathematical analysis, I'm concerned the entropy coder could have obscure malicious table jumps which could be quite problematic in the security world.

Finally, there's the performance issue. JPEG-XL isn't nearly as bad as JPEG-2000 (which is a truly amazing standard if impractical), but coding the large amounts of hand-tuned assembler needed to implement the standard is going to take some time.

There just isn't a strong enough use case for JPEG XL to justify its existence outside of niche areas such as newspaper photo archives.

ICANN responds to Ukraine demand to delete all Russian domains

When the war is over

The Russian people are not our enemy. In any war, people are convinced by their leaders through multiple routes that they have to fight. When this war is over, for the health of the world and the human race, we have to embrace the Russian people and integrate them back into our civilization. We can’t have a repeat of the end of the Cold War. Putin is in power for one reason and one reason alone. When the Cold War ended, we treated Russia as beggars and losers and we made movies bragging about how families sold their daughters as sex workers in exchange for wood to keep from freezing. The Russian people responded by embracing a leader who they believed would give them back their pride through a show of strength.

I absolutely want Ukraine to come out of this as a free country who develops into a great nation. But I don’t want Russia coming out of this as a starving and desperate dog.

US reportedly pushed ASML to cancel chipmaking kit for China early

Re: Team america

And America needs to buy Greenland from them

Is it time for 6G already? Traffic analysis says yep

XR uses less bandwidth, not more

As a codec developer, let me point out that a proper XR codec which has access to depth data can handle motion vectors substantially better than 2D.

XR doesn’t have to transmit stereo video. In fact, it’s silly to do so as it would limit the receiver of the stream to only the angle of view of the two cameras.

Rather, XR will transmit scenes created via photogrammetry where frame to frame, motion vectors will move vertices but textures will remain mostly the same. For the most part, from frame to frame, static scenes will remain mostly unchanged. Objects in motion will be transmitted as additional texture data which will only include data not previously observed and transmitted.

Scene changes similar in 3D to what was an I-Frame in 2D will be very data intensive and most likely will require buffering. But the 3D version of P and B data will be considerably smaller than its 2D counterparts.

But, these are all things we don’t like thinking about because it would suggest 6G isn’t needed for XR.

Here's who thinks AI chatbots will eventually be smart enough to be your coworker

Re: An AI

Product cycles are an issue with cars. Any car made today, with the exception of accidents or being made by GM or Ford should last about 25 years.

That means, when automotive shops are effectively factories capable of completely and automatically disassembling a vehicle, inspecting every piece and reassembling it it will take at least 25 years for meat sack mechanics to be out of work.

And, well, somehow companies like GM and Ford will continue to exist for longer than that and they certainly will never invest in vehicle longevity. The new cars sold 25 years from now should hopefully last 50 years or more. GM can’t even make a connected vehicle that can receive remote software updates. I’m still laughing that Cruise had to recall all cars to update their software. If I ever managed a modern car project (anything from 2005 forward), the first thing I’d do is make sure every single computerized component could be fully remote managed.

AI won't take your job, might shrink your wages, European Central Bank reckons

AI stopped me from using people

I believe that using AI this year alone made it possible for me to avoid long hiring processes and numerous interviews. I found that although AI wasn’t a perfect human replacement, managing AIs and fixing their mistakes has been substantially more productive than hiring extra people. Altogether on my one project, AI has cut at least three potential jobs.

AI is not a human replacement, but it is a massive productivity augmentation tool. As a result, it stops the creation of new positions because the risk is substantially lower and the time savings by using AI makes the process of hiring humans and on-boarding them counterproductive.

As a note, this is in Europe and these are highly skilled IT jobs, positions that would have posted with “Masters degree plus 5 years experience required”.

Apple exec defends 8GB $1,599 MacBook Pro, claims it's like 16GB in a PC

Symbian all over again

Symbian destroyed Nokia (let's be fair, Nokia destroyed Nokia, but...)

Nokia made ridiculous blanked statements about how users didn't really care about what ran on their devices. What mattered was that they wouldn't have to charge their phone every day. Nokia bashed Microsoft Pocket PC over the insane system requirements that were of course battery hogs. They laughed that Pocket PC needed 100Mhz processor and 16MB of RAM but Symbian only needed 16Mhz and 2MB of RAM.

Well, long story short. So long as Nokia could dictate all the media and applications used on the phones, this was perfectly ok. After all, if you ran a large number of static applications where the content was closely controlled, there was no issue. You could optimize all the content to the specific device and it would work just fine.

On the other hand, as soon as you introduced a web browser where the web pages dictated the content on the screen... you had to have enough CPU to respectably run the Javascript, decode the images, and of course store the decompressed images. If your device only had 2 megabytes of memory and a downloaded photo decompressed to 4MB... and of course there was no swap space, well you see where I'm going.

Nokia honestly spent millions on teams of people who would contact website by website to attempt to convince them to alter their websites to be mobile friendly.

Pocket PC devices on the other hand increased CPU performance and RAM to support the content.

Then Apple came around and made a phone which had a pretty much full web browser, LOTS of CPU and LOTS of RAM. And Nokia kept insisting that Apple and Microsoft were stupid because if they used Symbian, they could somehow magically display 20MB of data on a device that only had 2-8MB or RAM and that it was better to have a phone that you didn't have to recharge than a phone which was actually useful for something.

Now we're in a world where most people will be working on high resolution media (even if just to look at the pretty pictures), HDR pictures and videos, etc... the Internet expects computers to have enough RAM to support this. I can understand buying a $200 PC from AliExpress with 8GB of RAM, but I would never consider buying any machine that I would need to be productive on with less than 16GB and if I want to keep the machine longer than a year or two, then 32GB.

Let's go one step further. When Apple enters the AI market, they'll be forced to do everything in cloud and almost nothing on the edge because the models won't fit on the laptops they're currently selling.

There's no Huawei Chinese chipmakers can fill Nvidia's shoes... anytime soon

Different perspective

As you may or may not know, Huawei is their own fab. The international press has been pumping a lot of nonsense for some time. Huawei is a massive producer of Flash memory for their own SSD devices. They also produce their own circuit boards. Huawei also fabs their own 7nm which appears to be about a year ahead of SMIC in development.

At Huawei Connect 2023, Huawei released the Atlas 9000 platform which like the H100 is a platform. H100 is not a card, it's basically an entire server. Atlas 9000 is a platform which scales from tiny to super-computer in size. Because of Huawei's supply chain, they can deliver have entire massive-scale AI supercomputers in transit in less time than it takes to negotiate a sale with NVidia.

I'm only relaying my own research because NVidia has extremely long lead times on delivery. And I would rather buy something which can be shipped this month rather than waiting two years.

It's a little funny, it wasn't until the US changed their message from "Chinese stuff isn't any good, you'll buy American because it's better" to "You can't use Chinese because it's a security threat" that I took Chinese tech seriously. It was actually the US campaign against China that made me realize that as long as I take the security aspect seriously, Chinese tech is good enough to replace American tech.

Uncle Sam is this keen to keep US CHIPS funds out of China

Wow, designed to backfire

The Chinese have managed to produce 7nm semiconductors without EUV. This is a task Intel spent billions on and failed. They will continue to innovate and advance their semiconductor production. They will continue to shrink their processes without outside technology.

Of course, their advancements mean that they will produce chips for a LOT less money than other companies.

By adding clauses blocking collaboration between China and the U.S., the U.S. will be unable to compete.

Oracle's revised Java licensing terms 2-5x more expensive for most orgs

Re: with 49,500 employees, all of whom are applicable

Help me out here. The applications I develop are, at a personal project level CAD software, operating systems and compilers. On a professional level, I code storage systems, physics simulations for things like bio computing and weather prediction.

I do actually have a storage system in production which provides object storage to CERN projects which is based on Java. Its system requirements are idiotic and if we could rewrite it in any language actually suitable for such tasks, it would save us tens of millions of dollars.

For business and accounting type systems, I code everything in C# because the language is insanely well maintained at this time.

What is a suitable application for Java? It always struck me as the square peg in the round hole. I used to develop a Java Virtual Machine and a clean room implementation of the Java libraries, and I suppose it was a fun project, and I needed to do it because I needed a platform independent JVM for embedding in a web browser on all kinds of platforms. But when DOM and HTML canvas became a thing, I didn’t really see the point anymore.

So what kind of software would you develop using Java and is that because of habit (shamefully, I still use C and C++ for some things because they’re comfortable even if they’re terrible), or does Java offer something which makes it well suited for some task?

Jury orders Google to pay $340M patent-infringement damages over Chromecast

Re: How does one get such patents in 2010

I read the patents and the first google I did voided all three patents irrefutably. "DLNA".

HDMI CEC covered it to.

IR blaster as well

These are all modern enough that there are probably active patents.

So let's try Real Player for expired patents.

H.263 patents also covered all 3

Logitech has to have expired patents for the control aspect.

Google's lawyers were totally incompetent.

Make chips, not trade wars, says Semiconductor Industry Association

Forfeiting leadership

All that the US has achieved by doing this has been to motivate China to become an alternate source/supplier for all these technologies. The result will be that they will compete and undercut all western companies forcing them to struggle. The China will continue to move forward with extremely high momentum while western companies (and Taiwan) struggle to recoup their investments on R&D. Eventually China will end up as the preferred suppliers for all these technologies.

The wins for the west are short term and short sited. Even if the restrictions are lifted today, China’s forward momentum will only slow, not stop. They will cut their dependence on western suppliers.

The only good news is, by the time China is far in the lead, the people responsible for starting this will be too old (or even dead) to care… as if they’re not pretty much there already.

Bad times are just starting for India's IT outsourcers, says JP Morgan

Re: Nobody mention AI

AI as a service?

1) Open Bing chat.

2) Ask “write a bash script to download and install Alpacca on my machine

3) Open Alpacca URL

Re: I can't stand...

The issue is the number of people needed to be productive.

No code and low code is a big driver for business systems today. It is getting to the point where a good low code platform is enough to handle most IT development tasks.

Add ML to the mix.

Since I started using copilot and Bing Chat to assist my work, my productivity has increased drastically. Not two or three fold. Substantially more than that. I am now completing projects on my own in less time than it used to take to spec the projects out. What’s worse is that the quality has improved greatly in comparison. I expect that these very early generation LLMs are going to improve considerably in very rapid succession. And once I finish my current project, I hope to switch to running an open source LLM which I expect to perform substantially faster and increase my productivity a great deal more.

I can honestly say that I could now complete 12-20 large scale IT projects in a year with these tools compared to the past where I used 3 months just to hire an outsourcing team to do one.

I think we’re about to see a considerable collapse of labor requirements thanks to a single LLM with one-two skilled developers greatly outpacing managed projects with 10-30 people and all the paperwork and bureaucracy.

I am tempted to design and build a large database project based on a previous project using one other developer and an LLM and establish real numbers for before and after for LLMs.

China the largest buyer of chipmaking machines as sales hit an all-time high

“Advanced technologies” aren’t always needed

This is a misunderstanding which seems to make a lot of people confused.

You don’t actually need more advanced nodes for nearly anything which the US is attempting to curtail.

Advanced nodes increase density which is a non-issue unless portability is the need

They also decrease power consumption which is also a non-issue if power is available. Thanks to China’s solar, battery and wind initiatives, total non-issue

They also decrease heat radiation. No problem if you can remove the heat. There are of course many profitable ways to consume the heat from data centers. And China does have ways to do that.

They can also increase clock rates which can be easily compensated for by simply adding more chips.

There is the issue of high frequency communication which is a limiting factor due to things like quantum tunneling. I won’t make this a quantum physics or regulation theory lesson, but the only way to deal with this is to have more parallel channels of serial communication. A better solution is to write better code which doesn’t need so much bandwidth. Use binary data rather than JSON for example.

The last and toughest issue is latency. Sometime you just need to send data from n less time. This is generally addressed with higher frequency busses. There are almost no places in high performance computing and AI where you genuinely need better latency. We use it when it’s available, but if you consider most super computers are running like slugs because scientists tend to write their code in Python, R or Matlab, this isn’t really the issue. Many HPC applications could run on laptops with a decent video card if they were coded properly.

The only thing these advanced fabrication technologies are really needed for is things like 5G. And this is a huge problem because China’s 5G solution is much cheaper and quite a bit better than their competitors. Even now, for example Nokia is struggling to deliver millimeter wave or SA. Two features Huawei could deliver on day one of 5G and also the two most interesting features of 5G. If the US hadn’t bullied everyone to avoid Huawei RANs, the whole western would would be shipping crates full of US treasury notes to China never to be seen again.

Overall, these semiconductor fabrication equipment sanctions have very little impact on China except that they are being forced to up their game will eventually catch up with the west and leave them perfectly positioned to compete aggressively against western companies.

The ship has sailed. There is no going back. If all the sanctions were lifted today, China would work just as hard to become the world leader in semiconductor technology. What Trump and Biden have accomplished is to make it so China might simply choose to withhold advanced Chinese tech from the US and their friends.

The good news is, most of the people who have initiated this inevitable end more than likely, due to old age won’t be around to see the world when the US is pan handling on the streets of Beijing begging for bread. But I’m sure they’ll go down in the history books as the presidents who forced China’s hand.

Boeing's first-ever crewed mission in Starliner ISS spacecraft delayed to late July

I didn't realize they were still trying

Honestly, I was under the impression that they had just given up. I wonder if they have a plan to how to lobby for a billion dollar a flight once BFR starts launching.

Ransomware severs 1,000 ships from on-shore servers

DaVinci Virus?

I'm pretty sure this is the DaVinci Virus and it will likely eventually cause the ballast in the ships to ship and the little boats will flip over unless a million dollars is deposited into a specific numbered account.

I recommend you they track down a hacker named "The Plague"

China may have to reassess chip strategy in face of US sanctions

Re: US semiconductor industry???

Thanks for this write up. It filled in some stuff I didn't know.

I would like to point out that most of ASML's technology and the technology of their vendors are patented which means that where patent protections work, they are safe. But in countries being sanctioned in ways they see as a violation of trust, the patents serve as blueprints to reproducing said technology.

I like to point out to people that if you were to compare China to the U.S. in a very naive way which says that "Any American is twice as good as any Chinese person", considering that there are twice as many people in China and they're demographically distributed similarly as most other countries including the U.S., then China is still twice as good as America. Or at the very least, they have the capacity to achieve twice as much as the Americans.

This tells me that if the Chinese can outperform America consistently by at least a fact of two, the Chinese are absolutely going to pass America in every conceivable way in time.

Google datacenters use 'a quarter of all water' in one US city

I am not convinced.

If the water is already evaporating on scale, then there are some pretty interesting situations to consider.

If the evaporation is indoors, the humidity of the data centers would increase and cause condensation.

If the water is at a high enough temperature to evaporate, there is potential for extremely low cost water filtration within a farm. The water released from the data centers could be pumped into pools within green houses with properly shaped roofs and the water would evaporate and slide down the sides of the system and be reclaimed. While doing so, it would also cool slightly.

In that same system, the water as it falls and cools can be used to turn low resistance turbines and the energy can be collected. The additional energy removed from the water in the process would be a few more degrees of temperature.

An evaporator system the size of several indoor Olympic swimming pools could be incredibly profitable to everyone involved.

I don't think the water is quite ready for mass evaporation, when that much water is carrying that much energy, only fools would pass up the opportunity to reclaim as much of it as possible. In fact, making sure that the evaporator system can collect even more energy from the sun would make it vastly more profitable.

I think there has to be a reason the water and energy isn't being reclaimed. Either

- they are using so much water that the heat is being broadly distributed enough that it doesn't actually evaporate easily and as such energy isn't easily extracted

- they are mixing the potable water with non-potable water

- they are mixing potable water with chemicals

One thing I know for sure from following some pretty interesting people who talk about DC engineering, they can't account for pretty much every milliwatt of potential energy savings within a datacenter. If there is a penny to be saved or made by exploiting the waste output of a data center, they will find it and exploit it.

Latest US blacklist spells trouble for China’s biggest domestic 3D NAND supplier

Re: Too Far?

This was a fresh perspective for me to think about.

I think I'll give you a knee jerk reaction.

China subsidizes YMTC heavily and the result has been some extremely inexpensive flash for consumer and enterprise. I'm not completely sure, but I think I just spent about $100K on YMTC flash last month as the Huawei products were better than the Hitatchi, IBM, NetApp, Dell and SuperMicro solutions I evaluated. I believe the storage system I purchased was substantially faster for the price than their competitors because they had access to much less expensive enterprise SSD for cache.

Here's the thing though, China needs flash too.

In fact, China needs flash so badly that the government is willing to do what is necessary to make it competitive and affordable. As a result, China is a world leader in Flash technologies now.

Now here's the crux.

China is definitely going to lay on the funding thick now.

Consider what you just said. You're saying that YMTC is a threat the the rest of the world because China values YMTC so much they're willing to subsidize it. In fact, you're suggesting that these subsidies will run the competition out of business.

What do you think will happen now?

If YMTC is a threat, the Chinese government will ask them "What do you need?" and they will make it happen.

Already, China is growing and producing massive amounts of silicon ingots. They've already developed technology for slicing and polishing and transporting them. It's only a matter of time before the ingots grow large enough to start production (if they haven't already done so). That will cut Chinese semiconductor cost in half or even further.

Then there's lithography.

EUV isn't really a big secret and it's an evolution on earlier technology. Do you honestly believe that China can't make their own equipment?

When they do, they'll buy the machines for $1 million instead of $100 million and cut the price of semiconductors by 80% or more.

When Trump imposed those tariffs, he guaranteed that China would respond by investing trillions in making their own semiconductors. Also, they'll invest trillions to collapse TSMC which would collapse the Taiwanese economy which would make the west lose interest in Taiwan.

All this latest decision will do is accelerate China's timelines.

Trump and Biden will go down in history as the fools who signed the death of the western semiconductor industry.

Nvidia faces lawsuit for melting RTX 4090 cables as AMD has a laugh

What will this accomplish?

NVidia is now aware of the problem and will obviously take steps to resolve the tech.

Some people will have been affected by the issue and it is very likely they will have to pay to fix the problems. NVidia will most likely attempt to address the financial issues and may even have to issue a recall on the damaged boards.

This lawsuit however will place NVidia on the offensive which will either delay financial reparations to the consumers and possibly even delay a recall as their legal team will most likely advise them that it will cost a great deal more to admit fault.

In the end, the affected consumers will probably receive checks for $50 if they live in the USA and the lawyers will make millions.

This is an excellent example of how class action suits can seriously backfire. The consumers will be hurt and NVidia will end up paying less than if than if they actually fix the problem.

Your next PC should be a desktop – maybe even this Chinese mini machine

One step further

I'm using some relatively older NUCs for most things, but I came to the realization that it's a heck of a lot cheaper to pay for a game streaming service then to pay the electricity on my normal desktop which is a pretty decently specced gaming system.

We also have an XBox Series X in the house, no screen attached, we just connect remotely.

I don't think I'll ever pry the gaming rigs away from my kids, but I'm pretty sure they'll be fighting over my video card soon when I switch completely to streaming gaming.

Chip shortages still plague carmakers despite weaker semiconductor demand

Duh?

Cars use chips mostly for sensors.

These chips are chosen when each part is designed. So, for example if there is a temperature sensor on each brake pad of the car, it will have a microcontroller that connects the thermal sensor to the CANBUS of the car.

The brake thermal sensor will be tested extremely well over a long and excruciatingly detailed product approval cycle. It will be tested in the lab and in the road. Once the design is certified, it’s given a slew and things like changing parts will not happen… maybe passive components, but certainly not temperature sensitive devices such as semiconductors.

The car manufacturer will use that same sensor on as many cars as possible and for long as possible because designing and certifying a new part is very expensive and more importantly requires a new SKU. Then every car which uses the old SKU as a replacement part will need to be certified for the new SKU. Distribution of the new SKU will take time and updating all systems to use said SKU as a replacement will need to be updated world wide. Then, even if all you did was move from a microcontroller built on 180nm to one built on 28n with a 1 to 1 compatibility, it will take years to identify, resolve or work around, and catalogue the issues with the new part.

Even now, designing new cars, if you add a brake thermal sensor to the design, shortage or not, the car company will use the old part with the old chip because the engineers who even know the requirements and parameters of such a sensor may already be retired.

That means the automotive industry has to find chips made using the same process.

The problem is, the number of cars using that part has been increasing over the years and the number of replacements parts needed has increased proportionally. So, the demand has increased but the supply has not.

Designing new SKUs is an option, but it takes years and would probably bankrupt the companies needing them.

Also, no one, even under the chips act is scaling availability of manufacturing those parts. Everyone wants to make new and bad ass stuff to be competitive.

China is always willing to scale production of this tech. In fact, they can easily scale 28-250nm production with no supplies from outside of China. But that would require actually depending on China for these chips and why should they scale to help the US car makers when the US is attacking every company that can actually solve the problem?

The chips act is great because it helps the US catch up with Asia in terms of tech. The US companies Intel and GF are willing to produce unlimited amounts of 10nm and are even willing to scale 28nm production. But that would only be helpful if the companies who actually need the chips could cover the cost of changing the process. This is not an option… unless of course the US were to invest in the companies using said chips to get new SKUs designed and certified.

We can’t cry like little babies over this shit when no one actually is listening to the people who actually experienced a real chip shortage.

China may prove Arm wrong about RISC-V's role in the datacenter

Nope... not a risk... we just don't use them... ARM that is

We're currently planning 5 exa-scale computers in the next 5 years. We had been working with ARM as we're developing our own CPUs. Then after extensive research and money spent and of course the Trump embargos on China, we (a European collaboration) moved on to RISC-V. Trump and Biden taught us that no one is immune to being crippled by friends... and unlike China, we just don't have the resources to bounce back.

So, we've more or less dumped ARM beyond some POC related things. So we'll build systems with a few dozen or maybe even thousand CPU cores, but then we'll scrap them and replace them with a RISC-V ISA.

Here's the real reason why ARM offers nothing in the datacenter over RISC-V.

There are so many varieties of ARM that compilers just aren't optimized for them. It's not like Intel or AMD where compilers are written to optimize for each generation of chip. For example, one generation of compiler can optimize code for CPU cores organized in a ring bus, the next generation will be organized in a mesh. But with ARM, you never know what you're going to get and the compilers don't optimize for a generation or even a given core. As a result, ARM processors are kinda useless when you need performance. Consider that code optimized for a NVidia processor probably isn't optimized for Qualcomm and certainly code written for Apple M2 isn't optimized at all for NVidia. You end up using compilers which target the minimum common denominator which means no support for many instructions or architectures.

RISC-V will be worse. As most ARM processors actually are mostly designed by one company. On the other hand, RISC-V processors will be designed by pretty much anyone. They'll have nothing in common and their instruction sets are highly customizable. As a result, we will very likely become dependent on JIT compilers which on Exascale machines are actually optimal as they can optimize code for the platform as they run based on their current states.

Already, I can't wait to get my Alibaba RISC-V laptop to start optimizing code.

Micro Focus bought by Canada's OpenText for $6b

Successful enterprise software starts…

Successful enterprise software starts as products accessible to young and agile future engineers hoping to choose a direction.

Micro focus is a paywall.

If Microfocus owns a product, it will never see the lights outside of the Microfocus paywall. They are a buy first pay later shop. Their entire existence is all about justifying open source. Because even though Microfocus already has a solution, it will be rewritten in the open source because it’s less work to develop an entire system to replace it than to work with Microfocus.

Your job was probably outsourced for exactly the reason you suspected

What about liability?

When you outsource, you specify a project and set acceptance criteria and then pay a company who can be held liable to deliver.

Yes there are horror stories of how this can go wrong. The British and U.S. governments are absolute experts on botching their procurement of IT systems on a biblical scale. But overall, most projects outsourced generally are delivered accounting for the fact that the original price and deadlines are set by who can lie the best. Meaning that in order to win a tender against similar companies, all the vendors tend to intentionally underbid and quote delivery times over optimistically knowing they can ask for more once the project is in progress. So, it’s generally best to estimate at least 50% extra cost and time.

Even then, most systems are extremely similar. Using mature tools and platforms, vendors can estimate quite accurately the amount of time and resources required to deliver most any business system. For more technical projects, they can often throw large numbers of developers at making them happen.

When you employ your own development staff, you have no one to hold accountable but yourself. You need to have expert knowledge at planning and executing development projects. You can see salaries double in the time it takes to complete a project. Many such projects can see entire project staffs rolled over during their lifetime.

If you are going to farm out development to an external organization anyway, why waste time or money going local when going overseas will yield similar results? I haven’t met many great ITT grads, but I’ve known a bunch of good ones. And for most projects, you just don’t need great. You need good enough.

And don’t forget, most people in IT, no matter where you look are about equal. Each place has their rock stars, each place has their slow but steadies and each place has their “there’s gonna be a law suit or a lynching” and all in about equal proportions

If I were to outsource, I would look to Lithuania. I have worked with IT people all around the world. I have even been in one country training their IT staff and a month later in Lithuania training their replacements following a surprise outsourcing. The Lithuanians real impressed me on every count.

China seems to have figured out how to make 7nm chips despite US sanctions

Re: Another shoe of the US Clyddsdale falls.

I enjoyed your perspective even if I consider your writing style to be as annoying as my own.

Intel said for quite some time that 10nm from Intel could achieve the same density as 7nm from TSMC which was mostly true as Intel and TSMC measured differently. Of course there were exceptions, but overall, Intel and TSMC evolved their processes greatly over time and once the limits of TSMC 7 and Intel 10 were reached, they both moved a notch down in size.

Intel 4 should be able to employ 7nm fabrication to match the density and power foot print of 4nm from TSMC or Samsung.

I haven’t found details on SMIC’s 7nm node yet, but I would expect they have to work on the following

- Transistor size

- trace width

- insulator width

- yield

- more

I would suspect that SMIC has nailed the transistor size issue. But yield will probably be a beast for them at this point.

What I think is really interesting about this that SMIC had to work around a LOT of sourcing issues. Everyone has made a lot of noise over the EUV tech, but in truth, with enough money and engineers, EUV is not actually a mystery. The patent documents cover it it a lot of detail and they just had to figure out how to do it themselves. I would be surprised if it was actually difficult to do once China decided to ignore patent rights.

The big dog was wafer production. It takes an incredible team of people to produce crystal of such purity. It is also really easy to hide the process of purification, growth, slicing and polishing. 7nm is pretty far from subatomic, still 1-2 orders of magnitude from it. But crystal production for semiconductors probably require at least a 85% crystal accuracy (speculating from basic chemistry knowledge) to gate accurately. Layering must also be extremely difficult since layers are very likely less than 30nm high. (7nm trace, 15 degree angle of exposure, 120 picometer photon [estimation], 350nm wavelength [EUV estimation]).

This would mean that NMOS or PMOS at this scale would require some truly insanely accurate robotics and so ridiculously accurate centrifuges for layer application.

I think what many people didn’t recognize is that once something has been accomplished by anyone anywhere, whether through patents or marketing material, or other sources… reproduction of that technology is much much much easier. All that keeps people from copying it is patents and good faith. When it became public knowledge that diagonal EUV lithography did in fact successfully etch 7nm wafers… the financial risk associated with reproducing that technology was mitigated greatly. You really would only need a team of chemists, physicists and precision engineers to do it again.

What I think should also be considered is that documents published regarding sub-nanometer fabrication allows China as a whole to invest in skipping multiple generations and even beating their competitors to the goal. No one says China HAS to move ton5nm or 3nm first. They can be happy with 7nm and focus entirely on 500pm instead.

That of course raises the next issue which is that the really difficult stuff to reproduce is the software.

If China is blocked from Mentor Graphics and other synthesis tools, they will be forced to stick with versions licensed before the embargo. They can’t just call Mentor and say “I need a patch for our proprietary fab process”. They will have to produce their now synthesizers, physical and logical simulators. That actually takes a lot of time. There is a lot of graph theory for rule constrained synthesis. Also, field theory, especially when concerning quantum effects which is necessary at such low gate sizes takes teams of brilliant people years to achieve.

Elon Musk considering 'drastic action' as Twitter takeover in 'jeopardy'

Where is the ROI?

I have been seriously wondering what Musk would actually be buying.

Twitter isn’t a meaningful information sharing platform. It does not promote communication. It requires users to favor banter. I haven’t seen much use of Twitter as anything other than sharing links and making quirky comments. It is commonly used for taking swipes at other people on a platform which greatly limits follow up discussion.

I see absolutely no features or Twitter which provide real value. As an example, look at Facebook. As a platform; it is generally also a cesspool, but everyone tends to have a Facebook account and often have messenger because it is the modern equivalent of a personal phone book. Twitter doesn’t have a real networking feature. You just follow people, you don’t connect with them.

I wouldn’t consider Twitter a worthy investment. The only value I can see for Musk to buy Twitter is to hurt Donald Trump. And frankly, he is rich enough to do something petty like that.

Actual quantum computers don't exist yet. The cryptography to defeat them may already be here

Policy vs Technology

Keeping data protected for 75 years is a matter more properly managed by policy than technology.

It would be important to identify the reason you would encrypt there data and then identify why you would decrypt it.

Most data doesn’t actually need to be encrypted if it is intended for offline storage. Rather, a good physical security policy such as long term media storage in a secure location would be ok.

Encryption is primarily necessary for when data is to be transmitted. Until it needs to be transmitted plain text is probably ok. But when you transmit the data and it is intercepted, if it needs to remain private for 75 years from the time it is intercepted, it is clear that data will eventually be compromised. Therefore, such data should be physically transported rather than electronically.

You’re absolutely right in my opinion. Looking 5-10 years ahead may be achievable. Looking 25 years ahead would be entirely impossible. We honestly have little or no idea how quantum computer work today and we have no ideas what break through swill occur in the next 25 years. At this time, every quantum programming language is roughly equivalent to a macro assembler. We are literally programming with gate operations like assembler op-codes and of course macros. 25 years ago, chip design was very similar… almost identical. Then we figured out how to write high level code which would automatically compile to synthesized logic. And then we rocketed decades ahead overnight. We have absolutely no idea what will happen when the quantum world sees such a leap.

So, if data must be secure for extended periods, floppy-net will probably be the best option.

Cisco EVP: We need to lift everyone above the cybersecurity poverty line

Step 1

Provide software updates for old hardware.

Currently, most Cisco customers are running hardware that is locked to old ciphers and hashes. This is because their hardware is out of support.

Cisco has been pushing upgrade cycles for decades that deprive customers of patches for their equipment.

There are millions of Cisco routers and switches that use MD5 passwords, RIPEMD160 SSH hashes, outdated and insecure AES key cycling for SNMP.

In fact, there are many in production, supported network devices which use weak or seriously outdated AES and 3DES ciphers for SNMPv3.

Then there is the Cisco golden goose, YANG centric NetConf which which Cisco DevNet provides lots of documentation and training to push for the use of poor security programming practices such as storing keys insecurely in example code.

Then there are products like Cisco ISE which is Cisco’s golden security tool at the heart of all Cisco security which uses severely out of date Apache TomCat versions, Java libraries with insecure LDAP implementations, extremely insecure and easily compromised SAML implementations and more.

Then of course there are insane cases such as Cisco FirePower which make use of insecure network stacks beneath their secure implementations.

Oh… I don’t even know where to start when it comes to course system patching. Every single IOS-XE device runs Linux and lots of things like OpenSSL, OpenSSH, etc… but even after ~20 years of IOS-XE, you still don’t have granular software updating. So, you can’t just “apt update” for example and get new versions of most of the tools which regularly have critical CVE’s, you actually have to schedule downtimes and perform time consuming full system updates, sometimes with update cycles as long as 45 minutes.

I can go on for hundreds of cases of this.

Overall, Cisco is a mess when it comes to security.

But let’s be honest, they’re not much worse than their peers.

Export bans prompt Russia to use Chinese x86 CPU replacement

Re: Russian politics aside

Thank you! Compared to your normal posts… this was was nearly a marriage proposal ;)

Re: Russian politics aside

Honestly… I am an American and I’m just excited about where the tech is going. I really never had an interest in taking sides with anyone. Let the gorillas thump their chests and grunt. From a tech perspective, there is nothing but good things to expect to come from this.

Russian politics aside

WOW!!!

Ok... I'm in complete disagreement from a technical perspective from the author. I feel as if there's a bit too much "Queen and Country" happening here. Patriotism and propaganda is fine, but there is so much more to this article than just "We're so much better than they are." In fact, this article should make the author seriously reevaluate that.

So, not long ago, China was embargoed and they were cut off from western technology. Since then they have

- launched a multi-pronged plan to mitigate the loss

- used ARM under a strange "ARM China is not ARM LTD" clause that allowed them to keep using ARM like crazy.

- moved approximately 50-200% faster on die process advancements than their western peers... depending on how you evaluate this. But whatever the case, it simply makes sense since their peers have to do things never done before and they only have to learn from what is already known.

- either licensed, bought or created most of the peripheral infrastructure surrounding CPUs including audio, video codecs, USB, MMU, interrupt controllers, DMA controllers, ethernet and more.

- launched a slew of RISC-V based processors and advanced RISC-V technology at least in terms of synthesis more than anyone else on earth.

- managed to use loopholes in the Cyrix x86 license via Via Tech to get a hold of x86 ISA (at least until Intel figures out how to go after this) x64 never really had these issues.

- Grabbed what I believe is S3 graphics tech which is nothing to write songs about but is a truly solid foundation for building on as they should be able to tack on a pile of specialized RISC-V cores to build most of the processing capacity and with some serious focus on memory technologies... meaning finding an HBM2 like solution and solving some pretty serious cache coherency issues, make a modern GPU.

Yesterday I was in front of a classroom and a student asked me to look at his progress on a project on his laptop. His machine was a several generation old Core i3 with 8 gigs of RAM. It was sluggish, but it was entirely usable even when loaded down by a very processor intensive application. I'd only guess that if I searched for a benchmark of this machine relative to the motherboards seen in this article, they'd be quite similar.

For HPC, Intel is not a requirement. Huawei and others solved this problem by producing CPUs which are less energy efficient than Intel or AMD but using solar energy and batteries for power. If you join a meeting with Huawei to buy a super computer, they present to you systems which they can deliver at any scale (as in Top500 systems) using Chinese technology and they can deliver power through solar. This is not a problem. And frankly, so long as it has a fully functional Linux toolchain including Python, Numpy, Scipy, Julia, C, C++, Fortran... I really don't care which instruction set I'm using. The only really important missing tool on the systems is Matlab, but it does have Octave ... which isn't really a replacement, but it could be good enough.

For telephones, there's ARM and soon RISC-V

For normal desktop, this x86 solution looks like it should be perfectly suitable for most users. Toss on a copy of Deepin Linux or ReactOS and it'll be fine.

Gaming and graphics workstations... that will require something else. And while everyone loves western and Japanese game studios... China is producing a crap ton of pretty good games these days... mostly on Unity and Unreal (I think) which I think could cause issues, I can honestly see China pushing Chinese and Chinese owned publishers to produce for a Chinese architecture. And let's be honest... most of the best games out there these days are well known for "they could run on a toaster"

At China's current rate of progress, they should meet or exceed western tech on every front within a few years... not necessarily always by looking at benchmarks and such, but based on user experience.

We can thank Donald Trump and Joe Biden for this. If it weren't for them, China probably would have kept chugging along at a moderate pace and had been happen just to keep using and licensing western tech. But thanks to the embargos which forced China to increase their efforts so drastically, we're going to see some truly amazing advancements in tech. This will be simply because the speed China can and is moving at is IMMENSE and soon everyone else will have to really push it into overdrive to avoid being left in the dust.

The tech world is going to be truly amazing now. I can't wait to see what comes from this.

What is so exciting is that there's absolutely nothing that can be done to slow China down now. Not only are they hellbent on never being crippled by the west again, but they need to do this for their economy. For them it's full speed ahead or bust.

Elon Musk flogs $8.4bn of Tesla shares amid Twitter offer drama

I don’t get it

So… Twitter

From my use of it,

if you write threads, people hate you because that’s not Twitter

If you write tweets

- people hate you because even a haiku author can’t express themselves in the character limit without being (intentionally) misunderstood by most people

- people hate you because you sound like a raving lunatic for writing short meaningless messages

- people hate you for writing crappy jokes

- people hate you for writing anything

I have never experienced a platform so well suited for spreading hate and discontent. It’s a toxic platform. I am sure there is someone out there who isn’t a shit tweeter, and I think Twitter sometimes is a good platform for making announcements… but, it’s just not a good place.

Then there is the issue of Musk buying the platform… has Musk ever bought a company and made it successful? I was always under the impression that he’s a maker and builder… not a buyer.

Deere & Co won't give out software and data needed for repairs, watchdog told

Stack vs heap

Deere has a very odd internal development policy. For decades, they wrote their code in C and they have specific rules for coding which require all data to be on stack, not heap. Also; they don’t allow structures.

This means that there are decades of impressively shitty code on John Deere devices. There is a high likelihood that the engineers who wrote the code have no ability to read it themselves.

I think John Deere must have done some sort of internal assessment and realized that if they released source to their systems, they would likely have to also provide some degree of support for the code and it is entirely possible that’s not an option.

I have see lots of JD source code and what I would say is “if you drive a JD tractor, you’re lucky to be alive”

Cisco to face trial over trade secrets theft, NDA breach claims after losing attempt to swat Leadfactors lawsuit

2008 was long past the point of original thinking.

I am more than happy to talk smack about Cisco’s business practices. I think Cisco is one of the dirtiest companies on the market at this time and the funny thing is, they don’t even realize it. It’s one big family of “We drank to Kool-Aid” out there in San Jose. I mean really, you have never met a more brainwashed group of… groupies than the Tasman campus groupies.

That said, I was working at a major competitor to Cisco in the collaboration gig at the time. In fact, Cisco acquired them for something like 1.8 gazillion dollars. And one thing I’ll tell you about the social and conferencing market is that there hasn’t been an original thought in that market since Walt Disney built Epcot.

Everything in that entire field is strictly logical evolution and if there is even a single patentable idea is social collaboration anymore, I’ll eat my socks… and that’s just gross.

I think the last piece of innovation in that field was when Zoom first reared its ugly head and rendered video in a 3D oriented fashion. That really threw us through a loop. But really, by 2008, there was nothing left to invent. We are soooooo long past that point

FreeBSD 13.0 to ship without WireGuard support as dev steps in to fix 'grave issues' with initial implementation

I was about to

sweep in and complain about poor coding.

Whenever I write kernel modules in C (Linux ugh), I find myself spending far too long detangling unintuitive preprocessor crap that has no place in 2021. When implementing secure protocols, most ciphers allow in-place operations, but usually headers need to be prepended and you will never find a solution to this problem that permits for the headers to remain memory aligned or for the buffers to be encrypted do.

This means to effectively make protocols which encapsulate higher level data, good buffer management is necessary. And while C allows you to do anything you want, as efficiently as you want, almost all solutions to the problem tends to lead towards reimplementing object oriented language features... usually in preprocessor macros or using crazy compiler extensions for tracking offsets into buffers based on their positions in structs.

There is also the whole game of kernel buffers. Since the kernel is in privileged mode, performing allocs and frees is frowned upon. The structure of the kernel memory space is expensive and dangerous to randomly allocate memory, especially if it may trigger the kernel to need to further allocate memory beyond its initial pool. Since the MMU is mostly bypassed in this mode and since C memory management generally is not relocateable, the only real solution is to overprovision needlessly.

I could (and probably should) write a book on all the numerous problems related to kernel development as well as the endless caveats of coding kernels in C, but let me simply say that while good C code is possible, it’s rare and far too many trivial tasks are managed manually and repetitively when coding C.

I don’t particularly care for the syntax of Rust or Go. But both languages run with a great concept which is... if it’s a very common task that can be added to the language with no real additional cost, do it. As such, both languages understand strings, buffers and data structures. There is no need for disgusting hacks to implement things like macros that are all hacks to support something as trivial as foreach.

C could fix these things as well. But it’s a conscious decision by the people steering the language to keep it as it is and to leave it to libraries and compiler extensions to do it instead.

I love C because if I want to write a new C compiler, I can make something usable and likely self-hosting within a few hours. But this isn’t the characteristic of a programming language I would want to use in 2021. If I were to spend my time on such a project, the first thing I’d do is build extensions for strings, buffers and data structures ... and it wouldn’t be C anymore.

Oh... and most importantly, I would drop the preprocessor and add support for domain specific language extensions. And I’d add proper RTTI. And I’d add an extension for references. And of course relocatable memory. And probably make ...

You know what... I don’t think I’d do anything other than bootstrap the new language with C and then basically just ignore the standard from there :)

Flagship Chinese chipmaker collapses before it makes a single chip or opens a factory

Re: More to this than meets the eye

They actually seen what we call IP theft as free market economy. Rather than protectionism which is supposed to allow companies to recoup their investments through patents and such, they believe that everything that can be copied is open source and each copy will generally generate innovation and place pressure on all players to always make advances.

I am currently sitting just outside a semiconductor research facility as I write this. It makes high frequency radio semiconductors for communications. Almost all the technology inside is off-the-shelf and while they obviously make a buck off patents, everyone there knows that their real protection is research and progress. We have a real problem at this location because if the US government becomes protectionist against its allies, they’d have to just shut down.

The good news is, there is another building just off to the side which is a nanotechnology research facility that can produce everything without dependence on the US. So rather than investing massive sums of money in traditional semiconductors, they focus their efforts in nanotechnology as a replacement.

When China eventually catches up in semiconductor fabrication to the US, it will quickly surpass them as American companies will depend on protectionism. And even if the US were to undo all the restrictions Trump enacted this moment, China would still invest heavily in their own tech and will still pass the US since they have now learned that if it happened once, it can happen again. Not only that, but I imagine the facility I’m sitting at will start buying from China at least as much as from the US so they will never have to worry about being completely cut off.

The sanctions placed on China by the Trump administration will likely be the biggest boost to China that could have ever been possible. It will take China time to recover from it, but when they do, it will leave the entire rest of the world without any leverage when negotiating with the Chinese on political issues. At this point, Biden’s choice of leaving the sanctions in place does nothing other than provide a buffer to let countries outside of China get a running head start in a very very long race.

Please don’t assume I’m playing the Trump is evil card. With the rollout of 5G, if Huawei would have been able to keep going as they did, China would have been able to draw trillions of dollars more from the US treasury. I don’t agree with how Trump prevented this, I would have rather seen an executive order demanding Cisco or another US company produce a competitive offering. But it did accomplish mitigating the risk of China being able to simply collapse the US economy on a whim. The European approach of working on an open source RAN was a far better approach.

America, Taiwan make semiconductors their top trade priority at first-ever 'Economic Prosperity Dialogue'

What happens when...

One of the world’s top economies... heavily dependent on semiconductors is told that no one in the world is allowed to sell them semiconductors?

The easy answer is, they invest massive amounts of money, time and resources to never need to buy semiconductors from another country.

Then they build up enough manufacturing ability to produce semiconductors for every other country who doesn’t like the impending threat of being cut off.

Then they do it for a lower cost than any other country.

Then they weaponize their capacity and use extensive government grants to economically attack countries like Taiwan... after all, why not simply give semiconductors away for free until TSMC can no longer afford to keep their doors open?

Of course, in order to stay competitive, that country will innovate as well. They will make sure they’re not just competitive, but after throwing money at racing to equal with publicly traded companies in the US and Taiwan, they will have a momentum already in place to also surpass them.

So what happens when someone makes a decision that threatens many of China’s largest and most influential companies and treats them like this? You think if Biden gives a little and agrees to sell them chips again that China will just stop their almost space race like efforts to become entirely independent?

Who knew? Hadoop is over, says former Hortonworks guru Scott Gnau

Re: @tfb This is why

I've been saying this for some time about COBOL. (Oh and I work with FORTRAN in HPC quite often)

People make a big deal about COBOL programmers being in short supply and that it's an antiquated language. Honestly though, what makes programmers really confused about it is that you don't really write programs in COBOL, it's more of a FaaS (Serverless, function as a service) platform. You write procedures in COBOL. The procedures are stored in the database like everything else and when a procedure is called, it's read from the database and executed.

The real issue with "COBOL programmers" is that they don't know the platform. The platform people are usually referring to when they say "COBOL" is actually some variation of mainframe or midrange computers. Most often in 2020, they're referring to either IBM System/Z or they're referring to IBM Series i ... which is really just a new name for what used to be AS/400.

The system contains a standard object storage system... or more accurately, a key/value store. And the front end of the system is typically based on CICS and JCL which is job control language. IBM mainframe terminals (and their emulators) have a language which could be kind of compared to HTML in the sense that it allows text layout and form entry as well as action buttons like "submit".

Then there's TSO/ISPF which is basically the IBM mainframe CLI.

What is funny is that, many of us when we look at AWS, all we see is garbled crap. They have a million screens and tons of options. The same is said for other services, but AWS is a nightmare. Add to that their command line tools which are borderline incomprehensible and well... you're screwed.

Now don't get me wrong, if I absolutely must use AWS, it wouldn't take more than watching a few videos and a code along. I'd probably end up using Python even though I don't care much for the language. I'd also use Lambda functions because frankly... I don't feel like rolling my own platform from scratch. Pretty much anything I'd ever need to write for a business application can be done with a simple web server to deliver static resources, lambda functions to handle my REST API, and someplace to store data which is probably either object storage, mongodb, and/or some SQL database.

Oddly, this is exactly what COBOL programmers are doing... and have done since 1969.

They use :

- TSO/ISPF as their command line instead of the AWS GUI or CLI tools.

- JCL to route requests to functions as a service

- CICS (in combination with a UI tech) to connect forms to JCL events as transactions... instead of using Amazon Lambda. Oh, it's also the "serves static pages" part. It's also kind of a service mesh.

- COBOL, Java or any other language as procedures which are run when events occur... like any serverless system.

It takes a few days to learn, but it's pretty simple. The hardest part is really learning the JCL and TSO/ISPF bit because it doesn't make sense to outsiders.

What's really funny is that IBM mainframes running this stuff are pretty much infinitely scaleable. If you plug 10 mainframes in together, they practically do all the work for you since their entire system is pretty much the same thing as an elastic Kubernetes cluster. You can plug in 500 mainframes and get so much more. The whole system is completely distributed.

But like you're saying FORTRAN is its own entire platform/ecosystem is entirely true. Everything you would ever need for writing a FORTRAN program is kind of built in. But I will say, I would never even consider writing a business system using FORTRAN :)