The hour grows late, the enemy are at the gates... but could Intel's exiled heir apparent ride to the rescue? Intel has been many things. It has been a struggling startup, surfing the sudden successes and near-death experiences of 1970s Silicon Valley. It's been a relentless powerhouse, ruthlessly exploiting its dominance as personal computing rode Moore's Law to commercial and social revolution. Now it's in danger of seeing its …
The less powerful chips should be tagged as efficient and the more power hungry tagged as high performance. People purchase based on the features most important to them. Atom CPUs are an excellent idea as long as they can do well at what they are used for. A product gets a bad name if it can't do well what people were expecting.
That would be the RPI 2. Good enough to
+ store your office files
+ store your code
+ run your personal web server
+ run an efficient discussion board (not PHPBB !)
+ run tinyproxy to filter out the tracker stuff
+ run an XMPP server for your self-controlled messenger
+ run the sn USENET server for your self-controlled discussion board
+ compile code using tinycc
It consumes 3Watts and costs less than 30 Euros once. Use ddnss as a dynDNS proxy service.
The OP specifially mentioned ECC, QAT and 4x 10GBit and you're recommending RPi which is something altogether different??
I really appreaciated your article and I think you did a great job of outlining the situation Intel is in and the person Gelsinger is (and I envy you for having met him more than once).
I hope Intel is going to pull through for the same reason I'm glad AMD is having a moment in the sun : we need competition. We need people who can think of new ways to improve performance, reduce power consumption and generally make computing an even better experience.
So I'm looking forward to what Gelsinger is going to cook up.
Another industry expert's take on ts stratechery intel-problems = Intel need to split into a fab company and a chip design company, but they can't because they are behind on fab technology and their x86 server monopoly is going away
The new Intel CEO must follow the below check list :
* Stop rebranding CPUs year over year!
* Stop use rebranded CPUs to justify new motherboard Chipset.
* Buy a top notch GPU maker with lots of patents, like Imagination, instead of starting from scratch;
* Learn to build GPU drivers! Learn OpenGL, Vulkan and DirectX, for God sake!
* Listen your consumers! And stop trying to fooling them!
* Create your own ARM compatible CPU, even as hardware on the fly translator!
* Talk with AMD about a new CPU instruction set specification, otherwise you both will sink under ARM revolution - remember: the last sucessfull instruction set you've designed was the 32bit x86 for 80386. And remember you had to license the AMD64 from AMD to not sunk with the Itanium madness.
Different budgets and different technologies. Yes the web hosting would cost a lot less if they needed a less powerful server but it's a difficult enough task to get a complex website built without ruling out technologies that the designers are familiar with.
Wordpress is hugely popular but it annoys me that it takes 500ms to switch to the next page. Throwing computer power at it makes no difference beyond a point. Re-write the thing on more of a static basis and it would be very fast and require much cheaper hosting. But then it would not be Wordpress.
Define 'equivalent'?
It's often the frameworks that take up the resources, less so the language.
Cost applies elsewhere too, you've got to find developers to write and maintain the code you run. There are fewer pascal developers than Java ones. (Or at least there are round here, I work for a place with a pascal legacy, and we have issues finding devs to maintain it).
The infrastructure budget is often separate from the software project budget.
So software demand the fastest servers possible, then code away like crazy trying to meet project deadlines.
Infra teams have to trade off a usually fixed budget against service levels (performance, reliability, availability).
Each one can't really control what the other one does.
Me, I'm a performance-focused software engineer who lost interest when Java/C# etc. became popular. Now I work with FPGAs (no-one has yet figured out how to bog them down with Java).
I suspect if any of those things are seriously on the new CEO's list, Intel are well and truely screwed.
Why? They are either marketing or irrelevant to Intel at present.
Intel need to be able to produce leading edge chips - it doesn't matter what they call them if they can't produce them. Adding a better GPU won't help if the competition is faster. And an ARM CPU won't help if the revenure per wafer that drives your profits are rapidly declining while the costs of each new chip generation are rising.
Intel has to make leading edge chips again. Realisticly, that is either fixed already (fabs built and are starting on A0 silicon - A1 if they are ahead of schedule) or hopefully Gelsinger's name is enough to get them the next round of finance to fix it.
When the new chips are rolling, they won't stop the annual rebrands, only AMD's marketshare will stop them doing stupid things with chipsets, ML is where the money is rather than GPU's (although there is a lot of similarities) and they have 3 architectures already in addition to PC GPU's that do very well accelerating corporate requirements for video and math.
Consumers are a tiny share of revenue and profit - corporates and cloud account for the vast majority.
And do you think Intel would license ARM from nVidia? RISCV would be more likely but it presents the same issue as ARM - low margins versus x86...
The rebranding ones are easy. It pisses everyone off when they remove a pin just to make a chip incompatible. I think Intel do have knowledge of ARM.
Intel are also pretty good at OpenCL, OpenGL, Vulkan and DirectX. All the i-core CPUs do these.
FPGA and ARM could be added to the PC. Maybe in the GPU. Yes they do need a GPU.
Intel sounds all brutal and patriarchal, whereas a fragrant name like Penelope will make people think of soft and silken lady hands and clothes and stuff and inclusion. Maybe supply some free lipstick with every chip. And call them "gems", not chips, because chips are fattening.
I think that the Penelope Kiss-me 2 sounds far more approachable that the Intel Thorhammer Graphomatic Speculum Hardrod Megatron 3000 or whatever masculine horror their latest chip, sorry gem, is called.
They should make the gems pink too, and maybe put feathers on them or something.
That'd do it. Let me know if you'd like an article.
You don't have to be into pastel colours and Swarovski gems to feel that computer hardware doesn't have to look like a Lamborghini stealth jet as drawn by a 15 year old boy with fluorescent marker pens.
Hmm, I'm thinking a workstation that looks like a scale-model mini Cray supercomputer - the cylindrical ones with the vinyl-clad donut-shaped bench seats around the base. The supplied action man figure comes with a chance of clothing, white shirt and pocket protectors, or shorts, tie die t-shirt and beard.
Or: some cases for NUCs that are 1/4 models of Silicon Graphics workstations... the purple one or the weird teal coloured one. Soooo 90s.
What do I know? My computer looks like a Sony transistor radio, and my speakers are big and have wood veneers!
Have an upvote for the Silicon Graphics references - while working as in my first full IT Admin role after graduating, at a local University back in the "Windows for Workgroups running on top of Netware 3.11/3.12" days, I was then given the task of looking after our Science Faculty's SG Workstations... Indigos, Indigo2s, etc. What a way to learn unix (well, Irix actually...). Just don't remind me of the seriously broken TCP/IP stack that Irix had been cursed with, which meant that getting things like TCP wrappers to work was... interesting, to say the least.
Working in IT during the 90s was such a hoot!
‘Penelope’? Seriously? I’d be too busy looking for the old beggar with the bow and no sense of humor. And I’d be really careful about doing anything which might be interpreted as drinking from a handy cup.
The youth of today, no education in the classics...
..by reusing the Datamaster 8085 motherboard and not waiting around for Motorola delivery of 68000 in quantity.
Without out Estridges decision Intel would have been just another failed 1970's DRAM company.
Nothing about Intels last 40 year history makes any sense unless you understand Intels success was the result of blind luck and nothing else. And the reason why Intel has been so unrelenting in trying to crush all competition is that they know that. They have always been a technically mediocre company. Which is why they have failed at everything else they tried.
Just look at the x86 instruction set. Which was already looking klunky and obsolete in 1979. It took AMD to put any kind of sanity on it and even then its still painful to use. When writing x86/x64 you still have to jump through hoops to do very basic operations that were straight forward in other instruction sets forty years ago.
I'll be glad of the day when Intel fade away to irrelevance. And the abomination of the x86 architecture with it.
Intel delenda est is what I say..
Anyone got a shipload of salt that they can deliver to an address in Santa Clara. I can give directions.
Three points:
Almost no-one actually writes x86/x64 -- we have machines for that now.
Instruction decode is about 1% of the die and only that big because it is done 16-way parallel.
Intel *were* very good at the physical process of making chips. Their current woe is almost entirely due to a succession of mis-steps in that area. *This* has to be Gelsinger's priority.
No one writes asm? Really? Maybe in your world but not in mine. Try writing a compiler code gen without writing asm. Or some of the fun device driver glue. Or how about compiler bug fixes. Inline asm to fix someone elses bug.
Some of us do work where we have to check the C disassem to make sure the correct instructions and machine code is actually generated. One of the skills you pick up in my line of work is to be able to look at C/C++ code and predict exactly what the asm should look like. Could do it for 68k, MIPS, PPC and ARM but x86/x64 still makes my head hurt if I have to look at it too long.
The reason why instruction decode is 1% of the die is because 98% of the die is cache memory. L1,L2 and L3. As for 16 way decode it does not do you much good if the instruction reordering scheduler in the compiler does a crap job. There are only so many compiler optimizer phase sins you can fix in silicon. Usually only the simplier ones.
One fact has not changed in the last 40 years. I you know how the silicon works a little judicious rearrangement of the source code going into the compiler make a huge difference to real world performance. Not a lot has changed since the first time I tailored C code so as to keep the IPU's pipelines single cycle result for a superscaler CPU. Which the compiler superscaler instruction reordering could not do. That was with a PPC which was pretty straight forward to write a instruction reordering optimizer for. It actually did a pretty good job most of the time but not when I needed it to single cycle 4 results per clock cycle. The best the compiler could do was one and a bit.
Now optimal instruction reordering by the compiler for x64, you must be joking. Have you looked at those instruction timing tables published by Intel in the hardware manuals. They make calculating ptolemaic epicycle tables look straightforward.
The world does not change that much.
Thirty years ago most of the JS jockies would have been writing BASIC and 50 years ago, RPG IV. No problem with that. You use the tools that best fit the task at hand. In my case its often scraping the bare iron with hand crafted bits.
This post has been deleted by its author
Networking chips, I've usually found them pretty reliable from Intel though not always cutting edge.
That said the last few highend bits of kit have used Extreme networks chipage and I still admit I have no loyalty to Intel network chips because I just want something that works.
> I said earlier that Pat Gelsinger was transparently decent, a point that many godless Brits conflate with his Christianity, something he lives with an openness unusual even for an American. These things are connected but not contingent, he's not the sort who needs to get his morals from a book
There's no need to downplay his faith. Super unlikely you'd do it with any other one.
There's no need to downplay his faith. Super unlikely you'd do it with any other one.
How does that downplay his faith?
As someone without faith, if it's any help, I read it as 'Pat Gelsinger is a Christian, he is also a decent human being. These things are not connected'. Which seems pretty fair. (Evidence would suggest that not all decent human beings are Christian, just as not all Christians are decent human beings).
I took the 'morals from a book' comment as a statement of his inate decency, as it was ingrained rather than affected, or learnt.
Thing is with faith, it's deeply personal, so, like musical tastes, it's best not worrying what other people think; that way only leads to trouble.
"for the 80486, which was a modest success in the 1990s."
I was working at a chartered accounting firm in the '90s. It eventually got to the point where we'd left 1-between-three, through 1-between-2, to everyone getting a PC on their desk. Being a corporal rather than a private, I got a 486 DX-2 66. That thing could fly. I think I had a 14" CRT. Ah, the mameries (yes, we got connected to web not long after).
Intel is a company that mostly works well. What they are able to do whilst stuck on 14nm is amazing. They get far more out of their silicon than AMD.
How do you fix the problems without breaking things that work in such an established company?
They have a huge fan base so to speak. They should keep their fabs but be a bit flexible in using other contract fabs. I can see a massive problem when both AMD and Intel are trying to get TSMC to make their chips. They need to do a deal with another fab to get some talent in to sort out their process.
Engineering is supposedly based on science but I believe the engineers go beyond what the scientists are able to theorise. This is done by trying things and seeing what works. There is more to be discovered about light waves which will shrink designs further. I suspect TSMC have stumbled on this and Intel have not. When the scientists catch up with the engineers then expect amazing advancements.
Engineering is supposedly based on science but I believe the engineers go beyond what the scientists are able to theorise. This is done by trying things and seeing what works.
That's pretty much what science is. :) Engineering is a science too.
Also, it's probably worth remembering that one fabs nm is not quite the same as anothers.... (they measure different features when quoting gate size).
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
