Posts by theblackhand 881 posts • joined 1 Oct 2009
I'm wondering about power as well - while the standard may support 4GB/s there is also a lower 3GB/s option versus SDXC maxing out at 1GB/s which is much higher than you would typically see on an SDXC card reader.
The interface appears to be rated at 1.8W vs current cards that are under 100 mW.
Intels q4 server CPU sales were off the charts - up around 49%
I know Nvidia contributed a chunk of that with their Geforce Now DC's, but hadn't hard which of the other big cloud providers took the rest.
AWS/Azure/Google had all been delaying spending waiting for new chips, any evidence that it was AWS that won?
"I well remember the RISC anxiety at Intel when I worked there 35 years ago."
And when Intel moved from pure x86 CISC to x86 CISC instructions decoded to µops to run on a RISC like architecture with the P6, what happened to that anxiety? Sure, Intel hedged it's bets with Itanium/VLIW but reality wasn't kind to that...
"the process that sees hordes of overpaid junior lawyers and accountants poring over every tiny detail of a business before going ahead with a merger or takeover."
In most accountancy or legal firms, overworked is more accurate than overpaid. The juniors do the donkey work (usually with significant amounts of travel and unpaid overtime), the senior managers are paid well to apply the whip to keep things going and the partners rake in the money and put the entertainment on expenses.
Oh...and the juniors are responsible for all mistakes.
"they can't _buy_ enough outsourcing capacity to catch up because it's all booked out."
To address outsourcing high performance chip designs to another fab - you will likely take 1-2 years to redesign a working CPU operating at >2GHz if you take a working design from one fab to another.
Intel/TSMC/Samsung all have significantly different processes - a working design for one fab does not instantly translate into a working design at another fab, and if you are dependent on a design that is outside of the conservative fab design rules, you risk low yields and poor performance. For examples, look at AMD/nVidia GPU's on identical TSMC processes where one card substantially under-performs or has low availability - typically the first GPU or largest GPU does not clock as highly as its rival because of design issues. A year later, we see a decent boost as the updated chip addresses the issue.
"Being able to replace a full rack of Intel with two AMD boxes is music to the ears of any data center professional."
The "rack full" of dual CPU Intel boxes MAYBE being replaced by "less than a rack full" of AMD boxes.
Licensing makes high core counts hard to justify for Enterprises unless they have applications (like HPC) that aren't licensed per core.
The high core count chips are power/cooling hungry making it hard to sell them into cloud environments or enterprise blade chassis where density is more important. That's not to say moving from 20 servers per rack to 10 while lowering connectivity costs and increasing core counts doesn't make sense.
IO (typically storage) also kills your per rack density - you are likely limited in the amount of IO you can deliver to each rack. So moving from 20 x 8-cores to 2 x 128 cores likely doesn't unless you can avoid storage/IO/network/memory bottlenecks by scaling up connection speeds cost effectively.
Can you mix your new CPU's with your existing CPU's in a VM environment? Or do you need to start a new farm. That makes switching vendors hard unless your current environment is end of life and ditching it is an option. If you have a larger environment and replace a fixed percentage each year, changing may require considerable planning/budgetting.
And finally, what does it cost all up?
If any of the Super7+1 (or is that +2 now nVidia are trying to compete for mobile gaming?) announce major deals with AMD in 2020, AMD likely double their server market share.
TL;DR: high core counts look great, but AMD's pricing, power usage, low-to-mid range core counts and being able to deliver are more likely to win them this round.
"History is important, I'm not suggesting otherwise, but using it as a barometer against today's employment opportunities is bonkers."
Then I would suggest re-examining post-WWII history (you can go back further if you want but this is likely to be sufficient to see details) to see just how far we have come and how many jobs have changed or disappeared completely.
"We all know that the roles needed in society change and automation and AI (whatever that means in reality) drives this."
Yes...and in general, society as a whole has improved on the back of change. And generally change has improved both the lives AND jobs of the less well off. Look at jobs that had high accident rates 10-50 years ago and how they are done now - machines have replaced people and the people manage the machines with a subsequent increase in rates of production with less injuries.
You look and see Deliveroo and Amazon Fulfillment Centres and I would suggest both are likely to become more automated in the future as the human acts as as a robot. And there are questions around safety/injury.
What I see is an increasing requirement for human education and knowledge and the potential for that to drive future change.
Will people still be left behind by this? Yes and the aging population is likely to require a lot more social care as cheaply as possible. At least until we figure out how to automate that.
Will society as a whole benefit? I believe so. Based on history. Trying to retain the status quo has been where society crumbles and revolution is needed to move forward again. When there is change, those at the bottom have something to strive for. And technology revolutions tend to be a lot gentler than political revolutions.
"It's somewhat disturbing to see how comparisons are being made between working and living conditions today and a 100+ years ago, as if that means anything."
We navel gaze at the future where AI will automate people out of jobs while installing computer systems that allow one person to do the tasks of many or write code or reports that allow businesses to operate more efficiently.
So yes, history means a lot...those that are aware of history are aware of the significant changes that have happened over the last 200 years to get the standard of living above "you'll probably die soon" to "you better plan for old age, you've got a lot of years ahead of you".
I suspect it may also be part of the current generational gap - those brought up in cities where everything is available as long as you had money versus those from rural areas or before much of the current service industry was available who can make do with alternatives or without...
And yes, I'm an old grumpy git before my time.
<random dump of useless information>
Before PCI-bus IDE HBA's, ISA IDE HBA's required an interrupt per channel which limited the number of hard drives you could use (you could usually manage two-channels of 2 drives each once you added a few network cards) and generally multiple hard drives were the only way of expanding capacity in the days where the biggest hard drives weren't big enough. From memory, the Netware IDE drivers weren't great either although that may have been a hardware bottleneck rather than just a driver issue.
SCSI HBA's allowed upto 14 drives per HBA...more than enough.. And the drivers were better. If you could get updates.
</random dump of useless information>
Well Lynch probably made a few million when he sold his company to HP. Looks like it was ~£500 million (source: https://www.computerworld.com/article/3416571/update--mike-lynch-leaves-hp-autonomy.html). It should be relatively easy to find based on shareholdings and bonuses from the sale but I can't find a better source...
You'd never believe how much HP paid for Autonomy...
But what about all the good work the board did with share buybacks to prop up that share price and limit the decline to only 26 per cent? Anyone? Please?
*tumbleweeds roll by*
"Is it going UP to a graveyard orbit or being unusually de-orbited into the atmosphere?"
It's going up - there is a significant quantity of fuel remaining (73kg at present) which may account for the life time extension.
Regarding the batteries, the satellite appears to be OK operating on solar panels but the concern is that the coming eclipse season will require battery operation leading to the potential for catastrophic failure. Eclipse season starts on Feb 25th.
"If Boeing designed an aircraft for a 12 year lifespan without ensuring that it would not still be flying around at the same altitude fifteen years later, you might think someone had been careless."
Is the comparison to an aircraft valid? An aircraft is maintained as part of normal operations while a satellite is almost entirely unserviced during its lifetime. Aircraft parts prone to stress are regularly replaced to minimise risk and anything with an operational life is replaced before it is likely to fail.
And aircraft operate in a less hostile environment than space - the fault that has end the life of this satellite is likely repairable if it occurred in an aircraft and if it was caused by being hit by an unknown object, then that is a risk satellite manufacturers take and attempt to workaround with additional resilience. But resilience doesn't protect you from multiple failures particularly as you reach the later stages of your mission where you may already be operating with failed components.
According to the wiki article, this was launched in 2005 with a 12 year lifetime and was currently being used as a backup satellite.
While the battery issue is undesirable - are there any details of the fault (i.e. overcharging due to a damaged component or control system?) as bashing Boeing for this seems a little harsh unless I'm missing something.
"Suggest you look at the successful network forums eg. IEEE802 (specifically 802.3), WiFi Alliance..."
Are you sure it's bollocks? Using the comparison to Ethernet, USB has had to change from 4 pin plus ground via a separate pin or the ground shield to 9-pin including ground in order to separate high power and high bandwidth cable pairs. Ethernet avoided this by already having 4-pair connectors for UTP from the beginning because there weren't cost effective digital modulation methods available to provide higher data rates when it was introduced..
USB was designed to minimise costs in early implementations - while there is an argument that it could have provided a more future proof connector at version 2, the fact is it didn't. Given competing interfaces that didn't minimise cost and failed, I'm unsure if you can choose a single interface that is universally "good" and "cheap".
Is Ethernet or USB able to go beyond 10Gbps with current connectors? Unlikely - they already use QAM256 encoding to achieve those rates and Ethernet has already moved away from UTP connectors beyond 10Gbps with 10Gbps UTP Ethernet having a number of downsides (i.e. high power use, high latency) that make that unlikely to change.
Is 100W/10Gbps sufficient for a general purpose device-to-device interface? I would say yes - marketers will say no...
Datacentre is DCG and is currently $7.2bn/quarter or $23.5bn/year
CCG (desktops/laptops) is $10bn/quarter or $37bn/year
Even if AMD are unsuccessful in obtaining mobile market share in 2020, they will likely impact revenue as Intel will need to reduce prices/margins to maintain volume.
Intel could easily see a 10%-20% decline in CCG revenue...
Intel care more about DCG revenue because there's a very slim channel to support so better margins than CCG plus DCG is a growth area whereas CCG is flat at best.
"the mitigations bring performance down"
The software mitigation's reduce performance - the hardware mitigation's found in Cascade Lake generation chips address Meltdown/Foreshadow in hardware and the remainder of currently known Spectre issues to a similar level as other x86/non-x86 SMT platforms.
While OS/software fixes are still required (and may impact performance) to fully address issues, that isn't an Intel exclusive issue.
"copywriting APIs will benefit absolutely nobody, even Oracle themselves"
Long term, yes.
Short term, Oracle claim copyright over Android and claim very significant damages against Google, possibly to the point of destroying Google or breaking it up and making Oracle very rich.
Day n+1 when the world is burning doesn't worry Oracle. Yet.
This is basically evolution in the home router space.
To make installing a new router, Netgear (and others) initially directed users to access the router via the IP address. Which worked as long as users didn't have to change the IP address. A DNS name was deemed friendlier and browser makers were discouraging the use of IP addresses for security reasons. Then came the push to move to secure communications, so Netgear used self-signed certificates. And then the move to ensure certificates were valid so Netgear obtained valid certificates. And now the move to avoid distributing public certificate keys.
The bit missing here is how do you opt out of security policies that provide little additional security (in most cases - I'm assuming internal access to a device you own with a default policy of no external access to the device) without encouraging user behaviour that is undesirable (ignoring browser security warnings)?
My reading of what Netgear are doing is trying to get a valid hostname/certificate that allows connections from a browser without security warnings for invalid certificates etc. The .com/.net domains allow Netgear to register valid domain names and get trusted certificates to avoid browser warnings around insecure connections or untrusted certificates.
I'm not sure a different hostname or self-signed certificate where the private keys reside locally on the router is the issue - it puts you back to the point where you need users to click through security warnings to begin router configuration.
I can imagine a process for getting valid certificates generated per device IF you have a working Internet connection (i.e. connect to trusted host using a public key embedded in firmware, register unique DNS name via trusted host and trusted host issues valid certificate, redirect from http://router.local to https://my device name>) HOWEVER:
- what happens if there is no working Internet connection yet?
- will this still cause browser warnings when you connect the first time you use the device?
Given the target market (consumers), I'm not convinced the risk of embedded certificate keys outweighs the advantages of convenience for the user, particularly when the troubleshooting steps would like be accepting untrusted connections which establishes bad habits. Using short certificate lifetimes, separate certificates per device model and auto-updates for firmware would significantly reduce the risk AND allow Netgear to respond quickly to any threats that became public.
Or you ship a setup application that allows you to bypass browser security warnings and is only used if the automated setup fails.
Unless I'm missing an obvious solution...
"They could have called it something else, I don't know, like "Motlin"."
This is exactly my point - if MS had called it something else AND not licensed Java from Sun, the lawsuit would have gone in a completely different direction (or Sun may not have found grounds for a case).
Instead the case was about failing to uphold licence requirements for Java as required by the Java licence - it was purely contractual and MS had not complied with those contracts.
Google vs Oracle is the parallel Java universe where Microsoft did the opposite of what they did in this Java universe.
Sun vs Microsoft was around attempting to coerce Java to Microsoft's standard - Microsoft were sued by Sun for failing to implement the standard correctly while still using Java trademarks and Microsoft were fined as a result and discontinued the MS JVM.
I would dispute that the Sun vs Microsoft case mirrors the Oracle vs Google Java case outside of both being based around Java - drawing parallels with the effects on software development for the two cases isn't valid.
"What's needed is long term spending on prevention, such as not having forests right up to towns and a fire service that is able to contain fires that break out. But you don't get that from donations like this."
And what if the towns build right up to the edge of forests? Part of the problem is increasing populations and encroachment on existing forests as much as building too close to forests.
Combine that with fuel reduction measures being restricted by properties, changes in water/land usage within affected states resulting in dryer areas and increasing temperatures due to regional and global changes and the extent of the issue becomes clear. The comparison I would make with the UK is flooding - the causes of flooding are as much due to changes in river management and land use as they are to property locations.
Donations will allow the immediate effects to be reduced (i.e. replacing damaged or lost property) and increasing preventative measures because the wider environmental issues are likely to take 10-30 years to address assuming the Australian government has the cajones to tackle the issues rather than just let rich industrialists (of which Bezos is not one) ride roughshod over the environment to allow them to make a few more dollars..
As for the animals, the worst types of animals - politicians - appear to have been largely unaffected by the disaster.
"What HP really want is a win in this civil case (balance of probabilities) to use in their upcoming criminal case (beyond all reasonable doubt). Doubt if they'll get it."
What HPE want is not to lose. While winning this case may allow them to take assets from Lynch and Hussein and ensure both are imprisoned in the US for their part the flip side is that if they lose, future cases become "shareholders vs HPE" or "shareholders vs former execs of HP" and that is what worries Meg and HPE the most.
Leo has demonstrated he wasn't up to the HP CEO role with his testimony in this case, justifying the board's actions to remove him. The reality is that the board supported him and this is just as much about HP's actions post-purchase under Meg and the actions taken. Were they in HP's best interests or the boards?
Is this the driver to finally get ECC as standard in servers AND consumer devices?
AWS papers suggest the premium for ECC memory (i.e. ~1.25x cost of additional memory versus the current 2-3x premium) would all but disappear if consumer devices were fitted with ECC as standard.
The increases in stability would likely cover those costs as memory sizes increase (assuming error rates in the 2000-6000 errors/Gbit per billion hours of operation is still valid).
The Register - Independent news and views for the tech community. Part of Situation Publishing
Join our daily or weekly newsletters, subscribe to a specific section or set News alerts
SUBSCRIBE
Biting the hand that feeds IT © 1998–2020
