Intel processor security flaw requires OS kernel level fix

[spacein_vader]

I imagine it would be big sueballs as discounts don't return their money there and then. If we take it as worst case then 30% of AWS and Azures performance was just destroyed in one single sweep. For a single PC, 30% is a lot but manageable but across billions of dollars of cloud infrastructure, I don't think discounts will cut it.

Maybe refunds rather than discounts then. I imagine Intel would rather do that quietly with each big supplier than publically get sued by them.

[ik9000]

AFAIK P4 and Pentium M were Out of Order processors, so are potentially vulnerable. Atoms prior to Bay Trail, OTOH, should be fine

Yep, according to the beeb it affects all processors going back as far as 1994 - not sure which exploit and how true that is, but what would that even be? An old 286?

[DanceswithUnix]

Processor fanboys aren't Intels problem, particularly as it seems light desktop work and gaming are going to be impacted only lightly performance wise. Their big issue is going to be the big cloud hosts and VM providers. Things like database writes and some heavy compute work gets the biggest performance hit and the likes of Amazon AWS and Microsoft Azure will feel the difference hugely. Their respective top teams will NOT write it down to experience just "because Intel". They'll be out for blood (and by blood I mean cash.) There'll either be some very big sueballs launched by them or more likely, some VERY deep and long lasting discounts on any new Intel kit they order.

There are some big assumptions here.

Database accesses are slowed down by going through the loopback 127.0.0.1 network stack because code is written to assume that is fast but bouncing off the kernel is now expensive. Attaching to a remote database server will include more work in the network stack, wire delays and fewer kernel transitions, so probably won't be impacted as much. So databases might be impacted on a case by case basis.

Phoronix showed compiling code to be hard hit, but on Windows it won't be. Why? Linux assumes reading lots of small files is cheap so that is what you do, on Windows it never was cheap so the compiler crunches all those little header file reads into pre-compiled headers resulting in magnitudes less filesystem access.

My gut feeling is that gaming is the other way around, Linux works differently and also has all sorts of porting crud in the way so won't be hurt much, but AIUI Windows will be hurt more as it has more of the driver in kernel space.

I got the impression that Windows fast track update already had the patch, are there not benchmarks floating around already?

[Xlucine]

The register claims that the windows patch is available here:
https://support.microsoft.com/en-us/...date-kb4056892

Since intel have known since june, it's odd that the 8700k is affected so badly. I smell a class action lawsuit...

But that's the problem - they'll probably still order intel, so in terms of boosting AMD, it won't, not in the same way that switching camps would do.

For a VM provider epyc offers more cores per dollar, and supports more memory, and AMD hasn't been shipping systems with this glaring bug for months.

[Tabbykatze]

But that's the problem - they'll probably still order intel, so in terms of boosting AMD, it won't, not in the same way that switching camps would do.

Agreed, the main issue is that there is little/no optimisations for AMDs product sector at the moment. But this might spark developers to more aggressively push to optimise for AMD because of this.

[rabidmunkee]

Should this invalidate all those intel benchmarks then? and if not then all the benchers are going to be trying to use the old o/s's prior to the fix to get that extra little bit out of it.

Whats the rules on those sort of things? are there any?

[ik9000]

Is it me or is anandtech.com surprisingly silent on this at the moment? Apart from a small twitter thread suggesting it probably isn't that bad (not entirely anandtech's comments) there doesn't seem to be much on their website that I can find. Odd no?

[ik9000]

Should this invalidate all those intel benchmarks then? and if not then all the benchers are going to be trying to use the old o/s's prior to the fix to get that extra little bit out of it.

Whats the rules on those sort of things? are there any?

not sure, but I'll be benchmarking my systems intensively this weekend at stock and OC before installing any patches so I can do a compare and contrast once the patches are installed. I'm most concerned about my laptop which I paid a good whack for to get something that could handle low-latency sound recording. If that suffers I will not be amused.

[Biscuit]

Is it me or is anandtech.com surprisingly silent on this at the moment? Apart from a small twitter thread suggesting it probably isn't that bad (not entirely anandtech's comments) there doesn't seem to be much on their website that I can find. Odd no?

Not all that odd really.

[ik9000]

Not all that odd really.

Surely, you don't mean... you don't do you? Not anandtech, surely?

[Tabbykatze]

Should this invalidate all those intel benchmarks then? and if not then all the benchers are going to be trying to use the old o/s's prior to the fix to get that extra little bit out of it.

Whats the rules on those sort of things? are there any?

Basically the benches still apply but they could be invalidated by this update. Because of the uncertainty of the performance metrics being untainted by the Kernel Page Isolation then taking any Intel benchmark prior to the patch at face value would be a bad idea. I would imagine all the major testers and benchers are wondering if they're going to have to re-do all of their benchmarks just to update their scoreboards.

Because if games are going to be relatively unaffected, that's one thing but users don't always use their computers for one thing. I use my computer for compiling, video editing and manipulating databases as well as games so actually this patch could affect me a lot (using AMD so unlikely). Benchmarks these days aren't just games, i mean both Hexus and TTL have game benchmarks in but they are not the main part of the review any more. It's all the sisoft, PCMark and other raw curnchable data benches.

To me, I will be looking at every benchmark and subtracting ~25% from the performance on Intel vs AMD as a ballpark until its confirmed how much of an impact is where.

[Biscuit]

Surely, you don't mean... you don't do you? Not anandtech, surely?

Can I get a cheer from everyone who looks at anandtech for impartial reviews.....

[kalniel]

I said "considering", and I'm going to go out on a limb and suggest that the thought has crossed your mind even before this announcement

Heh, only for the last two years or so. But I literally pushed the button last night (though immediately regretted not going for a x5675 instead, but oh well )

[DanceswithUnix]

Since intel have known since june, it's odd that the 8700k is affected so badly. I smell a class action lawsuit...

Turn around on silicon is slow. Months to prove a fix, then 3 months for production silicon at the new mask rev level to turn up in the market place.

[MrJim]

There's a Youtube video giving some before & after benchmarks following the emergency Windows patch:



Edit!

As they mention in the video, they don't look at server loads in their assessment, but I think that won't be too much of a concern for most people on this Forum...

[DanceswithUnix]

No drop in games performance then, nice.