• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

Intel bug incoming? Meltdown and Spectre exploits

humbug said:
Not that bit it isn't ^^^^



The actual performance of Ryzen 1 ended up significantly higher than AMD and Glofo claimed.

Of course they could be lying this time round but Ryzen 1 is at such a low base in terms of Mhz "more than 10% higher performance" is frankly not even remotely unbelievable, i mean the best and highest clocked chip, the 1800X is clocked at 3.7Ghz all cores, if you bring that up to just 4.1Ghz its an 11% uplift in performance, i'm quit comfortable to say between 4 and 4.2Ghz would be Ryzen 2800X new base clock.

Intel are not the whole industry, some integrate may still exist outside of their offices, we have no reason to disbelieve Glofo or AMD on Ryzen.
Click to expand...

Yes it is!

Its a canned benchmark, FPS figures in civ6 are moot. Its turn time that indicates performance as already mentioned.
Yes @pete910 I agree its getting annoying!
 
Mods, can you please stop Gavin and Humbug from bringing every thread to a title for tat argument.

This is NOT the gpu forum, surely the 3 strike rule should be enforced here.

We're all sick of Gavin's crusade to pee on the cornflakes of anyone who praises AMD or anything Humbug says, and were all tired of Humbug taking Gavin's bait and continuing to fuel the fire.

It's tiresome.

I would put the pair on ignore but I actually find some of what Humbug says useful.
 
gavinh87 said:
What I'm more surprised is the bios updates have close to zero impact where the windows updates are the biggest offenders. (In this video anyway)
Click to expand...

This suggests that (for the game sequences in the video at least) the Spectre mitigation (IBRS - microcode + OS patch for Variant 2 CVE-2017-5715) has negligible impact, whereas the Meltdown mitigation has some. However, without an equivalent microcode update for AMD, I don't know whether Spectre Variant 2 mitigation on Windows will have a similar sort of impact. On Linux, the microcode update isn't mandatory as there is the alternative of Retpoline which has a minor performance impact.


On the phantom claims of a 30 percent hit, here is what the breaking news article actually said: "The effects are still being benchmarked, however we're looking at a ballpark figure of five to 30 per cent slow down, depending on the task and the processor model. More recent Intel chips have features – such as PCID – to reduce the performance hit. Your mileage may vary".

This is talking CPU performance, and not end user FPS, so there's no surprise that gaming FPS on recent CPUs are showing mostly very little impact - it's not a linear relationship between CPU performance and FPS, never mind all the other variables.

DragonQ said:
known about for months (I think around a year actually)!
Click to expand...

Anders Fogh suspected there was a vulnerability for well over a year, but it took many months to get traction with researchers.
 
I would welcome some informed in-house commentary on the 'bug', @Gibbo, @8pack, @OCUK what does your testing show? You must have some significant hardware and software to provide answers in areas other than gaming.
 
Someone is ******
Anyone nows him?

Here is a no ******** approach from MattZN commenting on ARS:

Ok folks, here is the jist of the two major bugs.

(1) Meltdown bug.

This is the 1000 pound gorilla. It's essentially a FULL kernel memory disclosure bug. Most kernel's also implement DMAPs which can extend the disclosure to all of physical memory.

This bug is Intel-specific. Some ARM cpus might also be affected. AMD cpus are immune.

This bug works as follows: All CPUs do speculative memory reads and speculative execution. INTEL CPUs will allow such speculative reads to cross protect domain boundaries, meaning that the speculative read can access kernel memory which is part of the shared MMU map with userspace (but protected with a bit in the page table entry).

AMD CPUs DO NOT ALLOW such speculative reads to cross this protection domain. Thus AMD CPUs are immune to this bug.

This is a major bug. Intel has to fix their (expletive) hardware. A lot of people won't be buying new Intel CPUs that still have this hardware bug.

The 'mitigation' is to break the user process MMU map into *TWO* separate MMU maps, one for the user process, and one for user->kernel and kernel->user transitions. This means that every single kernel->user and user->kernel transition must reload the MMU page tables twice (in x86 land, a mov *,%cr3 or equivalent).

The result is that all system calls and interrupts will now incur an extra 150 to 250nS worth of overhead. A system call normally has an overhead of around 100nS, so the mitigation increases this overhead to 250nS-350nS.

Certain cpu features, such as PCID, can reduce the overhead somewhat, but its still nasty.

Us kernel programmers have spent 20+ years reducing system call and interrupt overhead, and Intel blew it all up in one day. To say that we are all ****** would be a grave understatement.

I think many companies will be holding off new Intel CPU purchases because of this bug, until Intel produces new silicon that doesn't have the bug.

(2) Spectre

Spectre is a sidechannel attack whereby the normal operation of the system where a user program passes data to the kernel in a system call or to another user program which, combined with cache massaging and branch prediction cache massaging, can cause the kernel or other user program to issue speculative reads and do speculative execution within their valid memory that allows the original user program to discern the contents of kernel memory or the memory belonging to the other user program.

This is a much harder attack to perpetrate, and harder to mitigate. All CPUs are probably vulnerable to varying degrees. But its heads and tails harder to exploit this bug than it is to exploit the Meltdown bug.

Specifically, the mitigation for meltdown doesn't help with this bug.

The meltdown bug (which is Intel specific) is horrendous.

(3) There is a third bug called a boundary attack which is easy to mitigate and can be ignored for now.

--

Also, all of Intel's press releases on these topics are HIGHLY deceptive. Purposefully deceptive.

First, they try to revector and confuse the issue by saying these bugs cannot modify or delete memory... but nobody was ever saying that. These bugs DISCLOSE protected memory, meaning your cryptographic keys and web sessions aren't safe (among other things). Intel intentionally avoided mentioning that. Intel also didn't mention that Meltdown is essentially a FULL KERNEL MEMORY disclosure bug, and that it is easy to exploit. And that it is Intel-specific due to stupidity on Intel's part.

Intel is also playing up microcode and BIOS updates for these bugs. What they aren't saying is that these microcode updates amount to ONLY minor mitigations of the Spectre bug. There aren't a complete fix to Spectre or anything close. And, more importantly, THE MICROCODE UPDATES DO NOT FIX THE MELTDOWN BUG AT ALL. We kernel programmers have to implement the horrible performance destroying mitigation to workaround meltdown on Intel CPUs.

Intel is also trying to push all sorts of crap onto the programming community. They are pushing hard to implement horrible hacks in GCC and other compilers and are trying to push horrible hacks to indirect procedure calls as a mitigation for spectre. THIS WILL NEVER WORK!!!!!. 30,000+ applications would have to be recompiled with the changes and kernels would have more horribly hacked code pushed into them just to obtain a PARTIAL mitigation.

Spectre can only be completely fixed in hardware.

Intel is intentionally trying to deceive its customers and its audience. It is the WRONG RESPONSE to these extremely serious bugs, particularly to the Meltdown bug.

To say that we are ****** at Intel right now would be an understatement of epic proportions.

-Matt
 
Intel are doing what they can to mitigate the bug. And trying to dampen the outcry concerning the loss of performance.

The truth is it is going to need new hardware. but you cant say that to customers who have already bought hardware.

Thus the need for a PR spin.
 
So I updated my laptop's CPU microcode with the files Intel released this week (it's an Ivy Bridge Core i5-3320M). The microcode revision changed from 1B to 1C so I guess it did something. Not sure if anything has actually changed though:

oI1JESv.png
 
From the Wikipedia Intel page:
2018 security flaws

Main articles: Meltdown (security vulnerability) and Spectre (security vulnerability)

In early January 2018, it was reported that all Intel processors made since 1995 (besides pre-2013 Intel Itanium and Intel Atom) have been subject to two security flaws dubbed Meltdown and Spectre.

The impact on performance resulting from software patches is "workload-dependent". The Register estimated that fixes would result in a ballpark figure of a 5% to 30% performance reduction with a lower reduction on newer processors, though Intel and others, including Bryan Ma at IDC, believed this to be exaggerated. Using the initial Linux Kernel page-table isolation mitigation showed little impact in some benchmarks, but slowing in particular use-cases. Intel wrote "for the average computer user, [the impact] should not be significant and will be mitigated over time." Microsoft reported that the majority of Azure customers should not see a noticeable performance impact. It is believed that "hundreds of millions" of systems could be affected by these flaws.
Click to expand...
I have no doubt that Intel have a team of people monitoring this page 24/7 to make sure that they can make sure that it is "accurate". I guess that such people as Google, Amazon, etc., despite using vast numbers of Intel CPUs are not "average".
 
DragonQ said:
So I updated my laptop's CPU microcode with the files Intel released this week (it's an Ivy Bridge Core i5-3320M). The microcode revision changed from 1B to 1C so I guess it did something. Not sure if anything has actually changed though:

oI1JESv.png
Click to expand...
Remember to Set-ExecutionPolicy Restricted ;)
 
You must log in or register to reply here.
Back
Top Bottom