Intel CPUs: Upcoming OS updates might bring up to 35 % loss in performance


(Thies Schulz-Holland) #1

According to “The Register” almost all Intel CPUs sold in the last decade suffer from a massive design flaw that can compromise the security of the machine.

To fix this, all major operating systems have to be updated accordingly. The problem: These security updates might mean a decrease in CPU performance of up to 35 percent, depending on the respective CPU model. This may very well also hit common virtualisation environments including Amazon EC2 and Google Compute Engine, render farms etc.

AMD CPUs are not affected, yet it is unclear as of now if those security updates will be forced upon AMD systems running Windows as well. AMD has supplied a kernel patch for Linux that stops those updates from being applied.


(Ace Dragon) #2

PC World did an exhaustive FAQ on this upcoming patch and the amount of slowdown you get really depends on how many calls to the OS Kernel the program does.

For instance, there’s a chance that rendering in Blender won’t see much impact because a similar task (video encoding), did not show a drop in Linux. However, it might be (for Windows users) that the performance of Microsoft apps. (Edge, Office, ect…) and games using DirectX will see a significant drop.

In addition, I speculate that it could drive another nail into Microsoft’s effort to promote UWP via the Windows store (there’s gotta be a lot of Kernel calls in there).

This should also be good news for AMD (providing that Microsoft doesn’t deliberately hobble their CPU’s as well, especially if it’s at the request of Intel). A 30 percent slowdown will end up making the Ryzens stand out compared to the i-series chips (even against Coffee Lake if that line still has the flaw). The choice of mine to seriously look at an AMD system when Ryzen+ comes out is looking better now.


(Ace Dragon) #3

Meanwhile, in the comment sections of other tech. websites, conspiracy theories are already starting to fly.

One conspiracy involves Intel forcing the creation of special RAM for the Kernel and creating a system of planned obsolescence, and another alleges a move by Intel to discourage certain programmer tactics in a bid to cripple AMD and get their monopoly back. That is just wild speculation though (like most theories, that being until someone can bring up solid evidence).

If there is evidence though, then who knows just how large the inevitable lawsuit will be and how much user anger will be generated (maybe enough to make AMD into the king of PC chips).


(organic) #4

. . . . . . . . .


(Ace Dragon) #5

So Intel has released some information regarding the flaw, they argue that a good chunk of it is pure alarmism from The Register.

They also argue that it affects nearly every single chip in a computing device today, not just Intel PC’s (though AMD argues it’s still a non-issue for their chips).


(sdfgeoff) #6

A good summary with links to the papers that describe the problem: https://meltdownattack.com
The worst case: a javascript program running in a browser can potentially read the contents of you computers password manager that are in RAM (aka your unencrypted passwords). Extremely unlikely, but I suppose it’s technically possible.


(thomascheng) #7

Obvious damage control. Is it true? I don’t know, but I know Intel has a history of fishy marketing tactics.


(sdfgeoff) #8

Quote from the Sceptre Paper (aka the guys who found the bug):

These attacks represent a serious threat to actual systems, since vulnerable speculative execution capabilities are found in microprocessors from Intel, AMD, and ARM that are used in billions of devices.


(burnin) #9

:yes: there will always be a flaw…


(Thies Schulz-Holland) #10

And here is a “translated” version of Intel’s PR blah: https://www.theregister.co.uk/2018/01/04/intels_spin_the_registers_annotations/


(Thies Schulz-Holland) #11

AFAIK Sceptre is not the issue addressed by the upcoming fixes to Linux and Windows. There are two security flaws at play here:

a)
MELTDOWN, which seems to only affect Intel and ARM CPUs and will be fixed shortly, thereby probably causing a system slowdown on these CPUs of presently unknown proportions. This issue is rather easy to exploit, so it needs to be fixed quickly.

b)
SCEPTRE, which is said (by Intel) to affect CPUs of all vendors. From what I have read it is quite hard to exploit, but even harder to fix.

Bringing up SCEPTRE in this context is of course a PR smoke screen by Intel to drag AMD into this: “See? Seee? They suck, too.”


(Jason van Gumster) #12

Moved from “Misc: > Off-topic Chat” to “General Forums > Latest News” (wow… how often does that happen?)


(theoldghost) #13

Have Linux openSUSE and a Intel CPU and have never looked back. As my ASUS must be since I don’t hear the HD groaning when I’m doing nothing. And, up through Windows 7 I was a Microsoft fanboy. But, not anymore. Windows 10 made me a Linux believer.

///


(Herbert123) #14

You do realize all operating systems are affected by this? It’s an issue on a hardware level in modern CPUs - Intel for certain, AMD not quite so certain yet. So your Intel CPU IS the problem. Has nothing to do with whether Linux is “better” than other OSs in your perception.


(m9105826) #15

This has nothing to do with Windows. This exploit exists at the instruction level of basically every CPU produced this decade.


(dgorsman) #16

All I know it’s going to be a PITA to filter out actual information from the combination of anti-Intel fanboys and review-of-a-review-of-a-summary reports. For example looking at the post on the Register site, it doesn’t say “… up to 35%…”, it says “…we’re looking at a ballpark figure of five to 30 per cent…” (so where did that extra 5% slip in?).

The actual slowdowns may not be noticed by a whole lot of people seeing as the majority are surfers, lightweight gamers, and “productivity users”.


(Ace Dragon) #17

Some actual benchmarks have been published.

In all, worst case is a 2 percent performance loss in certain gaming and memory-related tasks. 3D rendering will also slow down, but the drop is less than 1 percent (which in Cycles will not even undo many of the smaller optimizations committed over time).


(Thies Schulz-Holland) #18

My bad. The first reports I found about this were these:



(dgorsman) #19

Thanks - not pointing fingers, just pointing out the complexities of trying to get good info. :slight_smile:


(Thies Schulz-Holland) #20

Seems the Windows bugfix is distributed right now as KB4056892.
Quickly ran a 3DMark benchmark on all my PCs - one older AMD desktop (FX8350) and an Intel notebook from last year (Intel Core i7-6700HQ) to be able to compare performance before and after the patch. Will report the results back when available.

EDIT
3DMark “Time Spy” benchmark (Top = pre-patch / bottom = post-patch) on the Intel machine. Seems to indicate a drop in CPU performance of roundabout 3 percent.


On the AMD the CPU performance dropped by 0.7 percent, but I guess this is within the error margin of the benchmark: