Intel bug incoming? Meltdown and Spectre exploits

kitfit1 · 2 Jan 2018 at 20:34

Code:

KAISER will affect performance for anything that does system calls or
interrupts: everything.  Just the new instructions (CR3 manipulation)
add a few hundred cycles to a syscall or interrupt.  Most workloads
that we have run show single-digit regressions.  5% is a good round
number for what is typical.  The worst we have seen is a roughly 30%
regression on a loopback networking test that did a ton of syscalls
and context switches.  More details about possible performance
impacts are in the new Documentation/ file.

Yes, the register has it at 5% to 30%. Not:

gavinh87 said:
From what I've been reading its more like 5% but its all speculation until the embargo lifts.

https://www.theregister.co.uk/2018/01/02/intel_cpu_design_flaw/

pete910 · 2 Jan 2018 at 20:40

Lol,

redesign of the Linux and Windows kernels

Am not saying that it's obviously been a PITA as appears to be fairly bad hole that's needed to be plugged, but stating a redesign is somewhat sensationalized I feel.

gavinh87 · 2 Jan 2018 at 20:41

kitfit1 said:
Yes, the register has it at 5% to 30%. Not:

https://www.theregister.co.uk/2018/01/02/intel_cpu_design_flaw/

Hmmm to believe patch notes or believe a news site??
I guess we'll see soon.

pete910 · 2 Jan 2018 at 20:44

gavinh87 said:
Hmmm to believe patch notes or believe a news site??
I guess we'll see soon.

Devs that's done patches wont know the actual perf hit until these things are tested either

Rroff · 2 Jan 2018 at 20:44

I'm always surprised more security issues with Intel's management engine don't happen TBH.

gavinh87 · 2 Jan 2018 at 20:55

pete910 said:
Devs that's done patches wont know the actual perf hit until these things are tested either

From his testing he actually states most workloads show 5%. Of course any performance hit isn't good but we know how these things go. Hyperbole most of the time.

jigger · 2 Jan 2018 at 21:00

Glad I'm moving everything over to AMD. Looks like I'll be moving over a little quicker now.

sideways14a · 2 Jan 2018 at 21:03

5% is still pretty bad, ok its not a catastrophic hit like 30% but its bad - not only that but the hit to Intels rep is not good either.

Great start to 2018 Intel.

pete910 · 2 Jan 2018 at 21:43

gavinh87 said:
From his testing he actually states most workloads show 5%. Of course any performance hit isn't good but we know how these things go. Hyperbole most of the time.

Which patch/dev are you referring too ?

DragonQ · 2 Jan 2018 at 22:10

So will Ryzen have better IPC than Coffee Lake after these patches?

Chuk_Chuk · 2 Jan 2018 at 22:17

Rroff said:
I'm always surprised more security issues with Intel's management engine don't happen TBH.

Does the intel management engine handle code prefetching?

~>Dg<~ · 3 Jan 2018 at 00:04

DragonQ said:
So will Ryzen have better IPC than Coffee Lake after these patches?

no with this issue they may come close to level LOL

pete910 · 3 Jan 2018 at 01:07

Rather costly then by the looks of it it seems

https://www.phoronix.com/scan.php?page=article&item=linux-415-x86pti&num=2

File IO and SQL seem to be the worst hit.

Hotwired · 3 Jan 2018 at 02:06

Can be worse than 30% then.

AndreiD · 3 Jan 2018 at 03:49

Some gaming benchmarks: https://www.phoronix.com/scan.php?page=news_item&px=x86-PTI-Initial-Gaming-Tests
Gaming is unaffected it seems.

Also want to point out that AMD hardware is targeted too, at least until it's properly confirmed to not be affected by whatever issue this is.

ajf · 3 Jan 2018 at 07:16

Just reading about this before I came on here.
Coding not being my strong point, but some sites are saying this affects everything on Windows, whilst others, and here, mention specifically VMs?
Can anyone explain in simple times the contradiction?

d_brennen · 3 Jan 2018 at 07:39

https://www.theregister.co.uk/2018/01/02/intel_cpu_design_flaw/

Don't believe this has been posted yet. A security floor that potentially allow user land code to snoop kernel level CPU memory has been in numerous Intel CPUs for the last ~ 10 years. Linux, Microsoft and Apple teams looking to patch via OS update. Worst case scenario is certain instructions will take significantly longer to execute. The penalty mulled over in the article is somewhere between 5 and 30% hit, varying with what the workload is.

Suddenly those Ryzen and Threadripper chips look so much better value. Hopefully we get a new round of benches post-fix.

This will be typical Intel, they'll be sued in the US, pay an undisclosed amount and settle with the plaintiffs while the rest of the industry takes the flack.

Cromulent · 3 Jan 2018 at 07:54

Hopefully Intel can fix the issue with a microcode patch and the operating system level patch which hurts performance will just be a temporary fix while we wait for a microcode patch.

I just bought a new Intel system yesterday. I'm a bit annoyed now but time will tell I guess.

Edit: Seems the article says a microcode update can't fix the issue. Damn.

Rroff · 3 Jan 2018 at 08:03

Be interesting to see if the impact is different between dual and quad channel architectures and how much can be mitigated via memory timings, etc.

Cromulent · 3 Jan 2018 at 08:53

Thankfully the two tasks I do most on my PC (gaming and editing / rendering video) do not seem to be impacted by performance problems according to the Phorenix benchmarks. I guess I'll just have to see how things turn out when the patch is deployed for Windows 10 on Tuesday next week. I'll run some benchmarks before and after and see what happens.