Question IPC revision for Raptor Lake / Zen4 in light of Spectre V2

[igor_kavinski]

I suppose there will be a minimum of 5% impact on Raptor Lake and Zen 4 IPC due to Spectre V2 (BHI) mitigation patches in most workloads. Alder Lake will be screwed shortly by a Windows update mitigating BHI. What else? Can Zen 5 and Meteor Lake/Arrow Lake be modified at this point to avoid this performance regression?

 

[nicalandia]

I suppose there will be a minimum of 5% impact on Raptor Lake and Zen 4 IPC due to Spectre V2 (BHI) mitigation patches in most workloads. Alder Lake will be screwed shortly by a Windows update mitigating BHI. What else? Can Zen 5 and Meteor Lake/Arrow Lake be modified at this point to avoid this performance regression?

These designs(Zen5 and Meteor Lake) have been set in stone for many years now, they can't avoid this Hit. The biggest hit will be at the Server/HPC level, people with Laptops will not be as impacted.

 

[igor_kavinski]

It might not feel like a hit to the normal user but 4K I/O for NVMe SSDs might get slower. Also, once the patches are applied to both Win10/11 and presumably before Raptor Lake/Zen 4 launch, it will be slower from the get go. I don't think anyone will load up a pre-patched version of the Windows OS and go, "This is what it was actually supposed to be like."

 

[Abwx]

From NBC who borrowed Phoronix numbers :

Spectre-v2 mitigation wreaks havoc on the performance of some Intel CPUs as AMD chips come out largely unscathed

Testing performed by Phoronix on some AMD and Intel CPUs has revealed that Intel CPUs suffer as much as a 35% performance drop due to the Spectre-v2 mitigation on Linux, whereas almost all AMD processors post a drop of less than 10%.

www.notebookcheck.net

 

[nicalandia]

We need Windows numbers Pronto, I mean Unix and Linux are cool and all, but they have not gained any market share since the late 1990s(at the Desktop, I know that Unix/Linux rule the world at HPC)

 

[The Hardcard]

These designs(Zen5 and Meteor Lake) have been set in stone for many years now, they can't avoid this Hit. The biggest hit will be at the Server/HPC level, people with Laptops will not be as impacted.

Actually, I believe Ian’s interview with the Zen head architect (Mike Clark?) that changes could be made in the last months. I think Meteor Lake and Zen 5 will be modded at least at the microcode level.

EDIT:

IC: Is it ever realistic to be able to pivot a design based on what competitors just released? Or is do you still have that two year lead on changes?

MC: It matters - we can. You'd be surprised at how quickly we can respond.

I’d expect changes in Zen 4.

 

[HurleyBird]

These designs(Zen5 and Meteor Lake) have been set in stone for many years now, they can't avoid this Hit. The biggest hit will be at the Server/HPC level, people with Laptops will not be as impacted.

AMD and Intel are not just reacting to issues as they appear, but also taking proactive steps to generally harden successive generations in anticipation of future exploits.

 

[igor_kavinski]

Things are not looking good for future PC ports of PS5/Xbox Series X games on Alder Lake that will depend on the DirectStorage API. 5800X3D is looking pretty attractive in this context.

 

[IntelUser2000]

It's going to affect them but it won't be as big performance-wise as some people think it'll be. Competitiveness is a bigger deal to performance than security mitigations. If it's a good CPU, it'll be class leading regardless. The Skylake generations got pretty much bombarded with everything.

Also, I question the proactive way of trying to find bugs and then putting out an article to call it out. Of course it'll have bugs and oversights. These are amazingly complex. Eventually it might lead to worse security as the whole world is aware of it - good or bad actors. And the install base of CPUs are massive.

2FAs, increasingly complex password requirements, recaptchas. They are a plague.

Why not just tell the CPU manufacturers in secret to fix it instead?

 

[The Hardcard]

It's going to affect them but it won't be as big performance-wise as some people think it'll be. Competitiveness is a bigger deal to performance than security mitigations. If it's a good CPU, it'll be class leading regardless. The Skylake generations got pretty much bombarded with everything.

Also, I question the proactive way of trying to find bugs and then putting out an article to call it out. Of course it'll have bugs and oversights. These are amazingly complex. Eventually it might lead to worse security as the whole world is aware of it - good or bad actors. And the install base of CPUs are massive.

2FAs, increasingly complex password requirements, recaptchas. They are a plague.

Why not just tell the CPU manufacturers in secret to fix it instead?

I haven’t seen the specifics of this case yet, but the normal is to wait 60 or 90 days after informing the responsible manufacturer/developer to go public.

For the gixes yo be out now I’d say Intel and AMD were properly told in advance.

 

[IntelUser2000]

I haven’t seen the specifics of this case yet, but the normal is to wait 60 or 90 days after informing the responsible manufacturer/developer to go public.

For the gixes yo be out now I’d say Intel and AMD were properly told in advance.

I think just keep it secret entirely. Because shouting out to the world you are raising the bar for how many are aware of the bugs exist.

When you want to sell an item, one of the first thing to focus is on marketing to increase visibility. Seems to me they are doing free advertising for people who want to get into the malware business. There are so many security news nowadays it's practically noise now. Massive banks, countries, military, they all get hacked and stolen at one point or another.

Sure making it public will help the high stakes scenario but I think it makes it worse for everyone else. It also makes using the internet increasingly annoying. One article was talking about how simple 6 letter password was all that was needed.

You cannot make it 100% secure. Impossible.

It also tells you how bad the world is when it comes to mistrust and you need to wall yourself from literally everyone.

One can imagine a story of a dystopian future where an average internet user needs a physical smart ID card, 12-letter password consisting of upper and lower case, special characters, a space in between after the 6th letter and a biometric scan to post on social media.

 

[igor_kavinski]

One can imagine a story of a dystopian future where an average internet user needs a physical smart ID card, 12-letter password consisting of upper and lower case, special characters, a space in between after the 6th letter and a biometric scan to post on social media.

All that and they may still get through using the NSA backdoors

 

[Carillon]

It also tells you how bad the world is when it comes to mistrust and you need to wall yourself from literally everyone.

That's not a bad thing, you shouldn't trust anybody, not even yourself. I myself once convinced me that I could jump a large ditch, I wish I didn't.
Also if all these vulnerabilities were to be kept secret, we'd either be suing manufacturers for the sudden performance drops, if they mitigated the vulnerabilities, or be completely at their mercy if they didn't.

 

[DrMrLordX]

I’d expect changes in Zen 4.

Correct me if I'm wrong, but "Spectre V2" doesn't even affect AMD CPUs, does it? The only problem AMD is facing (if you can call it that) is a flaw in their retpoline mitigation against Spectre V1.

 

[deasd]

This discussion should be in the sticky Spectre/Meltdown thread though... Since Zen4/RPL already finished in silicon, these two would have some impact inevitably, but would slightly better than nowaday CPUs which are on shelve.

I'm wondering why Microsoft/Intel still haven't issued forecast of mitigation/fixes for Windows. It seems this scene is like when 1st gen Spectre born, this V2 Spectre would end up as nightmare as 1st gen Spectre which lead to measureable perf drop. I guess Microsoft/Intel now both are in headache....

These designs(Zen5 and Meteor Lake) have been set in stone for many years now, they can't avoid this Hit. The biggest hit will be at the Server/HPC level, people with Laptops will not be as impacted.

I think if an architecture still hasn't finished in silicon they might have chance to fix. But the worst scenario is this would become endless nightmare when new security hole being discovered it would keep affecting everything in our hand....

 

[soresu]

This discussion should be in the sticky Spectre/Meltdown thread though... Since Zen4/RPL already finished in silicon, these two would have some impact inevitably, but would slightly better than nowaday CPUs which are on shelve.

I'm wondering why Microsoft/Intel still haven't issued forecast of mitigation/fixes for Windows. It seems this scene is like when 1st gen Spectre born, this V2 Spectre would end up as nightmare as 1st gen Spectre which lead to measureable perf drop. I guess Microsoft/Intel now both are in headache....



I think if an architecture still hasn't finished in silicon they might have chance to fix. But the worst scenario is this would become endless nightmare when new security hole being discovered it would keep affecting everything in our hand....

Zen4 isn't even out yet, there is still time to revise Zen5 µArch for security tweaks if Intel could do it ad infinitum for Skylake.

 

[lobz]

Actually, I believe Ian’s interview with the Zen head architect (Mike Clark?) that changes could be made in the last months. I think Meteor Lake and Zen 5 will be modded at least at the microcode level.

EDIT:



I’d expect changes in Zen 4.

Surely you meant Zen 5

 

[darkswordsman17]

Correct me if I'm wrong, but "Spectre V2" doesn't even affect AMD CPUs, does it? The only problem AMD is facing (if you can call it that) is a flaw in their retpoline mitigation against Spectre V1.

The link indicates its about Spectre v2 mitigations but seems to be about their previous mitigations, so yeah not sure what exactly the issue is. Which, I'll have to go somewhere else to find out since Anandtech doesn't even bother to pretend to give lip service to this stuff any more.

This discussion should be in the sticky Spectre/Meltdown thread though... Since Zen4/RPL already finished in silicon, these two would have some impact inevitably, but would slightly better than nowaday CPUs which are on shelve.

I'm wondering why Microsoft/Intel still haven't issued forecast of mitigation/fixes for Windows. It seems this scene is like when 1st gen Spectre born, this V2 Spectre would end up as nightmare as 1st gen Spectre which lead to measureable perf drop. I guess Microsoft/Intel now both are in headache....



I think if an architecture still hasn't finished in silicon they might have chance to fix. But the worst scenario is this would become endless nightmare when new security hole being discovered it would keep affecting everything in our hand....

We know why Intel isn't. Because they don't need to, seemingly no one is holding them accountable for this stuff so there's no need for them to respond to any of it. Well other than sick their leet haxors to try and break AMD's stuff so they can go "look how bad AMD is!" How long before we get part 2 of the "OMG the AMD is broken permanently no fix ever!" clownshow?

As for Microsoft, I'm guessing its a mix of Windows not being that big on servers and them buying up every AMD EPYC chip they can for their own servers? I don't think Microsoft cares about performance for consumers in Windows that much as there's the AMD option and it gets people to buy new PCs. People know the issue is Intel so its not like Microsoft should catch much flak for it. I guess maybe the people buying Surface products (but pretty sure that's largely Enterprise at this point as the former Apple people likely have moved back now). But this just plays into Microsoft's argument for their overall software solution in Enterprise and doing stuff like the security chip in AMD's stuff.

Considering how Win11 came out the gate though, sadly I'm expecting that Microsoft will let Intel bake in their own fix and it'll probably include some flag that happens to tank performance on AMD stuff as well just because. And then Intel will let places benchmark after the fix and AMD will call BS before Intel goes "oops, our bad, totally didn't do that on purpose " but most people will be too obtuse and we'll get people claiming AMD is also affected despite the glaring differences.

I'd almost say that it seems like maybe Alder Lake's benefit is the thread advisor and that these fixes seem like they might undo that (which would lead to questioning if that was known and intentional by Intel risking that people wouldn't care since seems few people cared about Spectre before), but then apparently older Intel is hit even worse so perhaps its what's keeping Alder Lake from really getting hammered in performance.

 

[TheELF]

From NBC who borrowed Phoronix numbers :

Spectre-v2 mitigation wreaks havoc on the performance of some Intel CPUs as AMD chips come out largely unscathed

Testing performed by Phoronix on some AMD and Intel CPUs has revealed that Intel CPUs suffer as much as a 35% performance drop due to the Spectre-v2 mitigation on Linux, whereas almost all AMD processors post a drop of less than 10%.

www.notebookcheck.net

https://www.tomshardware.com/news/intel-cpus-performance-hit-spectre-v2-migitations

https://www.tomshardware.com/news/amd-cpus-see-less-than-10-performance-drop-from-revised-spectre-v2-mitigations

From the sources they list.
Also the 10% for AMD is on epic and the 35% for intel is on a laptop cpu...depending on your workload you will be better off on one or the other but neither one is a clean victor.
Clickbaiting at its finest.

According to Phoronix's Core i9-12900K (Alder Lake) results, networking and storage performance went down the toilet after enabling Retpolines. The publication recorded a 26.7% performance loss on the former and 14.5% on the latter. That's the hallmark of these mitigations: Any external I/O from the chip takes a hard hit. Workloads like web browsing or image manipulation in GIMP didn't show a huge impact.

The Ryzen 9 5950X (Vermeer) suffered a 54% performance reduction with the Stress-NG (Context Switching) benchmark. Stress-NG is similar to Prime95 for Windows users, so it isn't the most relevant metric for measuring performance from a consumer workload standpoint. However, the Ryzen 9 5950X held up pretty well besides that specific benchmark. There was only a 5.3% and 5% drop in networking and storage performance, respectively. In comparison, the Core i9-12900K (Alder Lake) experienced performance hits of 26.7% and 14.5% in the networking and storage department.

Spoiler

 

[igor_kavinski]

5950X suffering in context switching is pretty bad for DB workloads running inside a VM. This could potentially be catastrophic for AMD in the cloud space.

 

[Hitman928]

https://www.tomshardware.com/news/intel-cpus-performance-hit-spectre-v2-migitations

https://www.tomshardware.com/news/amd-cpus-see-less-than-10-performance-drop-from-revised-spectre-v2-mitigations

From the sources they list.
Also the 10% for AMD is on epic and the 35% for intel is on a laptop cpu...depending on your workload you will be better off on one or the other but neither one is a clean victor.
Clickbaiting at its finest.

Spoiler

The notebookcheck article should be ignored. They seemingly don't realize that the Phoronix tests are testing mitigations for two different vulnerabilities for AMD and Intel and it is also not the same tests being run on both. For instance, the worst case performance penalty for AMD showing 54% faster with prior mitigations, this test wasn't even run on the Intel CPUs. Same applies the other way around. For the tests that were the same between the two, Intel CPUs showed worse performance degradation, but again, they are testing mitigations for different vulnerabilities. The ones for Intel are new and AMD CPUs aren't affected by this vulnerability whereas the AMD ones are updated mitigations for an older vulnerability which also showed (some times severe) performance degradation on the tested Intel CPUs at the time.

5950X suffering in context switching is pretty bad for DB workloads running inside a VM. This could potentially be catastrophic for AMD in the cloud space.

Intel suffered really bad too from this as well (ARM most likely too), it just wasn't tested in these Phoronix tests so it is unknown if Alderlake has hardware mitigations for it that avoid the performance penalties. It won't be catastrophic for AMD in the cloud space because their competitors suffer from the same thing.

 

[DrMrLordX]

5950X suffering in context switching is pretty bad for DB workloads running inside a VM. This could potentially be catastrophic for AMD in the cloud space.

Suffering by how much? The only org I'm aware of that has even done proper testing of Generic Retpolines vs AMD Retpolines is Phoronix, and in any actual test they did (versus a synthetic), switching to Generic barely did anything to Zen3?

 

[Ajay]

5950X suffering in context switching is pretty bad for DB workloads running inside a VM. This could potentially be catastrophic for AMD in the cloud space.

Oh nooos, the sky is falling!

 

[yuri69]

Actually, security patches lowering performance have been a blessing for Intel datacenters business. With the patches applied, there is suddenly a need for purchasing additional Intel machines.

 

[igor_kavinski]

It's at times like this I wonder how I got to be a mediocre Excel monkey instead of having my dream job of testing the latest and greatest hardware for the performance impact of just-landed-security-vulnerability-mitigations. Oh wait, I know. Coz most offices don't give a damn about how slow their hardware is gonna get by some mitigation patch but they do care about having an Excel monkey around to do their dirty work. *sigh*