Reviewers spot major regression in AMD's high-end Ryzen 9000 series Zen 5 CPUs

[WYP]

Multiple reviewers spot huge inter-CCX latency increase for AMD's Ryzen 9000 Zen 5 CPUs - Bad news for gamers



Read more about AMD's Zen 5 inter-CCX latency regression.

 

[AngryGoldfish]

There's quite a few funny comments going 'round about AMD at the moment.

"AMD never misses an opportunity to miss an opportunity."

"Isn't AMD kind? They're giving Intel a chance to come back."

"AMD fails to compete even with itself."

It seems clear that this architecture is not ready for launch.

 

[Dawelio]

There's quite a few funny comments going 'round about AMD at the moment.

"AMD never misses an opportunity to miss an opportunity."

"Isn't AMD kind? They're giving Intel a chance to come back."

"AMD fails to compete even with itself."

It seems clear that this architecture is not ready for launch.

Yup and that apparently Zen 5 was really made for the server environment and not the consumer market.

 

[NeverBackDown]

People are so quick to blame AMD. Blame MS instead, this is a Windows ONLY issue. These latencies are not seen on Linux. Why? Because windows sucks and they make AMD use core parking because of windows game bar. This is the best theory on the matter (game bar that is)

This is a scheduling problem. Look at the Linux reviews. These CPUs absolutely crush everything.

 

[WYP]

People are so quick to blame AMD. Blame MS instead, this is a Windows ONLY issue. These latencies are not seen on Linux. Why? Because windows sucks and they make AMD use core parking because of windows game bar. This is the best theory on the matter (game bar that is)

This is a scheduling problem. Look at the Linux reviews. These CPUs absolutely crush everything.

Anandtech's Ryan Smith says that they have tested on Linux and have found simularly high latencies.

https://x.com/RyanSmithAT/status/1824103659019628717

I think you are referring to the Ryzen Windows bug that is being reported on? That Windows issue seems to be a different issue. That said, if I am unaware of any details I would like to be better informed.

 

[NeverBackDown]

No that bug you reported on gives what a 3% average performance lift? I wouldn't really say that's a big deal personally, although extra performance is welcome.

I've seen people talk about windows game bar core parking hurting zen. It forces a different schedule for essentially game mode. I don't have a CPU I can't test it. From what I can see this must be a bigger windows issue than the 3% performance bug. On Linux these new CPUs are incredibly good. Absolutely smashing everything. I could be wrong on the reasoning but it wouldn't be the first time Windows messed something up

If this new tweet reports (from what I can tell slightly better but still bad) CCX latencies, then why is the performance uplift on Linux so much better all else being equal? It doesn't make sense.

For Linux testing reference: https://www.phoronix.com/review/amd-ryzen-9950x-9900x/15

 

[AngryGoldfish]

People are so quick to blame AMD. Blame MS instead, this is a Windows ONLY issue. These latencies are not seen on Linux. Why? Because windows sucks and they make AMD use core parking because of windows game bar. This is the best theory on the matter (game bar that is)

This is a scheduling problem. Look at the Linux reviews. These CPUs absolutely crush everything.

I'm not actually blaming AMD though. Why or how this happened or whose responsible is less what I was saying. What I'm really saying is that it's not ready for launch.

 

[Dicehunter]

No that bug you reported on gives what a 3% average performance lift? I wouldn't really say that's a big deal personally, although extra performance is welcome.

I've seen people talk about windows game bar core parking hurting zen. It forces a different schedule for essentially game mode. I don't have a CPU I can't test it. From what I can see this must be a bigger windows issue than the 3% performance bug. On Linux these new CPUs are incredibly good. Absolutely smashing everything. I could be wrong on the reasoning but it wouldn't be the first time Windows messed something up

If this new tweet reports (from what I can tell slightly better but still bad) CCX latencies, then why is the performance uplift on Linux so much better all else being equal? It doesn't make sense.

For Linux testing reference: https://www.phoronix.com/review/amd-ryzen-9950x-9900x/15

I've been messing around lately with Linux slowly learning it, Garuda is pretty nice, And even on 7000 series chips like the 7800X3D which is single CCD I'm seeing an uplift in certain games of 5-10% compared to Windows which seems very odd, This is compared to a completely fresh Windows 11 install fully updated, Latest drivers etc...

I'm guessing something under the hood of Windows needs looking at and likely a lot of initial 7000 series tests need to be redone.

 

[NeverBackDown]

I'm not actually blaming AMD though. Why or how this happened or whose responsible is less what I was saying. What I'm really saying is that it's not ready for launch.

This doesn't make sense though? If it's indeed a Windows issue it would not matter when it launches. It would still be an issue. I'm seeing people talk about launching for example games in administration mode/system admin privileges getting big results. This isn't a good thing. You don't want everything to have that level of access. This work around works for all CPUs however. It lets the programs schedule more work rather than letting the OS decide. I think MS just need to update their OS to better reflect the difference in CCX vs Intel's big.LITTLE architecture designs.

I've been messing around lately with Linux slowly learning it, Garuda is pretty nice, And even on 7000 series chips like the 7800X3D which is single CCD I'm seeing an uplift in certain games of 5-10% compared to Windows which seems very odd, This is compared to a completely fresh Windows 11 install fully updated, Latest drivers etc...

I'm guessing something under the hood of Windows needs looking at and likely a lot of initial 7000 series tests need to be redone.

Yep. No issues on Linux. The kernel just works. You will get big performance benefits on Zen 5 vs Zen 5 windows and that's even using an API compatibility layer in between for gaming.

 

[AngryGoldfish]

This doesn't make sense though? If it's indeed a Windows issue it would not matter when it launches. It would still be an issue. I'm seeing people talk about launching for example games in administration mode/system admin privileges getting big results. This isn't a good thing. You don't want everything to have that level of access. This work around works for all CPUs however. It lets the programs schedule more work rather than letting the OS decide. I think MS just need to update their OS to better reflect the difference in CCX vs Intel's big.LITTLE architecture designs.

But doesn't AMD design their products alongside Microsoft? They design their chips for Sony and Xbox with their platforms in mind, so why wouldn't AMD design their chips with Windows in mind? If they knew these issues existed and needed assistance from Microsoft to get the code in the OS to be correct, why not delay the launch until that is done? Does Microsoft need to use buyers to debug the problems? I genuinely am asking because I don't actually know.

Either way, if it's true, it's not a product ready for launch in my opinion. DLSS was not ready for launch when it was because it hadn't gathered enough data. Consumers were the generators of information. That in my opinion is an unfinished product. It was AMD's decision to use Zen 2 information for Zen 5 when Zen 4 data may have been better suited.

Again, not saying this is AMD's fault entirely. It looks to be an issue with Microsoft. But that doesn't change that AMD surely had a choice to delay the launch until Microsoft had their OS in order to receive the new chips.

 

[NeverBackDown]

But doesn't AMD design their products alongside Microsoft? They design their chips for Sony and Xbox with their platforms in mind, so why wouldn't AMD design their chips with Windows in mind? If they knew these issues existed and needed assistance from Microsoft to get the code in the OS to be correct, why not delay the launch until that is done? Does Microsoft need to use buyers to debug the problems? I genuinely am asking because I don't actually know.

Either way, if it's true, it's not a product ready for launch in my opinion. DLSS was not ready for launch when it was because it hadn't gathered enough data. Consumers were the generators of information. That in my opinion is an unfinished product. It was AMD's decision to use Zen 2 information for Zen 5 when Zen 4 data may have been better suited.

Again, not saying this is AMD's fault entirely. It looks to be an issue with Microsoft. But that doesn't change that AMD surely had a choice to delay the launch until Microsoft had their OS in order to receive the new chips.

MS bugs have been around for years. If you want hardware companies to wait for them to fix these issues, you'll be waiting forever.

In my opinion I don't consider your console analogy an equal argument. Sony and MS both come to AMD with their design specifications and make AMD figure out how to make it work. That's a little different because it's purely custom. In our case, its just a generic slap together parts and OS.

I'm not trying to abstain AMD from fault here, I do think they are taking to much heat over less than exciting performance in certain situations. Still you are correct in saying they collaborate. AMD could have pushed harder. For all we know they did and MS ignored them. Either way at some point hopefully it's all fixed.

 

[AngryGoldfish]

MS bugs have been around for years. If you want hardware companies to wait for them to fix these issues, you'll be waiting forever.

In my opinion I don't consider your console analogy an equal argument. Sony and MS both come to AMD with their design specifications and make AMD figure out how to make it work. That's a little different because it's purely custom. In our case, its just a generic slap together parts and OS.

I'm not trying to abstain AMD from fault here, I do think they are taking to much heat over less than exciting performance in certain situations. Still you are correct in saying they collaborate. AMD could have pushed harder. For all we know they did and MS ignored them. Either way at some point hopefully it's all fixed.

Fair points. I agree with you.