Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)

[DisEnchantment]

Speculate at will

 

[Rheingold]

that puts 9700X at ~28% faster than 7700X

That's 28% faster in Blender with the 7700X at 142W PPT and the 9700X using PBO. That's +11% from 7700X@142W to 9700X@88W and +15% from activating PBO on the 9700X. Compare this with the 9950X which has 22% at the same 230W PPT and gains another 6% for a total +29% when activating PBO. This also means that the 9950X will then use more than 300W.

Info: Whoops, this was all mixed up before.

 

[Josh128]

Another strange rumor we've heard is that ST scores tend to increase with PBO turned on. In all previous Ryzen gens to date, simply turning PBO "On" (no curve optimizer, etc.) simply raises PPT and give you better MT perf, but almost always lowers ST perf. Take for example the 9600X above-- were the two different scores with PBO off vs PBO on? ST shouldnt have changed at all, or maybe even slightly decreased if you were to do that in previous gens.

 

[Hail The Brain Slug]

Another strange rumor we've heard is that ST scores tend to increase with PBO turned on. In all previous Ryzen gens to date, simply turning PBO "On" (no curve optimizer, etc.) simply raises PPT and give you better MT perf, but almost always lowers ST perf. Take for example the 9600X above-- were the two different scores with PBO off vs PBO on? ST shouldnt have changed at all, or maybe even slightly decreased if you were to do that in previous gens.

I mean. I saw Cinebench 2024 ST increase on my 7950X just enabling EXPO and nothing else. I don't have a good explanation for that, either.

Until these chips are tested thoroughly either in reviewers hands or in end users hands to demonstrate these kinds of scenarios, we won't know for sure.

 

[PJVol]

Not too shabby for a 6 core i would say

Yeah, especially that pathetic CPU-z 12T score. +7% MT and +17% ST

 

[9949asd]

put the score well into the 2300s.

He posted another! 2244 and 17037!

this should be the pbo, 2130 is stock.

 

[Det0x]

Yeah, especially that pathetic CPU-z 12T score. +7% MT and +17% ST

Who da f cares about CPU-Z "benchmark" ? 🤣

I mean it showed 1% IPC increase for Zen4..
Whatever its measuring, its not "real world performance" ™

 

[PJVol]

this should be the pbo, 2130 is stock.

Since when did PBO improve ST performance?

 

[9949asd]

Since when did PBO improve ST performance?

Excuse me? What are you talking about? Pbo will not improve st and mt?

 

[Josh128]

Excuse me? What are you talking about? Pbo will not improve st and mt?

No, it historically has actually hurt ST perf on previous gen Ryzens.

 

[PJVol]

Excuse me? What are you talking about? Pbo will not improve st and mt?

Did I say MT?

I mean it showed 1% IPC increase for Zen4..

I mean who give a f about IPC increase, TDP or whatever?
We had a +27% MT and +20% ST increase in this benchmark for Zen3 > Zen4. While I agree that it's not completely representative, but it's still concerning.

 

[tsamolotoff]

No, it historically has actually hurt ST perf on previous gen Ryzens.

PBO per se didn't lower the ST performance, high EDC did. Looks like it does not really affect it for Zen4, at least I haven't noticed as big of a drop as with Zen2 or Zen3 (3900x and 5900x).

 

[9949asd]

Did I say MT?

I mean who give a f about IPC increase, TDP or whatever?
We had a +27% MT and +20% ST increase in this benchmark for Zen3 > Zen4. While I agree that it's not completely representative, but it's still concerning.

my 78x3d pbo improved both st and mt. I’m only talking about pbo, not even curve optimize. The 9700x 2280 21xxx is pbo -200

 

[Det0x]

I mean who give a f about IPC increase, TDP or whatever?
We had a +27% MT and +20% ST increase in this benchmark for Zen3 > Zen4. While I agree that it's not completely representative, but it's still concerning.

Did you miss this article about why CPU-Z benchmark shouldn't be taken serious ?

CPU-Z’s Inadequate Benchmark

CPU-Z is a hardware information tool from a company called CPUID, not to be confused with the CPUID instruction.

chipsandcheese.com

Let me quote the conclusion:
---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

Benchmarking is tough. No benchmark can represent the broad range of applications that users will run. For example, Cinebench can’t exactly mirror a gaming workload. However, the primary challenges facing modern workloads are branch prediction and memory accesses, and a lot of benchmarks do present these challenges.

What limits computer performance today is predictability, and the two big ones are instruction/branch predictability, and data locality.
Jim Keller, during an Interview with Dr. Ian Cutress

That’s not just Jim Keller’s opinion. I’ve watched CPU performance counters across my day-to-day workloads. Across code compilation, image editing, video encoding, and gaming, I can’t think of anything that fits within the L1 cache and barely challenges the branch predictor. CPU-Z’s benchmark is an exception. The factors that limit performance in CPU-Z are very different from those in typical real-life workloads.

From AMD’s slides, Zen 4 barely improves over Zen 3 for CPU-Z. AMD’s architects likely saw changes that could benefit CPU-Z wouldn’t pay off in other applications. Zen 4 received improvements like a larger micro-op cache, better branch prediction, and doubled L2 cache capacity. Those would help a lot of applications, but not CPU-Z. Thus, CPU-Z’s benchmark ends up being useless to both CPU designers and end users.

---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
I would actually lump CPU-Z bench together with userbenchmark as the IDL bastions last hope at this point

 

[Josh128]

PBO per se didn't lower the ST performance, high EDC did. Looks like it does not really affect it for Zen4, at least I haven't noticed as big of a drop as with Zen2 or Zen3 (3900x and 5900x).

Right, but either way (lowering or not lowering), it has never equated to +5% or better gains in ST is all Im saying. If thats the case, a lot of things have definitely changed from the status quo. Also, unless that 9600X is really tanking on its max boost in the ST run, its still way lower than AMDs +17% IPC claim. Strictly based on numbers AMD provided to the public, this should be hitting 2330+ in ST with no PBO. 5.7GHz 9950X should be knocking on 2400.

Based on this leak it looks thats all out the window now. If this is what it indeed is, Zen 5 will not beat out the current gen Intel in R23 ST. And before anyone chimes in claiming I shouldnt be comparing to Intels perf prior to the coming mitigations, Intel doesnt matter at all. AMD themselves gave the numbers to expect, even conservatively, that the R23 ST scores would all be in the 2300+ range.

 

[Abwx]

Did you miss this article about why CPU-Z benchmark shouldn't be taken serious ?

CPU-Z’s Inadequate Benchmark

CPU-Z is a hardware information tool from a company called CPUID, not to be confused with the CPUID instruction.

chipsandcheese.com

Let me quote the conclusion:
---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

Benchmarking is tough. No benchmark can represent the broad range of applications that users will run. For example, Cinebench can’t exactly mirror a gaming workload. However, the primary challenges facing modern workloads are branch prediction and memory accesses, and a lot of benchmarks do present these challenges.


That’s not just Jim Keller’s opinion. I’ve watched CPU performance counters across my day-to-day workloads. Across code compilation, image editing, video encoding, and gaming, I can’t think of anything that fits within the L1 cache and barely challenges the branch predictor. CPU-Z’s benchmark is an exception. The factors that limit performance in CPU-Z are very different from those in typical real-life workloads.

From AMD’s slides, Zen 4 barely improves over Zen 3 for CPU-Z. AMD’s architects likely saw changes that could benefit CPU-Z wouldn’t pay off in other applications. Zen 4 received improvements like a larger micro-op cache, better branch prediction, and doubled L2 cache capacity. Those would help a lot of applications, but not CPU-Z. Thus, CPU-Z’s benchmark ends up being useless to both CPU designers and end users.

---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
I would actually lump CPU-Z bench together with userbenchmark as the IDL bastions last hope at this point

You can read here what happened to CPU Z when the 1700X was released, they "updated" the "bench" because AMD scores were too good comparatively to Intel, there s also a comparison between CPU Z 1.78 and 1.79 scores on the link.

CPU Benchmark: CPU-Z 1.78 and CPU-Z 1.79 Scores | Geeks3D

CPU-Z, the popular CPU information utility, comes with a CPU benchmark module. This CPU benchmark is available since CPU-Z 1.73 (released in 2015). The CPU benchmark performs two quick tests: a single threaded test and a multi threaded test. In the latest CPU-Z 1.79, the CPU benchmark has been...

www.geeks3d.com

 

[PJVol]

Let me quote the conclusion:

You didn't have to, I've enjoyed the reading myself.
But still, whatever flaws this benchmark may have, they don't explain such a low increase in MT compared to ST, unless the 9600X is severely power limited in MT test.

 

[PJVol]

The 9700x 2280 21xxx is pbo -200

Reviewers usually just enable the PBO "Lazy Mode", not touching the curve or boost override, and that's mostly what people here mean referring to the PBO.

But you're right, ST score will surely increase with Fmaxboost.

 

[Hitman928]

View attachment 103889

This tweet seems to have been deleted, if that tells us anything.

 

[RTX2080]

View attachment 103889

This just reminds me of something that I heard few weeks ago. There were some problems in whatever pre-production or retail silicon, that cripple performance a bit. Now it seems if thats the case, I guess this is not a coincidence. Maybe the root cause is in the bios, if thats silicon level problem it might not be solved in 1-2 weeks.

 

[Saylick]

C&C posted some musings on Zen 5's decoder. I'll post it in the Zen 5 Architecture thread as well.

Zen 5’s 2-Ahead Branch Predictor Unit: How a 30 Year Old Idea Allows for New Tricks

When I recently interviewed Mike Clark, he told me, “…you’ll see the actual foundational lift play out in the future on Zen 6, even though it was really Zen 5 that set the table for that.” And at that same Zen 5 architecture event, AMD’s Chief Technology Officer Mark Papermaster said, “Zen 5 is...

chipsandcheese.com

 

[APU_Fusion]

At this point I think and should sell itself to nvidia and Lisa Su should commit Seppuku. AMD has failed for two weeks. Death to Zen5

 

[Gideon]

C&C posted some musings on Zen 5's decoder. I'll post it in the Zen 5 Architecture thread as well.

Zen 5’s 2-Ahead Branch Predictor Unit: How a 30 Year Old Idea Allows for New Tricks

When I recently interviewed Mike Clark, he told me, “…you’ll see the actual foundational lift play out in the future on Zen 6, even though it was really Zen 5 that set the table for that.” And at that same Zen 5 architecture event, AMD’s Chief Technology Officer Mark Papermaster said, “Zen 5 is...

chipsandcheese.com

Whoa, I thought I discovered it fast, but you literally posted it the moment C&C own tw(x)itter account tweeted about it.,

 

[Saylick]

Whoa, I thought I discovered it fast, but you literally posted it the moment C&C own tw(x)itter account tweeted about it.,

It helps to be subscribed via email.

 

[Rheingold]

they are really sure of themselves to reduce perf that much, they could had created an intermediary TDP between 65W/88W PPT and 105W/142W PPT for this SKU.

I think it was a conscious decision. AMD saw that People are buying the X3D models not just because of good performance in games, but also because of being so efficient. This is all over the forums, at least in the EU. Ryzen 9000 brings higher performance at lower power consumption. Freeing up PBO headroom is a welcome side effect to please enthusiasts.

Also, not going to the absolute limit serves as a reserve for out-of-box performance gains in future generations should they need it.

 

[yottabit]

I think it was a conscious decision. AMD saw that People are buying the X3D models not just because of good performance in games, but also because of being so efficient. This is all over the forums, at least in the EU. Ryzen 9000 brings higher performance at lower power consumption. Freeing up PBO headroom is a welcome side effect to please enthusiasts.

Also, not going to the absolute limit serves as a reserve for out-of-box performance gains in future generations should they need it.

It just goes to show you can't please everyone. I remember reading nothing but "AMD shouldn't ship these chips so far out of their efficiency range!" and now it seems the majority of the sentiment is "Why would they use such a low TDP if it's only XX% faster than Zen4?! They're leaving so much performance on the table!"