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

[DisEnchantment]

Speculate at will

 

[Josh128]

Someone cracked 3500 ST in GB6 Windows with a 9700X. 5.7GHz ST perf. Not bad.

ASRock B650M-HDV/M.2 - Geekbench

Benchmark results for an ASRock B650M-HDV/M.2 with an AMD Ryzen 7 9700X processor.

browser.geekbench.com

 

[Thunder 57]

Someone cracked 3500 ST in GB6 Windows with a 9700X. 5.7GHz ST perf. Not bad.

ASRock B650M-HDV/M.2 - Geekbench

Benchmark results for an ASRock B650M-HDV/M.2 with an AMD Ryzen 7 9700X processor.

browser.geekbench.com

But it doesn't do awesome in games so its a complete dumpster fire release like all the techtubers told me. /s

 

[poke01]

Someone cracked 3500 ST in GB6 Windows with a 9700X. 5.7GHz ST perf. Not bad.

ASRock B650M-HDV/M.2 - Geekbench

Benchmark results for an ASRock B650M-HDV/M.2 with an AMD Ryzen 7 9700X processor.

browser.geekbench.com

guesses on 5.7GHz power consumption?

I think 30-35 watts.

 

[Hail The Brain Slug]

Did you miss the following quote ?

This is not how SMT scheduling have worked in the past, nor how it should work
(for reference, check the numbers for 7700X how it behaves with SMT ON/OFF as a comparison)

Kinda seems more like its someone else that's "coping and grasping at straws" as you put it.. Why is that ?

I have a wild hypothesis for this behavior, hear me out - what if this is a feature?

What other related processor just came out where slamming as many threads of a process onto the first cores would actually be a performance benefit? Strix Point. Due to only having 4 high clocking cores in the primary CCX, if you had say, 8 demanding threads in a process, on Strix Point you would want them to be slapped onto the first 4 cores to get highest clockspeeds, more cache, and avoid cross-CCX penalty. It would most likely be faster than scheduling 4 on the first 4 cores and 4 onto 4 of the Zen 5C cores.

In addition, it may even be desirable to intentionally aggregate threads onto fewer cores for power efficiency reasons.

What if there's a bug somewhere and the scheduling behavior is treating Granite Ridge like Strix Point?

 

[gdansk]

But it doesn't do awesome in games so its a complete dumpster fire release like all the techtubers told me. /s

Personally, I think that's valid for some people. If a part doesn't offer them more gaming performance and that's what they care about then it is fine for them to disregard Zen 5 (for now).

For example, I don't quite care about overclocked results but I do appreciate that other people will like to play around with it.

 

[Thunder 57]

Personally, I think that's valid for some people. If a part doesn't offer them more gaming performance and that's what they care about then it is fine for them to disregard Zen 5 (for now).

For example, I don't quite care about overclocked results but I do appreciate that other people will like to play around with it.

It's valid, but I think the techtubers exaggerated some to get views. Clickbait, if you will. That's a better way of saying what I meant.

 

[marees]

I have a wild hypothesis for this behavior, hear me out - what if this is a feature?

What other related processor just came out where slamming as many threads of a process onto the first cores would actually be a performance benefit? Strix Point. Due to only having 4 high clocking cores in the primary CCX, if you had say, 8 demanding threads in a process, on Strix Point you would want them to be slapped onto the first 4 cores to get highest clockspeeds, more cache, and avoid cross-CCX penalty. It would most likely be faster than scheduling 4 on the first 4 cores and 4 onto 4 of the Zen 5C cores.

In addition, it may even be desirable to intentionally aggregate threads onto fewer cores for power efficiency reasons.

What if there's a bug somewhere and the scheduling behavior is treating Granite Ridge like Strix Point?

Its probably more of a feature than a bug.

AMD has probably optimized their resource layout assuming SMT is enabled. Disabling SMT could make the processor inefficient

Intel & AMD seem to be on different trajectories here. AMD is pushing a server first design to client (as AMD's primary market is server), whereas Intel has a large client market to cater to; so probably dedicates features for client

 

[linkgoron]

But it doesn't do awesome in games so its a complete dumpster fire release like all the techtubers told me. /s

According to TPU, the 9700x is 10% faster than the 7700 (non x) in non-gaming, while being 10% more power efficient while costing much more. Am I supposed to be amazed by this achievement? You can toss the TDP limit aside, and gain ~5% more performance, at the cost of efficiency. I'm OK with that tradeoff, but even then the 9700x is basically just ~10% faster than the 7700x.

At $360, the 9700x is just priced too high, even ignoring Intel. For gaming, you can get a 7800x3D for $366. For non-gaming you can get a 7900X for $358. It's in a tough spot, and AMD hasn't done enough to make it a winner.

 

[StefanR5R]

Its probably more of a feature than a bug.

AMD has probably optimized their resource layout assuming SMT is enabled. Disabling SMT could make the processor inefficient

That could be true. But on Personal Computers (mainly doing interactive work), per-thread performance in lightly threaded workloads is still typically of higher importance than performance/Watt (or task energy) when not running on battery and not in a cooling constrained form factor. That is, on generic PCs with Granite Ridge, utilizing only one hardware thread per core if possible should still be the preferred scheduling policy in most use cases.

 

[StefanR5R]

Issues like this *might* be caused by a power-saver or bugged power plan/scheduling scheme. "Filling" all physical cores first should be the energy efficient strategy.

The latter may have been true traditionally, but is it still true with Zen 5?

(I have no idea myself; I only ever looked at SMT-on vs. SMT-off power efficiency in rate-N loads, not in "rate-2…8". Apropos, for Strix Point's classic cores, a SPEC Integer rate-N performance-per-package-power graph with and without SMT was posted by David Huang.)

 

[PJVol]

At $360, the 9700x is just priced too high, even ignoring Intel. For gaming, you can get a 7800x3D for $366. For non-gaming you can get a 7900X for $358. It's in a tough spot, and AMD hasn't done enough to make it a winner.

I tend to agree with Coreteks (youtube) on that AMD doesn't care much about DYI consumers or what reviewer's verdicts are. With this in mind, the Zen 5 chips look quite impressive.

 

[Fjodor2001]

I tend to agree with Coreteks (youtube) on that AMD doesn't care much about DYI consumers or what reviewer's verdicts are.

So what do they care about then?

 

[Timmah!]

So what do they care about then?

I watched the video. Servers/HPC pretty much. He shows bunch of benchmark tables, like Apache server performance, where Zen5 performs extremely well compared to older CPUs/intel…. Maybe cause of avx512? I would not know.

I dont exactly think highly of these utubers (this guy especially after he thought couple of years back rtx cards gonna have separate RT coprocessor chip, hence the weird cooler design, which even me as a layman, knew it was ridiculous) but in this case he is likely not far from truth.

 

[gdansk]

Servers on one side and laptops on the other.
But they definitely do care about reviewers - otherwise they wouldn't have had reviewer guides and so on.
And they desire gamer $ enough to try to market them as "gaming leadership" parts. But yes, I don't think AMD were panicking due to Zen 5 performance characteristics. Well, hmm, maybe they were. The short precall and all that.

 

[PJVol]

And they desire gamer $ enough to try to market them as "gaming leadership" parts.

Well, X3D's fits that claim nice, aren't they (according to the supposed strategy) ? )

 

[CouncilorIrissa]

So what do they care about then?

There's an answer in there somewhere.

 

[Fjodor2001]

View attachment 105077
There's an answer in there somewhere.

Well both client and server/DC are up substantially, if that’s what you mean.

So they ought to care about client too?

Also, what in Zen5 compared to Zen4 that we seen so far makes it a success for server/DC? AVX512?

 

[CouncilorIrissa]

Also, what in Zen5 compared to Zen4 that we seen so far makes it a success for server/DC?

It shreds in PHP, Node, Python, databases and Apache (the last one is seeing especially absurd gains).

Well both client and server/DC are up substantially, if that’s what you mean.

Client is mostly laptops running javascript which STX is pretty good at.
DIY desktop is a small niche that's barely relevant to AMD's bottom line.

 

[DrMrLordX]

Well both client and server/DC are up substantially, if that’s what you mean.

So they ought to care about client too?

DiY desktop is a small slice of client. Mostly that's laptops or prebuilt desktop.

Also, what in Zen5 compared to Zen4 that we seen so far makes it a success for server/DC? AVX512?

@CouncilorIrissa beat me to it so props there. Though it isn't just "pretty good" at js, it's flipping the script there and blowing everything the competition has out of the water.

 

[Tuna-Fish]

Every time I look more closely at STX results, all I can think of is that I would much rather have a single-CCX 8-core with a larger L3.

It would lose on, like, cinebench, but do better on nearly everything people actually care about.

 

[exquisitechar]

Every time I look more closely at STX results, all I can think of is that I would much rather have a single-CCX 8-core with a larger L3.

It would lose on, like, cinebench, but do better on nearly everything people actually care about.

Absolutely. Zen 5C would have best been kept for Turin Dense, the hybrid config in STX is unnecessary and mostly good for Cinebench bragging rights.

 

[CouncilorIrissa]

Every time I look more closely at STX results, all I can think of is that I would much rather have a single-CCX 8-core with a larger L3.

It would lose on, like, cinebench, but do better on nearly everything people actually care about.

Yeah.
I wonder if Kraken Point will outperform STX in some scenarios as it's a single-CCX design (although it's 4 classic + 4 dense) that won't suffer from horrific cross-CCX latency.

 

[inquiss]

There are many MT workloads that don’t require that much memory bandwidth.

So you hobble the platform for the hope that people buy it for those workloads. How many people that want that do you think there are? Hardly anyone buys the highest core count CPUs as it is. If you really want higher core counts you get threadripper.

 

[gdansk]

So you hobble the platform for the hope that people buy it for those workloads. How many people that want that do you think there are? Hardly anyone buys the highest core count CPUs as it is. If you really want higher core counts you get threadripper.

You'd be surprised how many people buy based on some irrelevant benchmark.
It's the main reason Strix is 4+8 instead of something more rational. If the competition is spamming cores you better too.

I don't think it would be a good part for many people but it might sell anyway.

 

[Mahboi]

It shreds in PHP, Node, Python, databases and Apache (the last one is seeing especially absurd gains).

The real shocker for server is that Zen 5 is an actual slam dunk.
- DBs
- in memory DBs
- web server
- python/JS/php

All that's left is for someone to test a load balancer and we're set. It's basically THE chip to buy for anything cloud/general web services. Freaking 8 core Zen 5 is nibbling at 13900ks/14700ks, with the odd "destroy everything including 14900Ks" benchmark.

If I were a CP I'd just buy EPYC Zen 5 and nothing else. It may be a bust in games and in so many apps, but for web services it's obliterating any competition.