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

[DisEnchantment]

Speculate at will

 

[adroc_thurston]

So what happens to Zen 4?

It exists.

Do they stop making Zen 4 chips immediately or are they supposed to slot in below the Zen 5 SKUs for more pricing options for consumers?

DT/mobile Zen4 dies yeah.
Server, Genoa will live for some time on.

 

[StefanR5R]

one can always in hindsight claim that some specific benchmark was being referred to (but wasn't mentioned explicitly!)

It was explained that SPEC int rate was referred to. The claims which primarily irritated some readers were about 1-threaded 1-copy SPEC int rate.

 

[Fjodor2001]

It was explained that SPEC int rate was referred to.

Ok, and what was the claimed performance increase for that benchmark, and which CPUs where being compared?

 

[Kepler_L2]

Ok, and what was the claimed performance increase for that benchmark, and which CPUs where being compared?

Core for core Zen5 is >40% faster than Zen4 in SPEC.

 

[Joe NYC]

So what happens to Zen 4? Zen 3 is still being manufactured I suppose? Do they stop making Zen 4 chips immediately or are they supposed to slot in below the Zen 5 SKUs for more pricing options for consumers?

New_Zen4_Price = Old_Zen4_Price - $50
New_Zen5_Price = Old_Zen4_Price + $50

It will probably make sense to phase out Zen 4 SKUs once full slate of Zen 5 SKUs is shipping in volume, except some special SKUs notably:
- 7800X3D
- 7500F

 

[DisEnchantment]

Core for core Zen5 is >40% faster than Zen4 in SPEC.

INT or FP?
Would be more interesting if that is INT

 

[Kaffeekenan]

INT or FP?
Would be more interesting if that is INT

I am a technewb, please tell me why. Is integer used by more programms?

 

[Tuna-Fish]

Is integer used by more programms?

Yes. All programs use integer math (for pointers if nothing else), only some use fp. More than that, we can expect well parallelized fp (which spec has plenty of) to get a major boost, because simd execution width is doubled.

 

[Joe NYC]

INT or FP?
Would be more interesting if that is INT

INT is what most people care about. Both single thread and multi-thread.

If it is really > 40% single thread, it would be truly mind blowing performance gain.

 

[StefanR5R]

Turin spec OUT

https://twitter.com/x/status/1771516361329750298

View attachment 95732

The cache amounts are wrong for some of these, Zen 5C cores have a fraction the L3 available to them.

EDIT: Actually, there's probably a lot more wrong than that. Wouldn't make sense for there to be only one 500W SKU.

I don't read twitter, but it seems there was an update posted:
https://www.computerbase.de/2024-03...rver-cpus-mit-mehr-l3-cache-und-fragezeichen/ (scroll down to "Update 25.03.2024 07:34 Uhr")

This still claims that there is only one part with 500 W default TDP.¹ I also wonder about the cache amounts of the parts with multiple-of-3 core counts, of the F parts (frequency-optimized), and of some of the low core count parts (low core count combined with "low" cache overlaps with SP6/ Siena and possible successors).

¹) Edit: The table header says only "TDP". "default TDP" is just my own interpretation. Core count increase aside, given the widing of the cores, a bump of at least cTDP_up per core compared to Genoa, if not of default TDP, would make sense to me.

 

[Gideon]

INT is what most people care about. Both single thread and multi-thread.

If it is really > 40% single thread, it would be truly mind blowing performance gain.

Yeah, 40% was Zen 1 initial target (which they beat by actually achieving 52%).

As full "breaking, ground-up redesigns" of CPU's usually take at least ~5-7 years, I was hoping (ever since that famous Mike Clark interview) that Zen 5 would be it - with targets as aggressive as Zen 1. Containing all the results of the higher-risk / more conceptual research AMD has been doing ever-since 2017.

As beating Zen 4 by that amount (particularly on a similar node) is an insanely harder achievement than "clubbing baby dozer-cores", i'm still very sceptical ...

... but boy do I want to believe

 

[Joe NYC]

Yeah, 40% was Zen 1 initial target (which they beat by actually achieving 52%).

As full "breaking, ground-up redesigns" of CPU's usually take at least ~5-7 years, I was hoping (ever since that famous Mike Clark interview) that Zen 5 would be it - with targets as aggressive as Zen 1. Containing all the results of the higher-risk / more conceptual research AMD has been doing ever-since 2017.

As beating Zen 4 by that amount (particularly on a similar node) is an insanely harder achievement than "clubbing baby dozer-cores", i'm still very sceptical ...

... but boy do I want to believe

Exactly, after Zen 4 performance gains that were really quite decent, where Zen 4 already used up some of the N7 -> N5/N4 process technology gains.

Computex is in about 2.5 months from now, so not too long before we find out. Maybe some leaks between now and then.

Another thing to keep in mind is that unlike previous Zen generations, when AMD was struggling financially, Zen 5 team was not under the same financial pressure, and could pursue all of the avenues that looked promising, even more risky ones.

 

[adroc_thurston]

Core for core Zen5 is >40% faster than Zen4 in SPEC.

You weren't supposed to say that, little one.

Zen 5 team was not under the same financial pressure, and could pursue all of the avenues that looked promising, even more risky ones.

Well no, that's more of a willpower thing.
R&D money translated directly into more designs rather than more ambitious ones.

 

[igor_kavinski]

You weren't supposed to say that, little one.

Oh no! NDA breach! We'll miss you, Kepler

 

[AMDK11]

IPC increase for Zen 5 > 40% What does this refer to? Is this the average of the IPC growth curve? Up to 40-50%?

It's easy to throw around such numbers, but if there is no reference point and what type of load these numbers refer to, they don't tell us anything. It is equally easy to write that Zen 5 has an IPC >6%. And what does this tell us? Nothing.

Zen 3 with one specific workload (SPEC?) saw +106% compared to Zen 2. However, in the case of AMD's presentation, the IPC growth curve ends at +39% with an average of +19%.

Zen 4, also according to AMD's presentation, the IPC growth curve ends at +39%, but the average is already +13%.

 

[inf64]

IPC increase for Zen 5 > 40% What does this refer to? Is this the average of the IPC growth curve? Up to 40-50%?

It's easy to throw around such numbers, but if there is no reference point and what type of load these numbers refer to, they don't tell us anything. It is equally easy to write that Zen 5 has an IPC >6%. And what does this tell us? Nothing.

Zen 3 with one specific workload (SPEC?) saw +106% compared to Zen 2. However, in the case of AMD slide, the IPC growth curve ends at +39% with an average of +19%.

It was explained that the figure is from the aggregate SPEC score (which is calculated from the subscores).

 

[yuri69]

IPC increase for Zen 5 > 40% What does this refer to?

The statement concerns core-to-core performance. That metric doesn't really equal to IPC.

 

[adroc_thurston]

That metric doesn't really equal to IPC.

It really does, those cores already sit high 5s and can't really go much further wrt clkspeed.

 

[AMDK11]

So how much did Zen 3 gain in SPEC over Zen 2 and Zen 4 over Zen 3?

 

[adroc_thurston]

So how much did Zen 3 gain in SPEC over Zen 2 and Zen 4 over Zen 3?

IPC? 19 and 13% respectively.

 

[igor_kavinski]

19

Double jeopardy event for Intel compared to their embarrassment with Zen 3!

 

[adroc_thurston]

Double jeopardy event for Intel compared to their embarrassment with Zen 3!

Yeaaaa let's just say IDC needs something far meaner than LNC to compete.

 

[AMDK11]

IPC? 19 and 13% respectively.

+19% and +13% are the average of the IPC growth curve from programs and games. I asked about the SPEC result.

+40% for Zen 5 is certainly not average and will this translate into average in many applications? I am not a carefree optimist, but to err is not a sin.

on average +40%, how much would the Zen 5 core have to be expanded? IPC does not come out of thin air, but requires reconstruction and strong expansion of the logic controlling the core and resources. On average IPC +40%, i.e. a curve from approximately +10-15 to 70-90%? Oh, it's going to be a disappointment.

 

[adroc_thurston]

+19% and +13% are the average of the IPC growth curve from programs and games. I asked about the SPEC result.

The SPEC results are the exact same lmao.

+40% for Zen 5 is certainly not average

AMD SPEC results translate into averages in just a truly wonderful fashion.

On average IPC +40%, i.e. a curve from approximately +10-15 to 70-90%? Oh, it's going to be a disappointment.

You're shadowboxing stuff.
The numbers are SIR2017 rate 1/n.

 

[leoneazzurro]

According to this Anandtech review:

AMD Zen 4 Ryzen 9 7950X and Ryzen 5 7600X Review: Retaking The High-End

www.anandtech.com

The average SPECINT 2017 ST increase from Zen3 to Zen4 is 23% considering the average of all substests, considering the platform differences there is probably some lower margin due to the cores only because of memory BW advantage of Zen4, which should not exist with Zen5.