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

[DisEnchantment]

Speculate at will

 

[csbin]

AMD Strix Point Blender benchmark

https://twitter.com/x/status/1791030712436682942

 

[ToTTenTranz]

Matching 16 core Zen3 Threadripper is crazy! That thing has a 280W TDP and over 200GB/s bandwidth from 8-channel DDR4.

 

[Kepler_L2]

Matching 16 core Zen3 Threadripper is crazy! That thing has a 280W TDP and over 200GB/s bandwidth from 8-channel DDR4.

3900X has the same score too. Looks like a weird benchmark. Also we don't know clock speeds.

 

[leoneazzurro]

3900X has the same score too. Looks like a weird benchmark. Also we don't know clock speeds.

38 in the OP code means it's limited at 3.8 GHz, doesn't it?

 

[Kepler_L2]

38 in the OP code means it's limite do 3.8 GHz, doesn't it?

I'm not sure that's what 38 means. And since this is STX1 we don't know if it's Zen5 or Zen5c clocks

 

[leoneazzurro]

I'm not sure that's what 38 means. And since this is STX1 we don't know if it's Zen5 or Zen5c clocks

That's also true, power limit is also unknown

 

[Gideon]

Matching 16 core Zen3 Threadripper is crazy! That thing has a 280W TDP and over 200GB/s bandwidth from 8-channel DDR4.

Easy now, from the same results:

• 7845HX gets 356.92 - this is a chiplet-based 12 core (repurposed desktop) CPU (1 result in database)
• 7700X (desktop) gets 269.02 (24 results)
• 7940HS gets 224.22 (16 results)
• 7840U gets 216.09 (1 result) and 153.31 (for 5 results) not sure why they are separate
• 8840U gets 205.62 (1result)

So in the end it all depends on what TDP that chip was running at. Let's not forget it's a 12 core chip.

These numbers are very-very impressive numbers if configured for a TDP of a 15 - 28W U level chip (e.g. 7840U).
Not so much so for a 35-54W HS chip (as it's only about 20% faster than a 7940HS with 50% more cores)

 

[Abwx]

I'm not sure that's what 38 means. And since this is STX1 we don't know if it's Zen5 or Zen5c clocks

In principle this is the base frequency, this was always the case for AMD s ESs since a decade if not more.

 

[Kepler_L2]

In principle this is the base frequency, this was always the case for AMD s ESs since a decade if not more.

That would mean lower IPC than Zen2, which I think is a bit unlikely

 

[poke01]

Easy now, from the same results:

• 7845HX gets 356.92 - this is a chiplet-based 12 core (repurposed desktop) CPU (1 result in database)
• 7700X (desktop) gets 269.02 (24 results)
• 7940HS gets 224.22 (16 results)
• 7840U gets 216.09 (1 result) and 153.31 (for 5 results) not sure why they are separate
• 8840U gets 205.62 (1result)

So in the end it all depends on what TDP that chip was running at. Let's not forget it's a 12 core chip.

These numbers are very-very impressive numbers if configured for a TDP of a 15 - 28W U level chip (e.g. 7840U).
Not so much so for a 35-54W HS chip (as it's only about 20% faster than a 7940HS with 50% more cores)

Before I can offer some input, what’s the difference with a Zen5 and Zen5c core?

I know they are the same architecture but what’s different about them?
Thanks

 

[DisEnchantment]

Easy now, from the same results:

• 7845HX gets 356.92 - this is a chiplet-based 12 core (repurposed desktop) CPU (1 result in database)
• 7700X (desktop) gets 269.02 (24 results)
• 7940HS gets 224.22 (16 results)
• 7840U gets 216.09 (1 result) and 153.31 (for 5 results) not sure why they are separate
• 8840U gets 205.62 (1result)

So in the end it all depends on what TDP that chip was running at. Let's not forget it's a 12 core chip.

These numbers are very-very impressive numbers if configured for a TDP of a 15 - 28W U level chip (e.g. 7840U).
Not so much so for a 35-54W HS chip (as it's only about 20% faster than a 7940HS with 50% more cores)

Similar to the GB5 results, FP/vector ops seems to not have as much gain as the rest.
Likely locked microcode/boost is inhibiting full perf, otherwise the FP speedup seems kind of mediocre if not stagnant.

 

[Abwx]

That would mean lower IPC than Zen2, which I think is a bit unlikely

That s an ES, and likely with configurable TDP, so without specification about the TDP used for this run the only sure info is that they have chips that could sustain 3.8GHz base frequency, wich btw is likely the c cores max frequency.

 

[deasd]

Computerbase has results of efficiency tests like 7900X(45w) and 7950X(45w)

AMD Ryzen 7000 im Test: So schnell sind 7950X und 7700X: Benchmarks in Apps und Games

Ryzen 9 7950X & Ryzen 7 7700X im Test: Benchmarks in Apps und Games / Leistung in Single-Core-Lasten (klassisch)

www.computerbase.de

Both 7900X(45w) and 7950X(45w) are 20-30% slower than a stock 7700X(105w) under blender test.

So yes, I can only say the leaked STX sample(45w?) is 20-30% faster than 7900X(45w) and 7950X(45w).

Ryzen 7 7700X DDR5-5200CL32	10:40
Ryzen 9 7950X (45 W) DDR5-5200CL32	12:45
Ryzen 9 7900X (45 W) DDR5-5200CL32	13:42

Also, the leaked STX sample seems to be the same as the old 1.4-2.0Ghz one running Geekbench (100-000000994-14_N)

AMD Ryzen 9050 "Strix Point" APU with 12 cores spotted on Geekbench - VideoCardz.com

AMD Strix Point engineering sample Over at Geekbench there is an entry featuring AMD yet unreleased processor. The APU, featuring an OPN (product) code of 100-000000994-14_N, is part of AMD’s upcoming deployment of the Strix Point processor series. This information is confirmed by Family 26 and...

videocardz.com

the new one 100-000000994-38_N seems to have 3.8Ghz stock frequency

 

[leoneazzurro]

Already posted?

https://twitter.com/x/status/1790989196225925318

 

[SteinFG]

Don't wonna log into twitter, what does this one say? Please summarize the tweet if you're just posting a link

 

[moinmoin]

Don't wonna log into twitter, what does this one say? Please summarize the tweet if you're just posting a link

It's just posting following image:

 

[SteinFG]

They fit 16 chiplets plus giant IO die on one package? Honestly It was already bonkers with 12 CCDs

 

[adroc_thurston]

They fit 16 chiplets plus giant IO die on one package?

yessir.

Honestly It was already bonkers with 12 CCDs

Really funky, world-class substrate design!

 

[Abwx]

Easy now, from the same results:

• 7845HX gets 356.92 - this is a chiplet-based 12 core (repurposed desktop) CPU (1 result in database)
• 7700X (desktop) gets 269.02 (24 results)
• 7940HS gets 224.22 (16 results)
• 7840U gets 216.09 (1 result) and 153.31 (for 5 results) not sure why they are separate
• 8840U gets 205.62 (1result)

So in the end it all depends on what TDP that chip was running at. Let's not forget it's a 12 core chip.

These numbers are very-very impressive numbers if configured for a TDP of a 15 - 28W U level chip (e.g. 7840U).
Not so much so for a 35-54W HS chip (as it's only about 20% faster than a 7940HS with 50% more cores)

All those references are meaningless since TDP is unknown for all these parts excepted for the 7700X.

Using this CPU as reference and assuming that cores power is 125W then Strix Point has something like 20% better perf/clock if the run was made at 45W, this as an estimation based on power/frequency curves, assuming power scales at an average 2.8 power of frequency between 80 and 125W and 2.36 average between 45 and 80W.

FI computerbase measured the 7700X as 26% faster in Blender than a 7940HS at around 70-90W, their run over 100s display about 85W but Blender test last much more at 800s and they state that the CPU end running at 70W.

If we extrapolate from the Computerbase test, and using the 70W lower bound to not overestimate the calculation, then this Strix Point ES perform 26% better than a 7940HS@70W and about 52% better than a 7940HS@45W, so assuming a run at 45W that would make 1.52x the perf with 1.5x the core count, and this would also imply 20% better perf/clock.

Razer Blade 14 (2023) mit 7940HS und RTX 4070 im Test

Die 3. Generation Razer Blade 14 setzt auf AMD Phoenix und Nvidia Ada Lovelace. Statt 16:9 mit 144 Hz gibt es 16:10 mit 240 Hz. Der Test.

www.computerbase.de

 

[adroc_thurston]

If we extrapolate from the Computerbase test and using the 70W lower bound to not overestimate the calculation then this Strix Point ES perform 26% better than a 7940HS@70W and about 52% better than a 7940HS@45W, so assuming a run at 45W that would make 1.52x the perf with 1.5x the core count, and this would also imply 20% better perf/clock

You're close!
Very close.

 

[H433x0n]

.. FI computerbase measured ..

Off topic but what do you mean by "FI" in this statement?

 

[Abwx]

Off topic but what do you mean by "FI" in this statement?

For instance.

 

[Kaffeekenan]

You're close!
Very close.

Isn't 20% more IPC kinda low? You said something around 30-40 if I remember correctly? So were you wrong or what happened?

 

[inquiss]

All those references are meaningless since TDP is unknown for all these parts excepted for the 7700X.

Using this CPU as reference and assuming that cores power is 125W then Strix Point has something like 20% better perf/clock if the run was made at 45W, this as an estimation based on power/frequency curves, assuming power scales at an average 2.8 power of frequency between 80 and 125W and 2.36 average between 45 and 80W.

FI computerbase measured the 7700X as 26% faster in Blender than a 7940HS at around 70-90W, their run over 100s display about 85W but Blender test last much more at 800s and they state that the CPU end running at 70W.

If we extrapolate from the Computerbase test, and using the 70W lower bound to not overestimate the calculation, then this Strix Point ES perform 26% better than a 7940HS@70W and about 52% better than a 7940HS@45W, so assuming a run at 45W that would make 1.52x the perf with 1.5x the core count, and this would also imply 20% better perf/clock.

Razer Blade 14 (2023) mit 7940HS und RTX 4070 im Test

Die 3. Generation Razer Blade 14 setzt auf AMD Phoenix und Nvidia Ada Lovelace. Statt 16:9 mit 144 Hz gibt es 16:10 mit 240 Hz. Der Test.

www.computerbase.de

Does that account for 4 of the cores being zen5c?

 

[Abwx]

Does that account for 4 of the cores being zen5c?

That doesnt matter since on MT and a power like 45W the regular Zen 5 cores wont clock higher than the c cores because of the limited TDP, it s highly likely that they all work at a same frequency.