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

[DisEnchantment]

Speculate at will

 

[Abwx]

That's pretty awful for 50% more cores and higher IPC.

Theorically they could get 27% better perf at same power if they used the same process and with 17% better IPC in CB R23, so the number is very good, it just require to know what is possible technicaly.

FTR 31.9% better perf at iso watt amount also to 1.92x better perf/watt at isoperf, so we can expect a 32W Strix Point to perform as well as a 60W Hawk Point.

 

[fastandfurious6]

Strix point R23: ~22000
strix point doesnt make much sense as product tbh, they just wanted the NPU "AI cpu" as soon as possible
it's only 4 full cores, 8x 5c is not just lower clocks but lower cache, smaller cores etc etc
definitely not worth upgrade from phoenix/hawk

Strix Halo R23 gonna be ~37000-41000 R23 MT

calling it now

 

[inf64]

The Zen 5C cores appear to be running at 3.65ghz in GB6.

Per AMD’s keynote slides CB R23 is one of the better performing workloads for Zen 5. They claimed a +17% IPC increase against Zen 4 at isoclocks. So not sure how that will effect the math of approximate clocks.

Well GB MT is not exactly the same as R23 MT workload. To me it looks like a fine result.

 

[adroc_thurston]

The Zen 5C cores appear to be running at 3.65ghz in GB6.

?

Per AMD’s keynote slides CB R23 is one of the better performing workloads for Zen 5

One of the worst, actually.

 

[H433x0n]

?

One of the worst, actually.

I have to be careful otherwise I’ll probably get reported and banned.

Dispassionately as possible: It was the 4th best result out of 13 total tests and the result was higher than AMD's stated geomean of 16%. I would say that qualifies as one of the better improvements for the new core.

 

[adroc_thurston]

I have to be careful otherwise I’ll probably get reported and banned

?

I would say that qualifies as one of the better improvements for the new core.

Nope, Blender, or anything else not scalar SSE from year 2007 does better.
Pretty sure even Cinememe'2024 does better.

I get it, you want to win so hard, but alas!
Now show me 3.65G dense on Strix and do try to correlate that with Strix shedding 1.8G in rated bclk.

 

[H433x0n]

?

Just got back from vacation and it was from posting in this thread.

Nope, Blender, or anything else not scalar SSE from year 2007 does better.
Pretty sure even Cinememe'2024 does better.

That is probably(?) true but it doesn’t invalidate my statement that out of the data provided by AMD’s Zen 5 launch the CB R23 benchmark is one of the favored workloads above their geomean.

Edit: The 3.65ghz was the clocks captured from GB6 nT test. I’ve got no idea what the clocks are during cinememe.

 

[Kaffeekenan]

?

Nope, Blender, or anything else not scalar SSE from year 2007 does better.
Pretty sure even Cinememe'2024 does better.

I get it, you want to win so hard, but alas!
Now show me 3.65G dense on Strix and do try to correlate that with Strix shedding 1.8G in rated bclk.

Yeah. That's why he used the word "better" and not the word "best" I guess?! Nobody denies, that there are better results...

 

[adroc_thurston]

AMD’s Zen 5 launch the CB R23 benchmark is one of the favored workloads above their geomean

Let's just say that contradicts stuff I saw on a GNR ES.
Is it latency-sensitive?

Either way, where's muh 3.65G Dense on Strix.
I'm waiting.

 

[gdansk]

Geomean 16% or "favored" 17% it doesn't impact the calculation, clocks must be lower than I expected.

Moreover, GB6 MT is not a good indicator of MT clock rates. It doesn't use enough of the cores, in my experience. I can lower my 12600K from 150W to 95W and the MT score and reported peak clock rates stay the same. But if I run Cinebench it does show lower scores and clock rates.

 

[adroc_thurston]

Geomean 16% or "favored" 17% it doesn't impact the calculation, clocks must be lower than I expected.

Strix starts losing nT luster pretty quickly above 30W since you're pushing dense way above 2W@core.

 

[Elfear]

AMD rep says that vanilla Zen 5 won’t top Zen 4x3d in gaming.

https://www.tomshardware.com/pc-components/cpus/amds-ryzen-9000-wont-beat-the-previous-gen-x3d-models-in-gaming-but-theyll-be-close-improved-3d-v-cache-coming-too

Some other good tidbits from that interview (quotes from David McAfee, Corporate VP and General Manager of the Client Channel Business at AMD & Donny Woligroski, Senior Technical Marketing Manager of Consumer Processors):

"I think the other thing that we'll get into is the frequency residency in spite of the fact that the Fmax (maximum frequency) hasn't really changed," McAfee said. "As far as what's on the box, the frequency residency, the efficiency of the lid and thermal design in the 9000 generation gives your effective frequency a lift over the prior generation. So you actually do see a net, overall positive there with just the processor running faster with the Zen 5 architecture versus Zen 4."

"At the end of the day, we bought a non-X3D chip very close to an X3D chip when it comes to gaming,"

"All of those things are pretty big differences compared to previous small steps and launches. We're going back on TDP because it turns out our eight cores are so good we don't need higher TDP, so I think it's a pretty stark comparison"

Sounds like the 9700X should be within a few percentage points of the 7800X3D (9950X may reach parity?). Also interesting that the boost clocks will maintain a higher frequency for longer if what they said is true.

 

[Hail The Brain Slug]

Some other good tidbits from that interview (quotes from David McAfee, Corporate VP and General Manager of the Client Channel Business at AMD & Donny Woligroski, Senior Technical Marketing Manager of Consumer Processors):

"I think the other thing that we'll get into is the frequency residency in spite of the fact that the Fmax (maximum frequency) hasn't really changed," McAfee said. "As far as what's on the box, the frequency residency, the efficiency of the lid and thermal design in the 9000 generation gives your effective frequency a lift over the prior generation. So you actually do see a net, overall positive there with just the processor running faster with the Zen 5 architecture versus Zen 4."

"At the end of the day, we bought a non-X3D chip very close to an X3D chip when it comes to gaming,"

"All of those things are pretty big differences compared to previous small steps and launches. We're going back on TDP because it turns out our eight cores are so good we don't need higher TDP, so I think it's a pretty stark comparison"

Sounds like the 9700X should be within a few percentage points of the 7800X3D (9950X may reach parity?). Also interesting that the boost clocks will maintain a higher frequency for longer if what they said is true.

I'd like to know what, exactly, "The efficiency of the lid and thermal design" means. I wonder if they will go into more details at the official launch event?

 

[Philste]

Asus is doing their very best to spoil the reviews in July. This appears to be an absolute apples/apples comparison with same chassis and power consumption.

Interesting that all these slides are openly available for over one week now.

These two are against 7735U aka Rembrandt btw.

1.51× CPU and 1.34× GPU over Rembrandt results in ~1.25× CPU and 1.2× GPU over Phoenix.

 

[Joe NYC]

AMD rep says that vanilla Zen 5 won’t top Zen 4x3d in gaming.

https://www.tomshardware.com/pc-components/cpus/amds-ryzen-9000-wont-beat-the-previous-gen-x3d-models-in-gaming-but-theyll-be-close-improved-3d-v-cache-coming-too

Nice for someone (Paul Alcorn of Toms Hardware) to actually include full quotes of AMD rep - to separate what was said from speculation.

 

[techjunkie123]

Interesting that all these slides are openly available for over one week now. View attachment 100972View attachment 100973

These two are against 7735U aka Rembrandt btw.

1.51× CPU and 1.34× GPU over Rembrandt results in ~1.25× CPU and 1.2× GPU over Phoenix.

Where are the slides available? Sorry for asking, not sure where to look. I wanted to look through them .

 

[Philste]

Where are the slides available

 

[Abwx]

...........

Interesting that all these slides are openly available for over one week now. View attachment 100972View attachment 100973

These two are against 7735U aka Rembrandt btw.

1.51× CPU and 1.34× GPU over Rembrandt results in ~1.25× CPU and 1.2× GPU over Phoenix.

The 1.51x CPU is the average of Cinebench and Procyon, the latter is not a good indicator of CPU perfs, in Cinebench it say 1.6x.

So Strix Point Point is 31.9% faster than Phoenix wich is 21% faster than RMB.

 

[Goop_reformed]

AMD rep says that vanilla Zen 5 won’t top Zen 4x3d in gaming.

https://www.tomshardware.com/pc-components/cpus/amds-ryzen-9000-wont-beat-the-previous-gen-x3d-models-in-gaming-but-theyll-be-close-improved-3d-v-cache-coming-too

This is pretty bad.

 

[inf64]

This is pretty bad.

The difference between 7700X and 7800X3D is ~18% according to CB . Let's wait for the benchmarks before we declare 9700X DOA.

 

[Thunder 57]

LOL at the 7900X3D. It now costs $329 while the 7800X3D is $10 more at $339. Yea I get it, it's on sale. I wonder if AMD will bother with a 9900X3D though. If I were in the market for a 12 core CPU, I would opt for the non-3D variant. Six cores with V-cache just isn't enough to be worth it.

 

[Goop_reformed]

The difference between 7700X and 7800X3D is ~18% according to CB . Let's wait for the benchmarks before we declare 9700X DOA.

Zen 5 is definitely not DOA but it's most definitely disappointing

 

[H433x0n]

LOL at the 7900X3D. It now costs $329 while the 7800X3D is $10 more at $339. Yea I get it, it's on sale. I wonder if AMD will bother with a 9900X3D though. If I were in the market for a 12 core CPU, I would opt for the non-3D variant. Six cores with V-cache just isn't enough to be worth it.

All of Zen 4 is on fire sale. The other day on Newegg the 7950X was going for $430. That’s an amazing deal.

 

[CouncilorIrissa]

Zen 5 is definitely not DOA but it's most definitely disappointing

Well it depends.
What we know so far: 16% PPC in mixed (nT/1T) workloads with basically no area penalty (the core itself is bigger, but the L3 is smaller, so it cancels out) and no clock regression in 1T at least. What we don't know is how much power it pulls. So it may very well end up a very successful core PPA-wise.

If you were expecting a chonker INT max go brrr core, then I guess so?

edit: it also seems like Z5 is one angry little FP shredder.

 

[Timmah!]

"At the end of the day, we give you more performance without increasing power, and at the end of the day, we give you more performance without increasing the heat. At the end of the day, we bought a non-X3D chip very close to an X3D chip when it comes to gaming,"

Someone should tell the guy they are not giving all that for free.
And if they were not giving it, "at the end of the day", why exactly should anyone consider buying it.