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

Page 321 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

tamz_msc

Diamond Member
Jan 5, 2017
3,865
3,729
136
IPC isn't free, you're paying in area and power.
If the cost is indeed area and power, then the 8-core having a TDP of 170 W, ie. 230 W PPT means that the 16-core flagship will be power limited if 230 W is the limit of the platform.

That would mean that the gains from the 7950X to 16c-Granite Ridge would be lower than the gains from 7700X to 8c-Granite Ridge.

And you're saying that they're gonna charge an almost 100% premium over the 7950X for the 16c, which goes for $550 at the moment.

It doesn't add up.
 

gdansk

Diamond Member
Feb 8, 2011
3,109
4,823
136
230W is the limit of the power discussion. AM5 specified that.

I won't even consider a 170W (actually 230W) single CCd design. It'd throttle itself. And dual 4 core CCd is also unlikely because of interconnect latency. Has to be a typo or a test product.

And by Intel logic the 7950X is already "relatively" power limited compared to the 7900X and 7700X if you do the math as PPT divided by core count. But in reality? Not really, about 100mhz higher all core boost than 7900X and 75mhz lower than 7700X.
 
Last edited:

adroc_thurston

Diamond Member
Jul 2, 2023
3,649
5,271
96
If the cost is indeed area and power, then the 8-core having a TDP of 170 W, ie. 230 W PPT means that the 16-core flagship will be power limited if 230 W is the limit of the platform.
Every bit as limited as Zen4 one so nothing changes.
And you're saying that they're gonna charge an almost 100% premium over the 7950X for the 16c, which goes for $550 at the moment.

It doesn't add up.
Are you seriously comparing street prices to SRP?
I won't even consider a 170W (actually 230W) single CCd design. It'd throttle itself
Oh it's a sucker SKU indeed.
Much the same way 5800X was.
But you don't have a choice.
 
Last edited:

Abwx

Lifer
Apr 2, 2011
11,591
4,408
136
Be it Zen 3 or 4 current 8C are limited at about 130-140W, it wont be different for Zen 5, anyone saying otherwise is out of touch with reality, more power is simply no feasible due to thermal density.

In this respect current CPUs are already well above the accepted enginering rules of the 2000-2015 era, check the thermal density of past CPUs, even the once much decried FX9590 with its 220W TDP was at a mere 0.69W/mm2, and that was only when stressing the chip with Prime 95.
 
Reactions: Tlh97 and Rigg

tamz_msc

Diamond Member
Jan 5, 2017
3,865
3,729
136
Let's do some quick estimates:

IN CB r23, Zen 3 to Zen 4 was 9% higher perf at fixed 4 GHz comparing 8 core to 8 core (source).

Now since Zen 5 is rumored to have more gains in the front-end and since r23 is front-end bound, lets assume that under the same conditions Zen 5 is 20% faster than Zen 4.

Now in the actual 8-core SKUs if the new TDP is indeed 170W, then to account for a 62% increase in power, vs 105W, the frequency would have to increase by roughly 15% (power-frequency dependency roughly following cubic relationship).

So 15% frequency increase and 20% ppc increase would put this 8-core Zen 5 ~38% faster than the 8-core 7700X. This is in line with the rumored performance uplift of 40%.

HOWEVER.

This increase is coming partly from increased clocks, which would mean that for the 16c Zen5 and Zen4 parts, which are both 170W, the benefit of the frequency increase will be absent. So, theoretically, in r23, the gain over the 7950X would be 25% at most.

40% higher core for core perf suddenly doesn't seem that impressive when you consider the actual CPUs in question, especially at the high-end.
 
Reactions: Apokalupt0

gdansk

Diamond Member
Feb 8, 2011
3,109
4,823
136
7700X all core AVX clock rate is usually around 5.3GHz. Multiply by 1.15 ≈ 6.1GHz.
6.1GHz all core In an AVX workload like CB r23?
On that idea I call bogus.
 

tamz_msc

Diamond Member
Jan 5, 2017
3,865
3,729
136
7700X all core AVX clock rate is usually around 5.3GHz. Multiply by 1.15 ≈ 6.1GHz.
6.1GHz all core In an AVX workload like CB r23?
On that idea I call bogus.
I don't see why 6 GHz in r23 is impossible given that we assume the rumors to be true. r23 doesn't use much vector math, and AVX/2/512 throttling is mostly solved since Ice Lake.
 

gdansk

Diamond Member
Feb 8, 2011
3,109
4,823
136
I don't see why 6 GHz in r23 is impossible given that we assume the rumors to be true. r23 doesn't use much vector math, and AVX/2/512 throttling is mostly solved since Ice Lake.
It is 'solved' in Zen 4 by dropping clock rates dynamically but only a little. And like most raytracers r23 uses pretty much only SIMD instructions.
If AMD is doing a sucker 170W 8-core SKU I wager it's only going to get another 2-300MHz over the 16 core version and it'll always be at 95C limiting itself.
 

tamz_msc

Diamond Member
Jan 5, 2017
3,865
3,729
136
It is 'solved' in Zen 4 by dropping clock rates dynamically but only a little. And like most raytracers r23 uses pretty much only SIMD instructions.
If AMD is doing a sucker 170W 8-core SKU I wager it's only going to get another 2-300MHz over the 16 core version and it'll always be at 95C limiting itself.
because this is a very high IPC core.
Here's what I understand as to putting the "40% core for core" higher performance of Zen 5 in context.

The only way you are going to get close to that figure is by increasing clock speeds AND a new, high-IPC core.

It may be possible to do so by making the 8c part 170W to get more frequency, but for the 16c parts you're not going to get 40% higher performance in something like CB r23 as they are already at 170W.
 

Hitman928

Diamond Member
Apr 15, 2012
6,328
11,123
136
Here's what I understand as to putting the "40% core for core" higher performance of Zen 5 in context.

The only way you are going to get close to that figure is by increasing clock speeds AND a new, high-IPC core.

It may be possible to do so by making the 8c part 170W to get more frequency, but for the 16c parts you're not going to get 40% higher performance in something like CB r23 as they are already at 170W.

True or not, the 40+% rumors have been single threaded performance, so power limits won't come into play.
 

gdansk

Diamond Member
Feb 8, 2011
3,109
4,823
136
What will be the Geekbench ST for Zen 5?

Everyone keeps speculating on Cinebench
Unhappy with Cinebench speculation?
Then let me show you some happy data?

9950X

GeekBench 6
single 3628 more than 23868

GeekBench 5
single 2715 more than 27712

(Only test data, does not represent the final result)
 

Abwx

Lifer
Apr 2, 2011
11,591
4,408
136
Let's do some quick estimates:

IN CB r23, Zen 3 to Zen 4 was 9% higher perf at fixed 4 GHz comparing 8 core to 8 core (source).

Now since Zen 5 is rumored to have more gains in the front-end and since r23 is front-end bound, lets assume that under the same conditions Zen 5 is 20% faster than Zen 4.

Now in the actual 8-core SKUs if the new TDP is indeed 170W, then to account for a 62% increase in power, vs 105W, the frequency would have to increase by roughly 15% (power-frequency dependency roughly following cubic relationship).

So 15% frequency increase and 20% ppc increase would put this 8-core Zen 5 ~38% faster than the 8-core 7700X. This is in line with the rumored performance uplift of 40%.

HOWEVER.

This increase is coming partly from increased clocks, which would mean that for the 16c Zen5 and Zen4 parts, which are both 170W, the benefit of the frequency increase will be absent. So, theoretically, in r23, the gain over the 7950X would be 25% at most.

40% higher core for core perf suddenly doesn't seem that impressive when you consider the actual CPUs in question, especially at the high-end.

All this is hugely wrong, because N4P used for Zen 5 has either 11% better perf/isowatt or 22% lower power at isoclock (= 28% higher perf/watt at isoclock) than the N5 used for DT Zen 4.

So if the IPC is say 28% higher FI then at same frequency Zen 5 will use the same power as Zen 4 while providing 28% more throughput assuming uarchitectural efficency is similar.

That is, if we use you CB R23 exemple that means of course 28% higher score at same power, so they dont even need 40% IPC to increase perfs by 25%, about 20% better IPC is enough.

Now if they use N4X (15% higher perf/isowatt or 28% lower power/isoclock than N5 = 39% higher perf/watt isoclock) that would extend even further the possibilities, in this case they can increase the IPC and hence throughput by 39% and still have the same power as Zen 4, and if they have only 25% better IPC it will consume much less than Zen 4 and still provide 25% better perfs.

 
Last edited:

tamz_msc

Diamond Member
Jan 5, 2017
3,865
3,729
136
True or not, the 40+% rumors have been single threaded performance, so power limits won't come into play.
That is what makes the $1000 rumored price tag for the 9950X for a measly 20-25% gain in CB r23 over the 7950X sus.
All this is hugely wrong, because N4P used for Zen 5 has either 11% better perf/isowatt or 22% lower power at isoclock (= 28% higher perf/watt at isoclock) than the N5 used for DT Zen 4.
Those figures are almost always obtained using some small reference core like the Cortex A78. Doesn't mean a damn thing in reality when trying to extrapolate to high frequency x86 cores.
 
Reactions: Apokalupt0

branch_suggestion

Senior member
Aug 4, 2023
401
892
96
Assumed ST gains are hovering around ~45%, PPW is only around ~25% from what has been said, the small node bump is more than offset by higher dynamic capacitance.
IPC and a small clock bump are not free. So that leaves the third arm of the holy trinity, PPA. My guess is PPA is probably only mid teens better, getting that sweet core/CCD area would be a big leak.

Lastly is MT perf/efficiency, which is largely a thermally limited mess to figure out. Performance gains iso core at different PPT's will be interesting, the thing is probably at it's best at 3-5Ghz relative to Z4.
 

Markfw

Moderator Emeritus, Elite Member
May 16, 2002
26,258
15,388
136
Mark, that is you. Take China market which are big market for AMD and Intel, people there are price sensitive. If AMD/Intel price the CPU too high, they won't get much sales. That's why AMD has to adjust Zen4 pricing after two months due to lack of sales.
As was said before, if you want performance, and the best, its not free.
Here's what I understand as to putting the "40% core for core" higher performance of Zen 5 in context.

The only way you are going to get close to that figure is by increasing clock speeds AND a new, high-IPC core.

It may be possible to do so by making the 8c part 170W to get more frequency, but for the 16c parts you're not going to get 40% higher performance in something like CB r23 as they are already at 170W.
Why don't you save your arguments until its out, and published benchmarks and reviews are released. Arguing now is kind of pointless.

And I just saw you cb23 comment. The biggest reason I like Zen 4 is avfx-512, we use it a lot in the DC forum.
 

Abwx

Lifer
Apr 2, 2011
11,591
4,408
136
That is what makes the $1000 rumored price tag for the 9950X for a measly 20-25% gain in CB r23 over the 7950X sus.

Those figures are almost always obtained using some small reference core like the Cortex A78. Doesn't mean a damn thing in reality when trying to extrapolate to high frequency x86 cores.

That change nothing, it s a comparison between N5 and N4P/X, the core design is of no importance here, the process perfs and perf/watt difference will be the same whatever the core, if a X86 core get 25% better perf/isowatt when shrinking then a Cortex will also benefit from those 25%/isowatt once shrinked.
 

tamz_msc

Diamond Member
Jan 5, 2017
3,865
3,729
136
That change nothing, it s a comparison between N5 and N4P/X, the core design is of no importance here, the process perfs and perf/watt difference will be the same whatever the core, if a X86 core get 25% better perf/isowatt when shrinking then a Cortex will also benefit from those 25%/isowatt once shrinked.
No, as usual you understand very little of how process nodes can be tweaked to account for target PPA which is variable based on the design under consideration.

An A78 shrunk from N5 to N4 will certainly not be the same as a Zen 5 shrunk from N5 to N4 as they are vastly different PPA targets.
 
Reactions: Apokalupt0

Hitman928

Diamond Member
Apr 15, 2012
6,328
11,123
136
That is what makes the $1000 rumored price tag for the 9950X for a measly 20-25% gain in CB r23 over the 7950X sus.

Those figures are almost always obtained using some small reference core like the Cortex A78. Doesn't mean a damn thing in reality when trying to extrapolate to high frequency x86 cores.

Retail CPU sales aren’t driven by Cinebench scores, that’s mostly just forum fodder. Gaming drives retail sales. If Zen 5 can get 40% increase in gaming performance, there will be a lot of demand. Enough to sustain the increased asking prices being thrown out, I have my doubts, but we’ll see.
 
sale-70-410-exam    | Exam-200-125-pdf    | we-sale-70-410-exam    | hot-sale-70-410-exam    | Latest-exam-700-603-Dumps    | Dumps-98-363-exams-date    | Certs-200-125-date    | Dumps-300-075-exams-date    | hot-sale-book-C8010-726-book    | Hot-Sale-200-310-Exam    | Exam-Description-200-310-dumps?    | hot-sale-book-200-125-book    | Latest-Updated-300-209-Exam    | Dumps-210-260-exams-date    | Download-200-125-Exam-PDF    | Exam-Description-300-101-dumps    | Certs-300-101-date    | Hot-Sale-300-075-Exam    | Latest-exam-200-125-Dumps    | Exam-Description-200-125-dumps    | Latest-Updated-300-075-Exam    | hot-sale-book-210-260-book    | Dumps-200-901-exams-date    | Certs-200-901-date    | Latest-exam-1Z0-062-Dumps    | Hot-Sale-1Z0-062-Exam    | Certs-CSSLP-date    | 100%-Pass-70-383-Exams    | Latest-JN0-360-real-exam-questions    | 100%-Pass-4A0-100-Real-Exam-Questions    | Dumps-300-135-exams-date    | Passed-200-105-Tech-Exams    | Latest-Updated-200-310-Exam    | Download-300-070-Exam-PDF    | Hot-Sale-JN0-360-Exam    | 100%-Pass-JN0-360-Exams    | 100%-Pass-JN0-360-Real-Exam-Questions    | Dumps-JN0-360-exams-date    | Exam-Description-1Z0-876-dumps    | Latest-exam-1Z0-876-Dumps    | Dumps-HPE0-Y53-exams-date    | 2017-Latest-HPE0-Y53-Exam    | 100%-Pass-HPE0-Y53-Real-Exam-Questions    | Pass-4A0-100-Exam    | Latest-4A0-100-Questions    | Dumps-98-365-exams-date    | 2017-Latest-98-365-Exam    | 100%-Pass-VCS-254-Exams    | 2017-Latest-VCS-273-Exam    | Dumps-200-355-exams-date    | 2017-Latest-300-320-Exam    | Pass-300-101-Exam    | 100%-Pass-300-115-Exams    |
http://www.portvapes.co.uk/    | http://www.portvapes.co.uk/    |