AMD Ryzen (Summit Ridge) Benchmarks Thread (use new thread)

[bjt2]

I read that Zen overclocks to about 4.2GHz. So at least with Zen people will stop complaining about Intel's processors not overclocking high enough since Zen (even) worse. That same dude says it matches Haswell performance.

Don't know if this was already posted here.



https://webcache.googleusercontent....20/amd-zen-141214/+&cd=2&hl=en&ct=clnk&gl=eng
https://webcache.googleusercontent....41214/index3.html+&cd=3&hl=eng&ct=clnk&gl=eng

That clock is of an ES, provided that is not guessed, but verified. According to pipeline stage number, Zen should have the same FO4 of XV, that on the shitty 28nm BULK reaches 4.3GHz turbo, so not even overclocked... Since Polaris gained about 20% in clock from 28 to 14, how come that Zen loses clock instead? The maximum clock depends from FO4, not IPC... It's not automatic that an higher IPC design clocks lower... So i think that OC clock was invented or guessed, because i would espect an OC at least same that the A12 9800 (4.8 GHz on air), that is on the 28nm BULK...

 

[bjt2]

Which does't use AVX/-2...

Sure, but this is also the official version downloadable from the site, if i remember well... Here some user managed to recompile with AVX2, obtaining an interesting gain, but cleverly AMD choose a bench in which Zen could shine with the official release... In the golden days of Pentium4, INTEL did the same, maybe even worse: the INTEL compiler let run optimized code only on INTEL CPUs and many softwares were compiled with it...

It doesn't change the picture: they are not enough. Not even the L1C is enough. And we don't know the politics which are used to cache the uops.

Ok, but INTEL is in the same situation... Actually worse for the L1I cache, because Zen has double the value... I don't know the uop cache situation...

Sure: one of the most common case (LOAD_FAST) is very fast and needs a few instructions. But talking about "few" it means around ten if you follow the path from the bytecode fetch 'til the real executing, and the jump back to the fetch code section, and consider that 3 conditional jumps are executed, plus an unconditional one (at the very end).

I would guess that the fetch and other common code are exactly the same instructions that probabily sit in the L1I and uop caches, because they are used at every emulated instruction and the LRU algorithm don't let them be evicted...

 

[mikk]

Yes but not 2+2

2+2 could be good enough for Raven Ridge.

 

[AtenRa]

2+2 could be good enough for Raven Ridge.

Against a 4C 8T + 768 Shaders or higher iGPU Raven Ridge ?? It will only be faster at Single Thread and lose in everything else (performance related)

 

[mikk]

Against a 4C 8T + 768 Shaders or higher iGPU Raven Ridge ?? It will only be faster at Single Thread and lose in everything else (performance related)

Sure. Look at FX-9800P, 4 cores with 512 shaders and is losing in both CPU and GPU against Kabylake. Even if there is a big increase with Raven Ridge, let's say 50%, Cannonlake 2+2 could be good enough. Honestly I expect CNL 2+2 to be better than Raven Ridge in a 15W power envelope.

 

[AtenRa]

Sure. Look at FX-9800P, 4 cores with 512 shaders and is losing in both CPU and GPU against Kabylake.

Bristol Ridge 28nm PLANAR (Bulldozer + GCN 1.3) vs Kabylake 14nm FF
Raven Ridge 14nm FF (ZEN + Vega) vs 10nm CNL 10nm

CNL 2+2 doesnt have a chance in MT and iGPU performance against a 4C 8T + Vega iGPU Raven Ridge.

 

[mikk]

Bristol Ridge 28nm PLANAR (Bulldozer + GCN 1.3) vs Kabylake 14nm FF
Raven Ridge 14nm FF (ZEN + Vega) vs 10nm CNL 10nm

Pointless, because it doesn't say anything about the competitiveness from Raven Ridge against Cannonlake. They are too many unknown variables. But one is clear, historically AMD struggled in a low power envelope even with much bigger GPUs that were much better on paper. And you can be sure Raven Ridge will be power limited if it includes a 768 SP GPU. You can also expect that Intel has a big lead in IPC and frequency, in 15W more than 95W+. So you have to wait, personally I don't think Raven Ridge will be good enough for CNL 2+2.

CNL 2+2 doesnt have a chance in MT and iGPU performance against a 4C 8T + Vega iGPU Raven Ridge.

Could you give the link to your test and specs? You don't even know the EU count from CNL and didn't even mention that CNL comes with a new Genx version.

 

[KTE]

Pointless, because it doesn't say anything about the competitiveness from Raven Ridge against Cannonlake. They are too many unknown variables. But one is clear, historically AMD struggled in a low power envelope even with much bigger GPUs that were much better on paper. And you can be sure Raven Ridge will be power limited if it includes a 768 SP GPU. You can also expect that Intel has a big lead in IPC and frequency, in 15W more than 95W+. So you have to wait, personally I don't think Raven Ridge will be good enough for CNL 2+2.





Could you give the link to your test and specs? You don't even know the EU count from CNL and didn't even mention that CNL comes with a new Genx version.

Might be Zacate vs i3 all over again.

Sent from HTC 10
(Opinions are own)

 

[Moumoule]

Sure. Look at FX-9800P, 4 cores with 512 shaders and is losing in both CPU and GPU against Kabylake.

Do you have any source for this ? I doubt that the HD 615/HD 620 has better performance (since there are no other informations about the iGPU of Kabylake)

 

[mikk]

Do you have any source for this ? I doubt that the HD 615/HD 620 has better performance (since there are no other informations about the iGPU of Kabylake)

Notebookcheck for example, check the Intel thread. Kabylake is easily faster. There is a healthy GPU increase with Kabylake over Skylake @15W because of a much higher GPU Turbo which is driver related and possibly to some extent because of 14nm+. Skylake might get the same boost with same drivers.

 

[majord]

notebookcheck never tested a 9800P afaik?

 

[bjt2]

Pointless, because it doesn't say anything about the competitiveness from Raven Ridge against Cannonlake. They are too many unknown variables. But one is clear, historically AMD struggled in a low power envelope even with much bigger GPUs that were much better on paper. And you can be sure Raven Ridge will be power limited if it includes a 768 SP GPU. You can also expect that Intel has a big lead in IPC and frequency, in 15W more than 95W+. So you have to wait, personally I don't think Raven Ridge will be good enough for CNL 2+2.





Could you give the link to your test and specs? You don't even know the EU count from CNL and didn't even mention that CNL comes with a new Genx version.

Isn't Samsung 14nm a low power process? Even with the shitty 28nm BULK, XV manage to compete at 15W with latest 15W ULV INTELs APU... At least in gaming benchmark... For instance: https://www.youtube.com/watch?v=p3YWXka0R-4 . In standard CPU benchmark INTEL wins by about 20-30%, but the IPC of Zen and the slightly higher clock should compensate... Only if the 15W next INTEL APUs will be quadcore they should beat also Zen. Otherwise a 2+2 can currently beat excavator, on the old 28nm vs 14nm, but Zen I don't think so...

And on low power spectrum, i bet the advantages should be more than +30% clocks... The NEON FPU test chip manages to go from 1.17GHz to 2.41GHz in the same power envelope (330mW), that is more than 100% clock... You can use this power budget to have 4 zen cores instead of 4 XV cores and 768/1024SP instead of 512 and have comparable clocks and/or produce models with less SPs (maybe from harvesting high SP models) or disable some cores from the CCX, but this lead to higher clocks in the 15W.
The biggest advantages of the 14nm are indeed in the low power spectrum...

When we are talking of high frequencies, all pinpoint that is a low power process, then when we talk of low power CPUs, all ignore that is a low power process. Let's say that AMD is doomed and close this thread...

 

[bjt2]

Notebookcheck for example, check the Intel thread. Kabylake is easily faster. There is a healthy GPU increase with Kabylake over Skylake @15W because of a much higher GPU Turbo which is driver related and possibly to some extent because of 14nm+. Skylake might get the same boost with same drivers.

Kabylake does not exist yet. In my previous post I posted a youtube video of an FX9800P beating a 6500U in a game benchmark.

Zen APUs should be better...

 

[KTE]

Notebookcheck for example, check the Intel thread. Kabylake is easily faster. There is a healthy GPU increase with Kabylake over Skylake @15W because of a much higher GPU Turbo which is driver related and possibly to some extent because of 14nm+. Skylake might get the same boost with same drivers.

Problem is this.

Whenever AMD/Intel has a decent chip, they boast about it with early showings. It's part of marketing.

Look at what AMD did with Ontario/Zacate in 2010:
http://www.anandtech.com/show/3920/amd-benchmarks-zacate-apu-2x-faster-gpu-performance-than-core-i5
http://www.anandtech.com/show/3933/amds-zacate-apu-performance-update

They demo'd and gave full access to it a few times in September, 4 months before the late January launch. They even let the press run new benchmarks they've never ran!

It's late November now, and we're still not getting anything on Zen from AMD.

Here's hoping this product is not delayed to March.

Sent from HTC 10
(Opinions are own)

 

[KTE]

And on low power spectrum, i bet the advantages should be more than +30% clocks... The NEON FPU test chip manages to go from 1.17GHz to 2.41GHz in the same power envelope (330mW), that is more than 100% clock...

How much are you willing to bet that clocks are NO chance +30% XV at the same power?

Sent from HTC 10
(Opinions are own)

 

[Sven_eng]

Might be Zacate vs i3 all over again.

Sent from HTC 10
(Opinions are own)

50 million AMD Brazos sales vs how many hundred-thousand cherry-picked low power i3's?

 

[piesquared]

Except that Opteron came before A64 by a good six months or so. This time around though AMD has said desktop will come first.

Are you sure? I could very well be wrong on that. Makes sense since socket 940 for the first desktop A64 was the same socket as Opteron.

 

[bjt2]

How much are you willing to bet that clocks are NO chance +30% XV at the same power?

Sent from HTC 10
(Opinions are own)

On high clock, yes, you are right. Base and maximum clocks probabily will gain only 100-300MHz (from 3.6/4.2 to e.g. 3.9/4.5 maximum) in the 15W range.
I was talking of low clocks. Here the A12-9800 was shown to go lower than the 3.6GHz default, by others, probabily due to the cTDP limited at 12W AND/OR skin temperature going over 45C...
With 14nm the mean clock could be way higher. I foresee 30% gain, from the 2.9GHz shown here, namely 3.7-3.8Ghz, substantially losing only a few MHz.
This because the 14nm LPP is low power and low leakage and the mean off transistors in Jaguar were 90% all the time. With a so low leakage, the CPU will draw 15W only at full load and/or full turbo.

 

[KTE]

On high clock, yes, you are right. Base and maximum clocks probabily will gain only 100-300MHz (from 3.6/4.2 to e.g. 3.9/4.5 maximum) in the 15W range.
I was talking of low clocks. Here the A12-9800 was shown to go lower than the 3.6GHz default, by others, probabily due to the cTDP limited at 12W AND/OR skin temperature going over 45C...
With 14nm the mean clock could be way higher. I foresee 30% gain, from the 2.9GHz shown here, namely 3.7-3.8Ghz, substantially losing only a few MHz.
This because the 14nm LPP is low power and low leakage and the mean off transistors in Jaguar were 90% all the time. With a so low leakage, the CPU will draw 15W only at full load and/or full turbo.

I can agree with that logic

Sent from HTC 10
(Opinions are own)

 

[bjt2]

I can agree with that logic

Sent from HTC 10
(Opinions are own)

Excuse me... I was wrong with the clocks and/or the TDPs... I swapped the 65W and 15W clocks... The reasoning is the same though... For both 15W, 65W and 95W parts... But with different numbers...

 

[KTE]

Excuse me... I was wrong with the clocks and/or the TDPs... I swapped the 65W and 15W clocks... The reasoning is the same though... For both 15W, 65W and 95W parts... But with different numbers...

I know you did... But that's beside the point, right

Sent from HTC 10
(Opinions are own)

 

[AtenRa]

Pointless, because it doesn't say anything about the competitiveness from Raven Ridge against Cannonlake.

Whats pointless to the RavenRidge discussion is your commend about BristolRidge 28nm Planar to Skylake 14nm FF.

I don't think Raven Ridge will be good enough for CNL 2+2.

Could you give the link to your test and specs? You don't even know the EU count from CNL and didn't even mention that CNL comes with a new Genx version.

So you actually believe that 4C 8T Raven Ridge is not going to be good for CNL 2+2 but you ask me for tests and specs in the very same post ???

Anyway I dont want to continue this here.

 

[sirmo]

Are you sure? I could very well be wrong on that. Makes sense since socket 940 for the first desktop A64 was the same socket as Opteron.

He's right, Opteron (Sledgehammer) was months ahead of Athlon64 (Clawhammer).. I remember buying the former, because I couldn't wait for A64.

AMD was targeting server first.

 

[AtenRa]

Well, they were pretty much competitive in Desktop back then so they opted for the Server fat margins first with Sledgehammer. Today they cannot wait for the Server parts to be ready and they will release Desktop first because it doesnt need the same evaluation as the server parts.

 

[Shivansps]

50 million AMD Brazos sales vs how many hundred-thousand cherry-picked low power i3's?

I remember back in the day that i had to "fight" "people" who argued that E-450 was better in games than an I3 with HD3000...

Anyway, Brazos was a piece of crap and it sold a lot because OEM used it EVERYWHERE, from netbooks to 17" notebooks. It was better than Atoms but that was it, and OEMs placed where they never meant to be.