[UPDATE: Almost Certaintly Fake] AMD Zen branding, SKUs, and pricing

[Abwx]

Look at their TDP statements,consider their -similar systems- statement when they ran the blender test at the presentation and one can be fairly sure that the only way to get broadwell-e levels of performance is with a broadwell-e 8core limited at 65wtdp...

They said that in Blender and with both CPUs at 3GHz Zen consume "a little less than BDW"....

I have no doubt that it consume even less in respect of BDW than what AMD stated as Intel s HEDT are factory undervolted as proved by Computerbase.de review, there s no way such a plateform that is supposed to be high end is using a CPU whose voltage margin is below mainstream CPUs.

In principle high end is not only about perfs but also about extremely robust stability, just look at the frequencies of server parts, they had to be reduced to make room for the higher voltage margins..

 

[Arachnotronic]

Yes, i m aware of how it work, thing is that to have the higher possible ST perf the faster core should be loaded by the application but at the same time the main thread require the faster core, overall it s a mixed bag and Hardware.fr had to disable this "feature" to have higher perfs in their tests.

Turbo Boost Max is a weird "feature." Just overclock the dang processor.

 

[Abwx]

Turbo Boost Max is a weird "feature." Just overclock the dang processor.

Indeed, and i m not sure that there s a single user who use thoses CPUs at stock settings, user settings will also solve the voltage margin i mentioned above for professional apps that forcibly mandate perfect stability.

Btw, there s some firms that use scores of overclocked CPUs to get the job done faster, FI simulation of the analog behaviour of a complexe circuitry is extremely slow and require a lot of computing power to simulate the circuit at its stock frequency.

 

[DrMrLordX]

Let the RX 480 be your guide. How did AMD price this relative to the competition?

MSRP aside, real-world pricing on Polaris has been a bit wonky since it has been such a darling of the miner crowd. They bought a lot of those cards for ETH and ZEC mining.

 

[The Stilt]

Funny enough ZEN's selling feature seems to be just that.
Look at their TDP statements,consider their -similar systems- statement when they ran the blender test at the presentation and one can be fairly sure that the only way to get broadwell-e levels of performance is with a broadwell-e 8core limited at 65wtdp...
40% over XV is still ~20% lower than haswell,clock to clock and 3,2Ghz base is not really high.
http://wccftech.com/amd-zen-8-core-35ghz-cpu-spotted/

These slides are for Bristol / Stoney.

 

[Abwx]

These slides are for Bristol / Stoney.

They somewhat apply to Zen as well :

 

[AtenRa]

They somewhat apply to Zen as well :

Hell this is so confusing, looking this slide again it seems to me that ZEN could be 40% faster per Single Thread at the same 65W TDP vs BristolRidge (Excavator).

 

[coercitiv]

Hell this is so confusing, looking this slide again it seems to me that ZEN could be 40% faster per Single Thread at the same 65W TDP vs BristolRidge (Excavator).

Why would you mention chip TDP when ST performance is involved?

 

[whm1974]

Why would you mention chip TDP when ST performance is involved?

Maybe he wants better ST performance and more cores with a 65W TDP so he can stick it in a MinITX case like the Silverstone Milo ML08B:
http://www.newegg.com/Product/Product.aspx?Item=N82E16811163286
I'm sure a lot people if AMD releases an 8 core Zen CPU with a 65w TDP would consider putting it in a small footprint case.

 

[Qwertilot]

Depends how fast it is - there has to be a chance that 65w/8 cores means relatively poor single threaded performance. It could easily enough be non trivially behind Intels 45W quads for instance.

 

[AtenRa]

Why would you mention chip TDP when ST performance is involved?

Because 65W TDP is the Excavator SKUs and according to the slide above, ZEN will have 40% higher "Instructions per core" at the same energy per cycle vs Excavator.

Since the latest leaks talk about 65W TDP ZEN SKUs, im only trying to connect the dots

 

[coercitiv]

Because 65W TDP is the Excavator SKUs and according to the slide above, ZEN will have 40% higher "Instructions per core" at the same energy per cycle vs Excavator.

Since the latest leaks talk about 65W TDP ZEN SKUs, im only trying to connect the dots

And again I ask, why bring SKU TDP into a slide where only arch efficiency for ST loads is being discussed. TDP figures in modern CPUs are MT dependent, except maybe for very low power mobile cases. Unless that slide discusses core throughput, hence 2 threads and CMT/SMT is included, connecting the dots still requires guessing SMT efficiency.

 

[krumme]

As far as the market share, that huge unit share gain happened pre-Polaris and was obviously more related to cheaper OEM discrete GPUs than gaming focused products. Post-Polaris, NVIDIA actually gained gaming GPU revenue share.

Naa we know now this is not the precise way to frame it imo. Lisa said Asp for amd discrete gpu went up q3.
And the gpu marketshare nv won here was minicule and more like flat.
Considering initial gf 14nm woes and nv broad high quality top bottom portfolio clearly the result in share is surprising especially in connection with amd increasing asp. I am pretty sure both of use wouldnt forecast that development looking at the portfolio.
But i guess the aprox 200usd market is just huge right now.

 

[TheELF]

And again I ask, why bring SKU TDP into a slide where only arch efficiency for ST loads is being discussed.

They are not talking about ST loads specifically,the slide says per cycle/per clock not per core.
Doesn't matter anyway,what they say is that the same amount of cores(whatever amd means by cores) will produce 40% more work at the specific TDP * they are stating,those 65w,careful they don't say at any power level but are very specific on the 65w (previous slides) .

*or as they state on those previous slides, 65w delivers the performance of previous 95w

 

[LTC8K6]

Yes, it says "+40% work per cycle", which could mean far less than a 40% ipc improvement in single thread.

 

[inf64]

Yes, it says "+40% work per cycle", which could mean far less than a 40% ipc improvement in single thread.

Come on, we have had slides from Hot Chips that in no uncertain terms claim exactly 40% IPC improvement over previous core(XV). Then at the same event in Q&A session AMD fellow confirmed that 40% IPC increase is in single thread workloads and that "SMT comes on top of that". Can we stop putting a question mark after these claims already? Whether they are true or not remains to be seen, but trying to muddy the waters regarding these claims while we have official slides and words from AMD engineers (with witnesses around) is getting silly now.

 

[LTC8K6]

Come on, we have had slides from Hot Chips that in no uncertain terms claim exactly 40% IPC improvement over previous core(XV). Then at the same event in Q&A session AMD fellow confirmed that 40% IPC increase is in single thread workloads and that "SMT comes on top of that". Can we stop putting a question mark after these claims already? Whether they are true or not remains to be seen, but trying to muddy the waters regarding these claims while we have official slides and words from AMD engineers (with witnesses around) is getting silly now.

I didn't make the slide...I just quoted it and gave an opinion. I don't have any effect on AMD sales as far as I know.
I don't understand why it says what it says if the gain is a simple 40% IPC improvement over Excavator.
I also don't understand "illustrative purposes only". I think that means the drawings are probably exaggerated compared to reality? The 40% ramp isn't really quite that tall, perhaps?
Just leak some solid results already, so that buyers can begin to make decisions and don't purchase the competition while waiting for Zen.

 

[inf64]

Wow, I won't even dignify that with an answer. Sorry.

 

[Abwx]

Hell this is so confusing, looking this slide again it seems to me that ZEN could be 40% faster per Single Thread at the same 65W TDP vs BristolRidge (Excavator).

That s it, a 4C Zen consume the same as a 4C XV on a clock/clock basis, that s for the cores but the uncore will consume less, so if BR is 65W at 3.5GHz Zen should be a little below this TDP at this frequency.

Yes, it says "+40% work per cycle", which could mean far less than a 40% ipc improvement in single thread.

Lol....

 

[whm1974]

Come on, we have had slides from Hot Chips that in no uncertain terms claim exactly 40% IPC improvement over previous core(XV). Then at the same event in Q&A session AMD fellow confirmed that 40% IPC increase is in single thread workloads and that "SMT comes on top of that". Can we stop putting a question mark after these claims already? Whether they are true or not remains to be seen, but trying to muddy the waters regarding these claims while we have official slides and words from AMD engineers (with witnesses around) is getting silly now.

Considering that AMD is treading water right now, a 40% improvement in IPC and double the threads is no small feat.

 

[Abwx]

And again I ask, why bring SKU TDP into a slide where only arch efficiency for ST loads is being discussed. TDP figures in modern CPUs are MT dependent, except maybe for very low power mobile cases. Unless that slide discusses core throughput, hence 2 threads and CMT/SMT is included, connecting the dots still requires guessing SMT efficiency.

You dont measure the work/cycle with several threads, the principle is to send one thread and measure how much more instructions are executed each cycle for this given piece of code, throughput is another matter, for this latter i posted the peak numbers in another thread..

 

[TheELF]

Yes, it says "+40% work per cycle", which could mean far less than a 40% ipc improvement in single thread.

Actually it's the other way around,CPUs loose a lot of IPC in MT scenarios,just look at CB's MP ratio for the FX-8xxx,ZEN of course will be much better hopefully.
The main thing is that this 40% + might only work at 65W and drop a lot at higher TDPs/clocks.

 

[coercitiv]

I don't think that people who say this +40% could be for MT realize that such a scenario would imply even stronger ST improvements, making Zen a warp engine among rocket propellers.

 

[Arachnotronic]

I don't think that people who say this +40% could be for MT realize that such a scenario would imply even stronger ST improvements, making Zen a warp engine among rocket propellers.

Oh hey, you're right. +40% perf/MHz for 1 core/2 threads versus +40% perf/MHz for 2 core/2 thread would imply an even bigger boost for that 1 core

 

[TheELF]

I don't think that people who say this +40% could be for MT realize that such a scenario would imply even stronger ST improvements, making Zen a warp engine among rocket propellers.

Only if Zscaling is working like Hyperthreading but there is nothing being scaled in HT.
Also as inf64 already said,amd confirmed that 40% IPC increase is without SMT.
So if ZEN is at 3,2 that would equal a XV core at ~4,5Ghz and those cores at ST are a real far cry away from haswell,hardly warp engine material.