[News/Rumor] Benchmarks for AMD's Zen ES CPUs have been leaked

[StyM]

http://www.overclock3d.net/news/cpu_..._been_leaked/1

Performance data from AMD's Zen Engineering samples have been leaked online, showcasing some huge performance gains over their previous architectures, even when running at much lower clock speeds.

With these early benchmarks, several factors need to be considered, first that these come from early engineering samples and do not run at the same clock speeds as final products and secondly that these CPUs are running on motherboards with pre-release BIOS versions.

Looking at the data from WCCFTech below we can see that these leaked benchmarks show a 38% performance improvement with AMD's 8-core Zen engineering sample over AMD's older FX 8350. This is a significant performance gain given the fact that the Zen Engineering sample has CPU clock speed of 2.8-3.2GHz, and the FX 8350 has clock speeds of between 4.0 and 4.2GHz.

When compared to Intel GPUs on the same benchmark and GPU configuration the Zen CPU sample outperforms Intel's i5 4670K with ease, but performs behind Intel's i7 4790, though it is worth noting that AMD's Zen based CPU runs at much lower clock speeds.

[AliG]

I hope these true. If they are, that's very, very encouraging.

Clockspeeds can be raised as the architecture and manufacturing processes mature. As long as AMD is competitive in IPC (in this case actually beating Haswell) they can be a big player in the server world.

[Aleki]

Are all eight cores being utilized in this benchmark? If not, welcome back AMD!

[AliG]

Are all eight cores being utilized in this benchmark? If not, welcome back AMD!

I don't think this benchmark scales very well beyond 4 threads. Note the difference between i5-4670 and i7-4790. Clearly the 20% increase can't be explained by clockspeed alone, but it also isn't nearly around 60% (which is what you would expect for full hyperthreading scaling).

Or I can flip the math to try to extrapolate the effects of scaling:
4790 is 5% faster, but performed 23% better. Let's assume that means that core scaling is around 18%.
Zen performed 10% faster than 4670, but is clocked 18% lower.

If we kinda play in the grey area, that would imply that the IPC is somewhere around 5-10% better than Haswell's. Again, I want to be clear this is not true engineering, just some hazy math.

[CrimInalA]

If you want to show cpu speed , don't do it by running a game which is gpu bottlenecked .
Run a cpu benchmark for crying out loud .

[Andi64]

I hope these true. If they are, that's very, very encouraging.

Clockspeeds can be raised as the architecture and manufacturing processes mature. As long as AMD is competitive in IPC (in this case actually beating Haswell) they can be a big player in the server world.

It isn't beating Haswell. It's an 8 core 16 threads processor losing against a Haswell 4 core 8 threads CPU.

Comparing both intel CPUs shows that this test likes threads over clocks. There is only 200Mhz difference between the two (5%) but gains over 24% performance improvement, just by running HyperThreading...

[Sparky]

I withhold judgement based on early A0 silicon running only 1 test.

[Andi64]

I withhold judgement based on early A0 silicon running only 1 test.

I agree. I won't conclude anything. Just analysing the available data.

[iddqd]

Welp, somebody's getting fired for this.

[AliG]

It isn't beating Haswell. It's an 8 core 16 threads processor losing against a Haswell 4 core 8 threads CPU.

Comparing both intel CPUs shows that this test likes threads over clocks. There is only 200Mhz difference between the two (5%) but gains over 24% performance improvement, just by running HyperThreading...

Did you see my math? The Intel CPUs gained only 18% for doubling the thread count - that is not good core scaling.

[iddqd]

Ashes is a really wonky CPU benchmark - sometimes, things like cheap locked Haswell i5s are beating 5960X

[zanzabar]

Ashes is a really wonky CPU benchmark - sometimes, things like cheap locked Haswell i5s are beating 5960X

it only cares about latency and for some strange reason also cares about memory and pci-e controller latency/bandwidth.

[drmrlordx]

Look at that lead over Vishera . . . it's tempting to extrapolate the minimum gains possible, but without knowing core utilization and turbo speeds it's hard to take away anything truly meaningful from this benchmark.

Even if Summit Ridge was running the bench @ 3.2 GHz while Vishera was at 4.0 GHz, assuming 8 threads or less were utilized and assuming the Vishera "module penalty" was not in full effect, that would give Summit Ridge a 72.6% lead over Vishera in this game. 72.6%! That's big, especially in a scenario where GPU limits on speed might be a factor.

[Andi64]

Did you see my math? The Intel CPUs gained only 18% for doubling the thread count - that is not good core scaling.

It actually is good core scaling. That "double" thread count is just Hyperthreading. It gained far more performance from Hyperthreading than from being 200Mhz faster than the i5.

So, the game is actually using those threads. It is pretty much obvious that the game can use more than 4 threads.

Lets do an excercise, supose the game engine can only use 8 threads simultaneously, and draw conclusions from there.

In this particular case, which is ONE game engine, ONE test, (without taking into account core usage, the fact that it's an ES, etc) you have:

Core i7 4790 8T @ 3.6/4.0: 65.4 FPS
AMD Zen ES: 8T @ 3.2: 58 FPS

And there you have it, an 8 cores CPU losing against an HyperThreaded 4C CPU.


If you want to look at it as "finally AMD can beat a Core i5", sure, be my guest. In this particular test, it sure can. It's a freaking Core i5, ffs. It's an 8 core 16 threads beast!

[AliG]

Actually hyperthreading scales very well. And unlike your hand waiving - I have proof!

[qcmadness]

It actually is good core scaling. That "double" thread count is just Hyperthreading. It gained far more performance from Hyperthreading than from being 200Mhz faster than the i5.

So, the game is actually using those threads. It is pretty much obvious that the game can use more than 4 threads.

Lets do an excercise, supose the game engine can only use 8 threads simultaneously, and draw conclusions from there.

In this particular case, which is ONE game engine, ONE test, (without taking into account core usage, the fact that it's an ES, etc) you have:

Core i7 4790 8T @ 3.6/4.0: 65.4 FPS
AMD Zen ES: 8T @ 3.2: 58 FPS

And there you have it, an 8 cores CPU losing against an HyperThreaded 4C CPU.


If you want to look at it as "finally AMD can beat a Core i5", sure, be my guest. In this particular test, it sure can. It's a freaking Core i5, ffs. It's an 8 core 16 threads beast!

You won't know if 1 thread of each core are under test or 2 threads of the 4 cores are under test. These two scenarios are different.

[Sparky]

I think it is just funny that anyone is trying to draw much of a conclusion of the CPU based off of so little information on early test runs lol.

[Metroid]

Pure speculation, its funny AMD testers let it leak, funny it looks just like what they did to rx 480, although 99% of us want it to be real and if it is then we will have more choices to choose from, the mainstream high end market will welcome the competition which as it stands its laughable. We have to give AMD some points cause if this processor is indeed to compete x i7 6700 or i7 7700 and if is 8 cores/16 threads then we will have a winner here if price is right. There is no point AMD launching an 8 cores 16 threads with prices similar to 6800k. Now if they manage to price it very close to 6700 or 7700 then AMD has a winner and that i doubt it will be the case cause AMD messed up on the rx 470 and rx 480 prices. So in the end AMD will likely say --> buy Intel and Nvidia and forget us cause we want your money and we hype so much before of things that we don't deliver what we promised.

Expectation --> 8 cores / 16 threads, priced at $349 to x I7 6700, i7 7700.
True story --> 8 cores / 16 threads, priced at $699 to x 6800k.

[Andi64]

I dont want an 8 core 16 thread CPU to be priced against a 4 core 8 threads CPU. If they are going to release 8 core CPUs I want them to be priced against 8 core Intel CPUs.
AMD is not stupid, and even thou they used to be the price/performance winners, they won't sell you a CPU that can match a Core i7 6900K for less than what a 6900K cost.

So, if you are expecting Zen to be priced againt an i7 6700K, it surely means that it can only beat a 6700K. That would mean that performance of an 8 core AMD CPU will only beat a 4 core Intel.

I want AMD to compete against Intels best... 6900/6950, that is what I expect. If that does not happen... Intel will continue to be unchallenged on upper mid-range and high-end products.

I really hope Zen succeds, and I'd love to go back to AMD CPUs... I'm not jumping to any conclusions based on this test, but I really don't get why would anyone call this a good trend nor "very encouraging". It looks like crap on this test, and I don't think anyone wants AMD to fail once again.

[zanzabar]

i expect that they will have to sell for less than intel does, they also are not including an IGPU in the 8 core parts so that should keep the cost down. maybe they will be selling them for 6800k money, but no way they are getting $800+ for any 1p chip.

[vario]

First, x264 HT scaling is pretty much best case scenario.Mostly it gives not that much or even nil, also 4790K has more cache+clocks, so all in all, While ashes do scale with cores, most probably it aint much .

Second, thing we should really look at is FX.
Huge clock disparity, and still comes strong on top.This bodes well for the architecture.As for clocks its another thing.
It may just be ES silicon being ran at the frequency they like (2,8 is somewhat standard on those parts) , OR they may have troubles to get much more.First signs about 14nm process at fishkill doesnt look good, but that was another AMD division and CPU guys will have much more time to tweak the process.
All in all, if this results are true (and they look legit) it does look good.

[drmrlordx]

Actually hyperthreading scales very well. And unlike your hand waiving - I have proof!

Um . . . about that . . .

First, x264 HT scaling is pretty much best case scenario.

Exactly. Let's look at core/HT scaling for AotS instead of x264:

https://forums.anandtech.com/threads...#post-38419022

[AliG]

You missed the entire point.... I said that AotS DOES NOT scale well beyond 4 threads.

[zanzabar]

Expectation --> 8 cores / 16 threads, priced at $349 to x I7 6700, i7 7700.
True story --> 8 cores / 16 threads, priced at $699 to x 6800k.

the 6800k is $440
http://ark.intel.com/products/94189/...up-to-3_60-GHz

[Metroid]

the 6800k is $440
http://ark.intel.com/products/94189/...up-to-3_60-GHz

6800k --> 6/12, whereas amd if 8/16 --> $699. I don't think AMD if performance is given, will sell their 8/16 for 6800k. I do hope so AMD gives more for less but that is just a hopeful wish, we all know that will not likely to happen.