Leaked AMD Zen Engineering Sample Benchmarks?

H

HardOCP News

[H] News
Joined
Dec 31, 1969
Messages
0
The crew at WCCFTech have stumbled upon what appear to be AMD Zen Engineering Sample benchmarks from the Ashes of the Singularity benchmark database. The numbers have since been pulled but not before screenshots were taken.

This is the first indication of real world performance which does not originate from AMD’s marketing material. That said however, the benchmarks given in this article are that of the Engineering Sample of AMD Zen and NOT entirely representative of the final product. As such, the benchmarks only serve to give us a preliminary understanding of what we can expect from the final product.
 
AMD_Zen_ES_Aot_S_Benchmarks.jpg

Even with a base clock of 2.8 Ghz, we can see that it manages to beat out the Intel Core i5 4670k which has a base clock of 3.4 Ghz. It is a full 38% faster than the AMD FX-8350 and around 10% faster than the i5 4670k. However, in its current state, and in this particular benchmark, the Zen ES 1D variant is around 11% slower than the Core i7 4790 (which has a base clock of 3.6 Ghz).
Click to expand...

Very interesting, if they can clock this thing a little faster and price it right it could be a success.
 
I've always thought that is where AMD came in with big wins in the past. More processing power with lower clock speeds. But, that was a long time ago. I hope they have some architectural breakthroughs here.
 
I am getting sick of waiting for Zen. The current AMD line up is well past its prime.
 
Thank god, my 3570k was the most overpriced pos hardware I've bought. If i was half the price back then, then I wouldn't be annoyed with its lack of multitasking oomph. $300 on amazon.ca is current price for anyone won
 
Really hoping that Zen is as good as current intel cpu's. We need competition badly and they need that money even more.
 
Looks good. With higher clock speed it would probably be on par with that Haswell i7 or beat it, all while running a similar or slightly slower clock speed.
This could be AMD's Conroe.
 
I'm not gonna hold my breath on this one. I feel like the GloFlo FinFET yields are still really low and they probably need a whole nother year to mature their process.
 
If you wanted to show cpu advancement why on earth would you do it showing it on a gpu bound API ?
 
But this is an 8 Core CPU with 16 threads? Does ashes scale? 8 Core/16 Thread beating a 4 Core 4 Thread CPU isn't a big accomplishment.... Yes its early, but that's not that promising.
 
If this info is true when they release the final product it will be a huge competitor to intel and hopefully help drive down some of the prices.
 
azuza001 said:
Hope it's true, expect it to be as good as an i5. This way I can't be disappointed.
Click to expand...

Even modest expectations can lead to disappointment. I was hoping for an RX 480 that was between the 390x and Fury in performance.
 
By no means an I an AMD fan, never have been but I hope that this is a real winner for those who are. This needs to be a major success to help AMD stay alive and the industry badly needs the competition to drive down prices and improve products.
 
I really hope this is a good CPU. Intel is just becoming way too lax and overcharging now that they have like no competition on the desktop IMHO.
 
Just to temper expectations, this chip likely will be on par with Intel's Haswell, not skylake, but support modern architecture like M.2 and USB C, and a fraction of the cost of Intel. Given the modest improvements to Skylake over intel, and the rumored further diminishing returns of Kaby Lake, this might actually end up being a net win for AMD, particularly if they overclock well.
 
90-95% of Intel's performance at 80% of the price will suffice to get them back into the game and give Intel a kick in the pants. Anyone expecting something significantly faster than Intel's offerings should re-examine their reasoning. Intel has some of the smartest and highest paid engineers on the planet while AMD is bleeding all of it's talent. With that context this will be huge win for AMD if true. Unfortunately profitability doesn't always take context into consideration...
 
Booyaah said:
I'm not gonna hold my breath on this one. I feel like the GloFlo FinFET yields are still really low and they probably need a whole nother year to mature their process.
Click to expand...

The yeilds will increase as will the optimal clock speed range of Samsung's FinFET process. IBM is using it in the upcoming POWER9 CPUs, Which expect to have single cores clocking to 5 Ghz when called for. There's a ton of improvements getting contributed to 14 nM finFET we will benefit from.
 
Zion Halcyon said:
Just to temper expectations, this chip likely will be on par with Intel's Haswell, not skylake, but support modern architecture like M.2 and USB C, and a fraction of the cost of Intel. Given the modest improvements to Skylake over intel, and the rumored further diminishing returns of Kaby Lake, this might actually end up being a net win for AMD, particularly if they overclock well.
Click to expand...

You're not really tempering expectations with unsubstantiated claims like that. Zen will launch in 8-core version. There is no word if they have a native quad core version launch simultaneously, or if that will wait until the APU is available later next year. So I would expect mostly 8 and cut 6 core versions, with a low-availability of cut 4-core part. Those could be quite expensive, depending on how the performance ends-up.

These are big chips, so AMD will price them to make a profit IF THE PERFORMANCE IS THERE.
 
Last edited:
Gman1979 said:
The yeilds will increase as will the optimal clock speed range of Samsung's FinFET process. IBM is using it in the upcoming POWER9 CPUs, Which expect to have single cores clocking to 5 Ghz when called for. There's a ton of improvements getting contributed to 14 nM finFET we will benefit from.
Click to expand...

Power9 uses a special Glofo(IBM designed) 14HP Finfet process that got nothing to do with Samsung and comes from the IBM foundry part. But it got extremely low yields and high cost. Zen and others uses 14LPP developed by Samsung.
 
90-95% of Intel's performance at 80% of the price will suffice to get them back into the game and give Intel a kick in the pants.
Click to expand...

I say AMD fans will revolt if that is the price.. I mean an Intel 8 core / 16 threaded CPU is around 1 thousand dollars US. Although I hope / expect the 8 core / 16 threaded Zen CPU to be competitive against the lowest end Intel 6 core / 12 threaded CPU and to be priced to compete with the AMD discount. So I am now expecting between 275 and 350 US.
 
Should Zen be compared against Skylake/Kaby Lake or Broadwell? Based solely on the number of cores/threads, Broadwell should be its direct competitor, right?
 
8 cores with 16 threads of Haswell-like performance will be enough for me to finally leave my x58 setup behind. X99 is just too much of a clusterfuck.
 
I am thinking it will not be enough of an upgrade for me. I currently want to replace my 6 core / 12 threaded i7 970 with the 16 core / 32 threaded Zen on the server platform. Even if it is a $1500 CPU. I mean provided the IPC is where AMD said it would be..
 
Monkey God said:
To minimize how bad your CPU's performance looks.
Click to expand...
That makes even less sense then running the Ashes benchmark. In the end people will have to know about performance otherwise there would be no incentive to buy the cpu.
 
drescherjm said:
I am thinking it will not be enough of an upgrade for me. I currently want to replace my 6 core / 12 threaded i7 970 with the 16 core / 32 threaded Zen on the server platform. Even if it is a $1500 CPU. I mean provided the IPC is where AMD said it would be..
Click to expand...
Well, if AMD kicks out a 16 core part, I'll certainly look at that as well.
 
Zion Halcyon said:
Just to temper expectations, this chip likely will be on par with Intel's Haswell, not skylake, but support modern architecture like M.2 and USB C, and a fraction of the cost of Intel. Given the modest improvements to Skylake over intel, and the rumored further diminishing returns of Kaby Lake, this might actually end up being a net win for AMD, particularly if they overclock well.
Click to expand...
There's hardly a difference between Haswell and Skylake. You don't see Haswell K part owners going out to buy a Skylake CPU for like 3% difference.

Most of us would be happy if AMD could match Haswell but offer things like hyperthreading with lower costs. USB C and M.2 are nice but we really need them to match Haswell performance. I don't expect ZENs to overclock well cause looking at Polaris cards 14nm FinFET doesn't show much promise in this area.
 
So Haswell level performance, as expected. The clock still being low despite this almost certainly being a later stepping is also an indication that GlFo's LPP node is holding back clocks. I think this is going to be fairly close to final performance levels, which is *meh*. Would have been more competitive two years or so ago.

I think we'll have the same dynamic we do in GPU land: AMD going more for the upper-mid tier performance crown, but with Intel clearly on to performance wise.

EDIT

Look at it this way, AMD doubled the number of cores, but only squeezed out a 32% performance gain. Work the math, and IPC in AoS actually DROPS compared to the FX-8350.

One one hand, this could be due to better scaling. On the other, FPS still lags Intel DESPITE that better scaling.

Honestly, this is not that impressive, and if GF's LPP node can't ramp up clocks [which I suspect is the case, given the node is optimized for <3GHz parts], then per-core performance is going to lag Intel by a significant amount. Even in AoS, which scales well and should be a *best case* for AMD, Zen's per core performance is 57% less then Intel's.

Right now, this is very disappointing for AMD. Even moreso then I initially thought.
 
Last edited:
Shintai said:
Power9 uses a special Glofo(IBM designed) 14HP Finfet process that got nothing to do with Samsung and comes from the IBM foundry part. But it got extremely low yields and high cost. Zen and others uses 14LPP developed by Samsung.
Click to expand...

`IBM signed on with Samsung to help with their process problems. IBM gave up pushing their own process development, instead putting their hat in with GlobolFoundries, AMD, and Samsung to move forward using an evolvolving form of Samsung's finFET process. It would have made production costs higher than needed to support an identical node in the same factory with no obvious advantages. The earlier sales materials stated they'd be using IBM's process, but the latest I've seen refer to using a process codeveloped with Samsung.


EDIT: I can't find a public source I can post to back this up as it's stuff from work. It was also on a marketing memo so it's not infalable. My understanding though is the partnership between Samsung, GlobalFabs, and IBM involves them sharing development effort and costs on process development. Continual process refinements are suppose to be fed back into the development. I just hope we gain the benefits of that as production goes on. It sounds like IBM is concentrating hard on 10 nM as well. They've been talking up developments with it quite often.


TLDR: I think it's probably a hybrid Samsung/IBM 14 Nm finFET effort. Samsung had good yields, IBM had a higher speed capable process with yield issues. It would make sense to share some of the better points of each process refinement.
 
Last edited:
gamerk2 said:
So Haswell level performance, as expected. The clock still being low despite this almost certainly being a later stepping is also an indication that GlFo's LPP node is holding back clocks. I think this is going to be fairly close to final performance levels, which is *meh*. Would have been more competitive two years or so ago.
I think we'll have the same dynamic we do in GPU land: AMD going more for the upper-mid tier performance crown, but with Intel clearly on to performance wise.
EDIT
Look at it this way, AMD doubled the number of cores, but only squeezed out a 32% performance gain. Work the math, and IPC in AoS actually DROPS compared to the FX-8350.
One one hand, this could be due to better scaling. On the other, FPS still lags Intel DESPITE that better scaling.
Honestly, this is not that impressive, and if GF's LPP node can't ramp up clocks [which I suspect is the case, given the node is optimized for <3GHz parts], then per-core performance is going to lag Intel by a significant amount. Even in AoS, which scales well and should be a *best case* for AMD, Zen's per core performance is 57% less then Intel's.
Right now, this is very disappointing for AMD. Even moreso then I initially thought.
Click to expand...

Yeah the manufacturing process and the "problems" never tend to favour AMD. Then again AMD does not have much money to spend if x86 design was easy there would be a number of players beside Intel and also for that reason it tends to make anything AMD does look really bad ....
 
DukenukemX said:
There's hardly a difference between Haswell and Skylake. You don't see Haswell K part owners going out to buy a Skylake CPU for like 3% difference.

Most of us would be happy if AMD could match Haswell but offer things like hyperthreading with lower costs. USB C and M.2 are nice but we really need them to match Haswell performance. I don't expect ZENs to overclock well cause looking at Polaris cards 14nm FinFET doesn't show much promise in this area.
Click to expand...

can you prove Skylake is 3% faster than haswell?.. =) I would like to see that both in high-level multi-thread apps and specially gaming scenarios, thanks in advance. I'll be waiting for the proof of your claims.

but in the mean time.. I would reconsider that 3% claim with this video.

 
Zion Halcyon said:
Just to temper expectations, this chip likely will be on par with Intel's Haswell, not skylake, but support modern architecture like M.2 and USB C, and a fraction of the cost of Intel. Given the modest improvements to Skylake over intel, and the rumored further diminishing returns of Kaby Lake, this might actually end up being a net win for AMD, particularly if they overclock well.
Click to expand...

That's a huge win if it comes to pass.
 
You must log in or register to reply here.
Back
Top