• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

Leaked AMD Zen Engineering Sample Benchmarks?

razor1 said:
Yeah, who is talking piss when they can't use google?

broadwell_e_core_i7_6950x_ashes_of_the_singularity_beta_cpu_benchmark-100663335-orig.png


We can see even more scaling on a 10 core chip 20 threads.
Click to expand...
Exactly my point. If you attempted to extrapolate Intel IPC based on this, one would assume the high core parts have TERRIBLE IPC.
We all know that isn't true.
Right????
 
I expect Zen core for core will be around the same size as Intel's 14nm (within a few % points)
Click to expand...

I expect AMD to have slightly larger (physical size) core than Intel and Zen will have less transistors per core (much wider than bulldozer but still not as wide as Intel).
 
Last edited:
funkydmunky said:
Exactly my point. If you attempted to extrapolate Intel IPC based on this, one would assume the high core parts have TERRIBLE IPC.
We all know that isn't true.
Right????
Click to expand...


As the work load breaks across the cores, you aren't going to get a real look at IPC, there are many other things going on that would influence frame rates even on the CPU end outside of IPC.

But the point being, there is scaling going on even with a 10 core 20 thread chip, that benchmark pic, if correct, for Zen, is not good...., people see what they want to see, but a 8 core 16 thread chip not beating matching a 4 core 8 thread that is 2 years old, doesn't matter if its an ES as explained before, even if clocks are 10% more on retail (which qualifying ES samples that kind of gain is a large wish for retail). Hopefully its something software related.....
 
drescherjm said:
I expect AMD to have slightly larger (physical size) core than Intel and Zen will have less transistors per core (much wider than bulldozer but still not as wide as Intel).
Click to expand...

because of process differences yeah I expect that too.
 
Who knows how early that sample is as well. I am not expecting miracles but I am sure this chip is not properly optimized on the firmware side either.
 
razor1 said:
As the work load breaks across the cores, you aren't going to get a real look at IPC, there are many other things going on that would influence frame rates even on the CPU end outside of IPC.
But the point being, there is scaling going on even with a 10 core 20 thread chip, that benchmark pic, if correct, for Zen, is not good...., people see what they want to see, but a 8 core 16 thread chip not beating matching a 4 core 8 thread that is 2 years old, doesn't matter if its an ES as explained before, even if clocks are 10% more on retail (which qualifying ES samples that kind of gain is a large wish for retail). Hopefully its something software related.....
Click to expand...

The problem for AMD with cores remains the same as it was previously with many games. Unless the bulk of them start scaling like mad the 8 cores and 16 threads will seem useless to most people as you can see in some benchmarks on the other side.

Then you have a scaling issue where "people" expect it to be really faster and you get something as an example: 4 cores performance is X so 8 cores and 16 threads must be at least X*2. That will certainly not be the case in most games...
 
Last edited:
What is AMD's rationale for going to 8 core / 16 thread? The only consumer software that uses that many threads is like... HandBrake or something like that.
 
jwcalla said:
What is AMD's rationale for going to 8 core / 16 thread? The only consumer software that uses that many threads is like... HandBrake or something like that.
Click to expand...

I say to compete with Intel's 6 core / 12 threaded CPU. Also the 8 core will be used as a module on 16, 24 and 32 core server parts.


I expect to see 4 core / 8 threaded and 6 core / 12 threaded Zen CPUs in 2017 that have higher frequencies than the 8 core / 16 threaded CPU that is released this year.
 
Last edited:
Did they do a power usage comparison? I see Zen will (as expected) have a large increase in performance / watt.
 
drescherjm said:
Did they do a power usage comparison? I see Zen will (as expected) have a large increase in performance / watt.
Click to expand...

It is a speculative piece where they used some of the leaked "benchmarks" and redid the ones with the Intel cpu
 
Why would anyone benchmark anything in Ashes of the Singularity...


What a useless benchmark, especially on the CPU.

I want to see how it performs in single threaded Cinebench.
 
It's not quite the 40% IPC they promised, comparing 8-core Piledriver to 8-core Zen.

Also, since we don't know the effects of those extra threads on the game engine, the single-threaded performance per-clock could be even closer.

This is looking like a trainwreck for AMD if they can't up the clocks to at least 3.6 GHz, especially because Piledriver LAUNCHED at 4.0 GHz base clock at 125w for $200. If it's not faster than it's predecessor, why didn't AMD just update AM2+ to accept 8 cores, and release an 8-core Excavator?

This is beginning to feel like the Bulldozer launch all over again.
 
Last edited:
defaultluser said:
It's not quite the 40% IPC they promised, comparing 8-core Piledriver to 8-core Zen.

Also, since we don't know the effects of those extra threads on the game engine, the single-threaded performance per-clock could be even closer.

This is looking like a trainwreck for AMD if they can't up the clocks to at least 3.6 GHz, especially because Piledriver LAUNCHED at 4.0 GHz base clock at 125w for $200. If it's not faster than it's predecessor, why didn't AMD just update AM2+ to accept 8 cores, and release an 8-core Excavator?

This is beginning to feel like the Bulldozer launch all over again.
Click to expand...


I think it's too early to draw conclusions like this. This was an engineering sample, and may not be reflective at all of final silicon.
 
I do not see Zen as even coming close to being a train wreck. It is a new design on a new node which means it will take time to get the clock speeds up. Also, it is appearing that the ES is able to beat the FX chips well doing so at only 2.8 GHz which is definitely a step forward for AMD. Although I do I agree, I wish they had made Steamroller and Excavator 8 core variants but I think because they did not, it hurt their sales. I am certain I would have upgraded to both myself.
 
I will lower my expectations and endeavor to not be disappointed...
 
I originally predicted Ivy Bridge like performance per core, but with better overclocking potential than Ivy Bridge, and I am still sticking to that preditction.
 
Zarathustra[H] said:
I think it's too early to draw conclusions like this. This was an engineering sample, and may not be reflective at all of final silicon.
Click to expand...

This may be true. Though the current leaked benchmarks seem to indicate not much of a performance advantage over my 9590. At least not enough to make me salivate.

Still adopting a wait and see. I'll let the release happen and watch closely the real results and make decisions then. Perhaps production silicon will be clocked higher?
 
Spirit_Retro said:
This may be true. Though the current leaked benchmarks seem to indicate not much of a performance advantage over my 9590. At least not enough to make me salivate.

Still adopting a wait and see. I'll let the release happen and watch closely the real results and make decisions then. Perhaps production silicon will be clocked higher?
Click to expand...

That is almost always the case. If the top Zen part on launch isn't at least 3.8Ghz, I'll be very surprised
 
Spirit_Retro said:
Oh I agree with you. The part of me that is an enthusiast is really happy to see these numbers. The part of me that is a consumer says "wait it out".

The rest of me just wants a pizza.
Click to expand...


Pizza good, Zen pizza with 3.8 lbs of pepperoni better :)
 
I'm building my next computer with Zen regardless. Hopefully AMD can pull through.

Also, I'm not going to put any weight to a leaked benchmark of an engineering sample. I'll wait til I see 4 or 5 different games benchmarked with the actual chip from a legit site.
 
cybereality said:
I'm building my next computer with Zen regardless. Hopefully AMD can pull through.

Also, I'm not going to put any weight to a leaked benchmark of an engineering sample. I'll wait til I see 4 or 5 different games benchmarked with the actual chip from a legit site.
Click to expand...

I'll be doing the same, as long as in launch reviews it is at least as fast as my 3930k in single threaded benchmarks, and has 40+ PCIe lanes.

I'd be willing to do a sidegrade to support AMD, but not a downgrade.
 
ManofGod said:
I do not see Zen as even coming close to being a train wreck. It is a new design on a new node which means it will take time to get the clock speeds up. Also, it is appearing that the ES is able to beat the FX chips well doing so at only 2.8 GHz which is definitely a step forward for AMD. Although I do I agree, I wish they had made Steamroller and Excavator 8 core variants but I think because they did not, it hurt their sales. I am certain I would have upgraded to both myself.
Click to expand...

No it didn't. The chips were compared at the same clock speed, at 3.2 GHz.

The Zen showed about 30-35% faster performance than the FX 8370 AT THE SAME CLOCK.

Piledriver FX 8370 at 4 GHz stock is clocked 25% faster than the 3.2 GHz sample. Hence my assertion that it would have to be clocked at at least 3.6 GHz to not be a total failure, as a it's current clocks performance would be almost identical. Not very good result after all that time spent implementing the new hotness.

We'll have to wait for final clocks and more thorough tests though.
 
Last edited:
Read the bottom of the graph posted yesterday by shintai. The test conditions and clock speeds are clearly marked.
 
I think an apples to apples comparison should clear these results right away. A 5960x is also an 8c/16t part with the base clock at 3 GHz that jumps to 3.5 GHz, so just lowering the multiplier to achieve 2.8 GHz base clock is easily doable! And use the same gpu and same amount of ram. Simple, no? And let's see the result! Regardless of whatever people will say, ah is just an ES sample, ah AMD is not showing anything. AMD maybe is not showing anything intently, but the secret is out! And because we know that DX11 is not that scalable with more cores, that means automatically that performance in any DX11 based title will be worse than titles using DX12. My impression is that AMD designed Zen with server market in mind, with a heavily threaded architecture. Having a desktop part derived from this will be just something extra! AMD desperately needs the money coming from the enterprise sector and will be using desktop Zen as test platform to refine and improve the enterprise product.
 
I think an apples to apples comparison should clear these results right away. A 5960x is also an 8c/16t part with the base clock at 3 GHz that jumps to 3.5 GHz, so just lowering the multiplier to achieve 2.8 GHz base clock is easily doable! And use the same gpu and same amount of ram. Simple, no? And let's see the result!
Click to expand...

Isn't that what was done? I mean pic shows all cpus were run at 3.2 GHz. Or was turbo enabled on the 5960x and the FX 8370 and not Zen? Or the Zen was run at 2.8 + turbo and not really 3.2 GHz?
 
Last edited:
Zarathustra[H] said:
I'll be doing the same, as long as in launch reviews it is at least as fast as my 3930k in single threaded benchmarks, and has 40+ PCIe lanes.

I'd be willing to do a sidegrade to support AMD, but not a downgrade.
Click to expand...

Same story here and I can put 8 cores to use via video editing either way. Looks promising but not getting hyped up over it - will just wait and see.
 
I'd be willing to do a sidegrade to support AMD, but not a downgrade.
Click to expand...

I would not do a sidegrade to support AMD or Intel for that matter. That is why I am hoping that the 40% IPC improvement was not a lie and also hoping that the 16 core on the server platform in 2017 will exist at a frequency that I can live with..
 
drescherjm said:
Isn't that what was done? I mean pic shows all cpus were run at 3.2 GHz. Or was turbo enabled on the 5960x and the FX 8370 and not Zen?
Click to expand...
Yes, sorry, I have seen it now.

drescherjm said:
Did they do a power usage comparison? I see Zen will (as expected) have a large increase in performance / watt.
Click to expand...

Compared to what exactly? Piledriver? Haswell-E? We know that desktop Zen has a TDP of 95W and FX8370 is a 125W part, so that would be true for the comparison with Piledriver. However, when it comes to Haswell-E I see performance is almost double with just 50% more power!

defaultluser said:
It's not quite the 40% IPC they promised, comparing 8-core Piledriver to 8-core Zen.
Click to expand...

You meant of course 8 thread Piledriver to 16 thread Zen, right? Because this result shows that doubling the number of threads only resulted in just around 40% performance uplift for AMD. Of course, this is not the ideal situation, a game engine does not scale linearly with the number of threads, but when the same situation was applied to the Intel CPU, respectively doubling the number of cores in the scalability test, the performance uplift was almost 60%. So, I think AMD is moving in the right direction, but with just this test it is really difficult, if not impossible, to draw a clear conclusion about the real IPC gain.
 
drescherjm said:
Isn't that what was done? I mean pic shows all cpus were run at 3.2 GHz. Or was turbo enabled on the 5960x and the FX 8370 and not Zen? Or the Zen was run at 2.8 + turbo and not really 3.2 GHz?
Click to expand...

Supposedly Zen single thread turbo hits 3.2 ghz and the all cores turbo 3.05 ghz. That makes this rather weird we also do not know what stepping the retail version is at and how much more was left from other improvements. It won't be large improvements but can make a difference in some cases even if it seems trivial ....

What is worrying that we are already talking about it without getting anywhere on certain areas which really suited for showing performance there is no "data". Which could be as bold as programs need to be compiled again for the extra performance increase which we did not see in _any_ benchmark last time with Bulldozer ...
 
Compared to what exactly? Piledriver? Haswell-E? We know that desktop Zen has a TDP of 95W and FX8370 is a 125W part, so that would be true for the comparison with Piledriver.
Click to expand...
Compared to both of the CPUs displayed. I wanted to see actual power draw for all. I would like them to show a comparison versus Intel 6 core / 12 threaded CPUs.


However, when it comes to Haswell-E I see performance is almost double with just 50% more power!
Click to expand...

Good point. I was concentrating on the increased performance versus Piledriver at a lower power usage.
 
Last edited:
Marstg said:
You meant of course 8 thread Piledriver to 16 thread Zen, right? Because this result shows that doubling the number of threads only resulted in just around 40% performance uplift for AMD. Of course, this is not the ideal situation, a game engine does not scale linearly with the number of threads, but when the same situation was applied to the Intel CPU, respectively doubling the number of cores in the scalability test, the performance uplift was almost 60%. So, I think AMD is moving in the right direction, but with just this test it is really difficult, if not impossible, to draw a clear conclusion about the real IPC gain.
Click to expand...

Which is why I said that 35% IPC improvement was "best case" and could actually fall if the benchmark results from the previous page are accurate.

image.png


Sure looks like the game engine scales with 16 threads to me! Those are with stock clocks, so the 6700k is clocked about 30% faster than the 5960x.

But I think only about 10% of that improvement is SMT. The typical SMT speedup on Intel quqad cores is 20-25% in benchmarks, and this game engine only scales at about half the theoretical per-core performance increase. So I'm guesstimating 10%, and 25% IPC improvements.
 
Last edited:
I was referring to the graph with 5960x where he changed from 2 cores @ 3.9 no HT to 4 cores @ 3.9 no HT.
 
drescherjm said:
I would not do a sidegrade to support AMD or Intel for that matter. That is why I am hoping that the 40% IPC improvement was not a lie and also hoping that the 16 core on the server platform in 2017 will exist at a frequency that I can live with..
Click to expand...


Well, as I see it, my current 3930k has all the performance I need and then some, even at stock speeds of 3.2Ghz, let alone when overclocked to 4.8Ghz.

I don't NEED any more performance out of my CPU. It's not holding me back in anything I do. 5 years after buying the damned thing, there isn't a single game I have tested where my CPU is the bottleneck, and there aren't any applications or tasks I feel are running slow due to lack of CPU performance.

I could use some newer features though, M.2 slots, lower power consumption due to die shrinks, etc. etc. That, and my motherboard is just starting to get old, with th eplastic drying out and cracking in some places. It's time for a refresh.

I could go out and buy the very latest Broadwell-E or whatever, but it too wont be notably faster than my 5 year old 3930k, and when overclocked they would mostly tie.

So either way, if I buy a new CPU and motherboard I won't be seeing much - if any - of a performance upgrade. There will be a few percentage points difference, but nothing of note.

If I needed the extra performance because I was at the limit of what my current system could deliver, I'd get the fastest there is. This is why I have a Pascal Titan X. I needed every bit of juice I could possibly get out of a single GPU, so I went with it. As it stands however, I will need a replacement motherboard/CPU at some point, and I'd just be happy if it performs similarly to my existing system, with the same number of PCIe lanes or more.

All else being equal, I will buy AMD. Not because I am an AMD fanboy (though having an AMD CPU fills me with warm nostalgic fuzzies due to all the fun I had in 2000-2003 with Durons and Athlons) but because it is in everyone's interest that AMD does not die. If they do, our hobby is essentially over. Without competition, intel and Nvidia becomes even more like Comcast than they currently are.

I'd even argue that if you care about our shared hobby at all, you should buy AMD in every single case where they are good enough, and only spring for Intel/Nvidia when you absolutely need it.
 
I'll ignore this "experiment" until we get the final product, which reminds me, when does Zen actually release? Even if at low numbers, will we have any reviews before xmas? I just ask because I'm getting new cpu/mobo/ram this xmas, so I'd like to decide if I go AMD or Intel before then so I can put pieces on my xmas list :D. If I go Zen, despite CPUs not being available, I could put the RAM and mobo in my xmas list and buy the CPU a couple months later. Money saved is money saved! :D
 
euskalzabe said:
I'll ignore this "experiment" until we get the final product, which reminds me, when does Zen actually release? Even if at low numbers, will we have any reviews before xmas? I just ask because I'm getting new cpu/mobo/ram this xmas, so I'd like to decide if I go AMD or Intel before then so I can put pieces on my xmas list :D. If I go Zen, despite CPUs not being available, I could put the RAM and mobo in my xmas list and buy the CPU a couple months later. Money saved is money saved! :D
Click to expand...

I haven't heard anything new beyond the most recent statement to shareholders which is limited release in late Q4 2016, with full release to follow in Q1 2017. I feel fairly certain that as review sites will be sampled with enough time to do tests before launch day though, as usual for high profile releases.
 
You must log in or register to reply here.
Back
Top