• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

[bitsandchips] First Zen Benchmarks

Status
Not open for further replies.
S

Silverforce11

Lifer
http://www.bitsandchips.it/english/...ks-the-next-amd-uarch-is-a-major-step-forward

zhash1.png


zmem1.png


zraytrace1.png


According to our reliable sources, Zen represents a major step forward. This uArch will be a lot better than Excavator, and will be on a par with – or very near to – Intel Haswell uArch.

We have some screenshots about particular AIDA64 benchmarks (v5.70). The sample tested is an 8 cores Summit Ridge CPU (Myrtle Reference Board), and it runs at 3,0 GHz.

However, thanks to these benchmark, we can say that Zen seems to be very good. The DDR4 bandwitdh is almost on pair with the bandwidth of the i7-6700K, while the score about CPU Hash benchmark is three times better than the score of the i7-5820K. Good, but not fantastic, the FP32 Ray Trace score.

The sample used is the A0 revision (So, take these results with a grain of salt. Next revision will improve the results). The CPU runs at 3,0 GHz. The Turbo Boost feature is still disabled. The Hash score is very high thanks to the SHA hardware acceleration, but ... why is the FP32 Ray Trace score so low? It seems that Zen will have a weak AVX2 implementation, but the pure Floating Point Unit will be able to match up with the best Intel FPUs.

Infact, our source claims that the 8 cores Summit Ridge CPU will be able to perform like the 8 cores Broadwell-E CPUs. Because of that, Intel will commercialize a 10 cores Broadwell-E CPU, and because of that Intel has backtracked: next Broadwell-EX/EP server CPUs are still here.

-------------------------

April Fools or potentially legit?

Looks reasonable to me given Lisa Su said it's on track for greater than 40% IPC gains over excavator.

Zen 8 Core duking it out vs 5960X!
 
DrMrLordX said:
I was worried that AVX2 might be gimped on Zen, just as it is on XV.
Click to expand...

Based on Dresdenboy's analysis, there is reason to believe that Zen's AVX implementation will be less powerful than Haswell and following architectures. (It might come close to Sandy Bridge, though.) Fortunately, AVX isn't all that commonplace; MMX/SSE/SSE2 performance is much more important. Keep in mind that Intel's low-end CPUs don't even support AVX at all, so few applications require it (even though some can benefit from using it as an option).

Rvenger said:
Zen has HT eh? Fake...
Click to expand...

We've known for a while that Zen will have SMT. It could be that AIDA is labeling any SMT feature as "HT", because Intel's HyperThreading is the only public implementation of that feature as of this time. It is, of course, possible that the screenshots may be fake, but I don't think this is conclusive evidence one way or the other.
 
Legitimate or not, you can't conclude an awful lot from this.

The SHA1 test uses hardware acceleration and is therefore only a useful measurement of SHA1. Where the Zen number is probably vast overkill for any realistic scenario.

I disagree that the bandwidth is almost on par with i7-6700K. The comparison here is not using the same DDR4 clock speed. You can find a measurement of i7-6700K with DDR4-2400 here:

http://www.legitreviews.com/ddr4-me...nding-the-best-ddr4-memory-kit-speed_170340/3

So it's more like 35547 vs 27613 were Intel is still 28.7% ahead (with a 4C/8T part). Still not a totally perfect comparison since the timings aren't the same, but probably much much closer. Normally I'd say that's probably still fine for most code, but if you want to justify 8C/16T you really do want a lot of bandwidth sometimes.

That leaves the FP32 ray tracing bench, one that was apparently just released in AIDA64 a few days ago. When you correct for clock speed and core count among same generation Intel CPUs it scales extremely well with core count. So just a very well threading test, where it doesn't do that well for the core count (unless this 8 core CPU costs less than a 4 core Skylake).

On the other hand, this test gets really unusually high boosts of around 30% going from SB/IB to HW/BW and again to Skylake. Who knows what that's about. Makes it seem like not the most representative thing in the world.
 
my quick google says its FMA and AVX workload, so i assume its the increase in perf comes from the increase in load store from SB/IB to Haswell.
 
Yet Another Zen Thread, and the op got fooled by an April fools joke.

There's one Broadwell thread, one Skylake thread, how about wet stick to one Zen thread also.
 
Last edited:
Yet Another Zen Thread, and the op got fooled by an April fools joke.

There's one Broadwell thread, one Skylake thread, how about wet stick to one Zen thread also.
Click to expand...

@moderation this would be a moot point if CPUs and Overclocking had subforums like Video cards and Graphics
Can we has?
 
Put this in your "for whatever it's worth column" but when I was upgrading my licensed version of Aida 64 the other day, I read the update notes and it mentioned that it added support for the Summit Ridge. I think that adds a kernel of truth to these results. Especially since the results are for an Aida64 Engineer version which is quite pricey.
 
I understand about MS Paint, Gimp AND it is April Fools day. Thus my comment about a "kernel" of truth to the posting!😀

I just ran my Aida64 for the same 3 tests and the one I think that is out of wack is the Hash result. My Hash score for a 5960X at 4.4Ghz is 90,277. However, the other two scores seem consistent.
 
Last edited:
Status
Not open for further replies.

TRENDING THREADS

Back
Top