AMD Zen ES leaked benchmarks: performance similar to a Core i5 4670K?!?!

[DXMember]

11 hours ago, Prysin said:

he already left.

 

appropriate slogan is "Jim Keller: I'll be back"

probly, was like a trump reference though

 

On 8/10/2016 at 4:22 PM, SamStrecker said:

But how CPU bound is AOTS?

in the benchmark with "CPU test scenario" preset for AotS it's pretty CPU bound

[patrickjp93]

On 8/10/2016 at 8:17 PM, Coaxialgamer said:

True.  But the 5960x was a 22nm part.  Samsung's 14nm LPP should have better power characteristics... Anyway,  ill stay away from any performance for the time being,  especially from wccf... 

Under that argument, then why the high TDPs for the 6850K and 6900K?

 

I think AMD will try to pull a fast one and only give ratings for non-vectorized workloads if it intends to sell a 3.6GHz 8-core as a "95W" part. If you put an FX 8350 at stock speeds under V-Trace or another vectorized workload, it definitely spits out more than 125W.

[Coaxialgamer]

31 minutes ago, patrickjp93 said:

Under that argument, then why the high TDPs for the 6850K and 6900K?

 

I think AMD will try to pull a fast one and only give ratings for non-vectorized workloads if it intends to sell a 3.6GHz 8-core as a "95W" part. If you put an FX 8350 at stock speeds under V-Trace or another vectorized workload, it definitely spits out more than 125W.

Again,  you can't really trust exact TDP measurements from the manufacturer are often inaccurate because they are often used across a whole cpu family ( that realistically put out different amounts of heat),  and there isn't a standardized way of measuring it. 

[cesrai]

On August 11, 2016 at 1:15 AM, themaniac said:

that argument only works with gpu's because amd and intel cpu's both use the same architecture(or whatever x86-64 is called) aka x86-64 so you can't really optimize the cpu side for just amd cpu's

You're not serious right ? remember Intel cinebench scandal ? 

[Notional]

10 hours ago, zMeul said:

Guru3D did something: https://www.guru3d.com/news-story/amd-zen-engineering-sample-aos-further-analysis.html

they took a  i7 5960X Haswell-E (because it matches the core count) and downclocked it to match the ZEN engi sample, put a RX480 and ran Ashes at the same settings - it does not look good for ZEN

 

example:

That's nice. What Guru3D ignored is that there's already 2 engineer samples out of ZEN, but paired with a 480, both 8 core, 16 threads.

 

1D2801A2M88E4 32/28 scores 5,300.

 

2D2801A2M88E4 32/28 scores 3,100.

 

Engineering samples are really useless to conclude upon, as certain blocks can be deactivated, certain functions doesn't have driver/OS/thread manager support and it most certainly runs at a lower clock rate as well.

 

I don't know if the 8 core ZEN will be similar to a 5960x, but if it is, it will be very expensive. I'm sure it will be a bit slower, but bring 8 core to mainstream, which is very nice. But concluding on a product that's about half a year away is pointless.

[Marinatall_Ironside]

14 hours ago, Prysin said:

ladies and gentlemen. Here he is, the most paranoid hater in the news section. The one, the only, MIIIIIIISSAAAAAANTHROOOOOOOOOPEEEE!!!!

Cut the ad hominem bullshit.

 

On topic: If this is the best that AMD's Zen microarchitecture can compete, I am not impressed at all. And even if it's lower in price compared to Intel, if the PPD metrics are reasonable enough, Intel will still be the clear winner in the CPU market in terms of single-threaded performance. I want to AMD to compete again, but from what this is looking like, there's no way AMD can compete except in multi-threaded applications (which there's only a few games and a good handful of applications that can leverage multiple CPU cores and threads).

[magnumkobra]

On 8/11/2016 at 7:08 AM, ThinkWithPortals said:

till speculation of course, but I pray to God that Zen is a hit. AMD needs the success.

 

[magnumkobra]

man what a mess this topic has created. Every little rumor or bench is another war against people, but why do we war in the comments? Is not as if it will change the world, yet we still do. I hope in angst that zen will be the turning point in amd's loosing streak , but untill the first review samples are shipped. we will not know. 

[Morgan MLGman]

On 11.08.2016 at 2:06 AM, Dabombinable said:

It still has a 95W TDP.......and going from their past actions it wouldn't surprise me at all if the 95W TDP is only for the base clock.

So what? 5820K has 140W TDP and noone bats an eye. More than the FX-8350.

[Dabombinable]

Just now, Morgan MLGman said:

So what? 5820K has 140W TDP and noone bats an eye. More than the FX-8350.

Look at the benchmarks again-it was underclocked to the same clock speed-and FYI its also on a 22nm node with a FIVR, not 14nm without a FIVR so there is that as well. The performance of Zen going of the leak+Guru3D bench means it performs like shit-which is what we don't want if AMD is to become competitive again.

[Misanthrope]

Best way to deal with ad hominem is to just move on. If it was a joke only, it was fine. If it wasn't, I think everyone can see that I didn't just pull random conjectures out of my ass: it is extremely unlikely that anyone outside of AMD has this engineering sample and is not beneath 100 layers of NDAs so it probably did come from AMD directly. The rest is just an observation on how it seems to be benefical speculation to me since it builds hype.

 

Also

 

Spoiler

 

[Prysin]

3 minutes ago, Misanthrope said:

Best way to deal with ad hominem is to just move on. If it was a joke only, it was fine. If it wasn't, I think everyone can see that I didn't just pull random conjectures out of my ass: it is extremely unlikely that anyone outside of AMD has this engineering sample and is not beneath 100 layers of NDAs so it probably did come from AMD directly. The rest is just an observation on how it seems to be benefical speculation to me since it builds hype.

 

Also

 

  Reveal hidden contents

 

aww. that picture gets me all hot and bothered. That mustache. So sexy.

[Void_Viking]

Will this be a whole new socket type? Or will they continue to use the AM3+?

[Marinatall_Ironside]

Just now, Void_Viking said:

Will this be a whole new socket type? Or will they continue to use the AM3+?

Bristol Ridge and Summit Ridge (both powered by Zen) uses the AM4 socket.

[Void_Viking]

Just now, JurunceNK said:

Bristol Ridge and Summit Ridge (both powered by Zen) uses the AM4 socket.

Fuck, I'll have to get a new motherboard  lol

[LeStringMan]

12 minutes ago, JurunceNK said:

Bristol Ridge and Summit Ridge (both powered by Zen) uses the AM4 socket.

Bristol Ridge is 28nm and based on Carizzo, but it will use socket AM4.

[Morgan MLGman]

1 hour ago, Dabombinable said:

Look at the benchmarks again-it was underclocked to the same clock speed-and FYI its also on a 22nm node with a FIVR, not 14nm without a FIVR so there is that as well. The performance of Zen going of the leak+Guru3D bench means it performs like shit-which is what we don't want if AMD is to become competitive again.

Nah, I didn't mean anything about performance as I personally would like to wait for some more info that isn't 1. AOTS and 2. ES. This is also a misleading title imho, as people might think that an 8c/16t Zen CPU will be of the same performance as an 4670K, which is ridiculous, as FX-8cores outperform it multithreaded-wise already

 

What I meant is that there's so much talk in this thread about "TDP", as if it really mattered that much... People get Haswell-E/Broadwell-E i7s without even mentioning their TDP that's above FX-8 line "cause FX CPUs are hot as fuck and will make your PC a toaster", I've had an OC'd 125W FX-8350 and now I've got a 65W i7-3770S which has half 8350s TDP.

And guess what, there's no difference. What's funny, FX had lower temps under load.

[zMeul]

38 minutes ago, Morgan MLGman said:

What I meant is that there's so much talk in this thread about "TDP", as if it really mattered that much... People get Haswell-E/Broadwell-E i7s without even mentioning their TDP that's above FX-8 line "cause FX CPUs are hot as fuck and will make your PC a toaster", I've had an OC'd 125W FX-8350 and now I've got a 65W i7-3770S which has half 8350s TDP.

And guess what, there's no difference. What's funny, FX had lower temps under load.

Haswell, Haswell-E, Broadwell and Broadwell-E all have the VRM on the CPU package, it's called FIVR - that's why they have those hi TDP ratings

from Skylake onward, Intel went back and removed the VRM from the CPU package - lower TDP, higher performance

 

if ZEN isn't able to compete with Haswell, what the hell AMD is it gonna do against Skylake and KabiLake?

[Morgan MLGman]

16 minutes ago, zMeul said:

Haswell, Haswell-E, Broadwell and Broadwell-E all have the VRM on the CPU package, it's called FIVR - that's why they have those hi TDP ratings

from Skylake onward, Intel went back and removed the VRM from the CPU package - lower TDP, higher performance

 

if ZEN isn't able to compete with Haswell, what the hell AMD is it gonna do against Skylake and KabiLake?

Kaby Lake will be mainly power consumption optimizations etc, not performance ones, so performance-wise it will be only a little bit faster than Skylake

[Briggsy]

On 2016-08-10 at 5:10 PM, leelaa14 said:

Ashes of the singularity is AMD's marketing tool to con people into thinking their products are better than they are.

This is especially true on the GPU side. Its basically a Halo product but in reverse. Instead of showing the performance of their best product, AMD shows the potential of all their products. It's like an internet provider saying "up to 1Gbps" while on average providing less bandwidth than the competition that says "guaranteed 0.95 Gbps bandwidth minimum." AMD using the "up to" performance metric surprisingly charms a lot of buyers.

 

Whether AotS translates to better performance for Zen... who knows. It definitely makes the Bulldozer/excavator/piledriver/stoolpusher architecture look good (even though AMD still use Intel CPU's in their AotS benchmarks  ).

[MageTank]

8 hours ago, patrickjp93 said:

Under that argument, then why the high TDPs for the 6850K and 6900K?

 

I think AMD will try to pull a fast one and only give ratings for non-vectorized workloads if it intends to sell a 3.6GHz 8-core as a "95W" part. If you put an FX 8350 at stock speeds under V-Trace or another vectorized workload, it definitely spits out more than 125W.

I try to tell people not to trust TDP in general. Even when undervolted to 1.1v, my 6700k still pulls 120w of power under Prime95. At stock 1.23v, this was 140w. Pretty far from its 91w TDP advertisement. Granted, this is not the intended use, and under normal consumer workloads one is unlikely to encounter that level of stress. 

 

Intel further covers themselves with this definition of TDP:

Quote

Thermal Design Power (TDP) represents the average power, in watts, the processor dissipates when operating at Base Frequency with all cores active under an Intel-defined, high-complexity workload. Refer to Datasheet for thermal solution requirements.

 

I can't imagine this 8 core, 16 thread CPU being 95w unless it was 3ghz or so. Even then, that number would skyrocket under any real stress. 

[Godlygamer23]

On 8/10/2016 at 5:31 PM, valdyrgramr said:

Um, remember that Batman game where they sped up the video?  That was one game they depended on. 

Sped up the video?

[zMeul]

6 hours ago, Morgan MLGman said:

Kaby Lake will be mainly power consumption optimizations etc, not performance ones, so performance-wise it will be only a little bit faster than Skylake

and?

ZEN "has" issues keeping up with Haswell  an arch that's nearly 3y old

[LAwLz]

Haven't read the thread so sorry if this has been asked before. How CPU bound is AotS? Testing CPUs with games has never been great because of the high GPU bottlenecking. 

[Prysin]

2 hours ago, LAwLz said:

Haven't read the thread so sorry if this has been asked before. How CPU bound is AotS? Testing CPUs with games has never been great because of the high GPU bottlenecking. 

very.

 

To explain it.

Imagine Total war with two HUGE infantery units fighting. Then multiply the army sizes by 11. Add "more sophisticated AI". Add "fast changing FoV". Then subtract "highly optimized multi-threading", substract "low GPU driver overhead" and finally divide by the number of cores.

 

On a more serious note. i've run it on my 295x2 and i7 4790k. Here is some screens when i tried to disable hyper-threading and make my CPU into a i5......

I tested the i5 with both 1600Mhz and 2400Mhz memory. As @MageTank and myself has proven in the past, faster memory helps when the CPU is the bottleneck, (except when you have hyperthreading. It doesnt do much in those scenarios)....

http://imgur.com/a/wU1zV

 

Ill do a run with my i7 in a little bit.

Edit: here is tests with a i7 4790k at stock speeds (4.0Ghz base, 4.4Ghz boost). 32GB of DDR3 1866Mhz memory and AMD Radeon R9 295x2 both in single and dual GPU mode. I also included the CPU stress test, which is PURELY a CPU test (max CPU load the game engine can create)...

Also have logs and screens from MSI afterburner showing different metrics.

 

http://imgur.com/a/sinWH

 

EDIT 2:

i have screens of a AMD FX 8320 at 1080p, but since the benches of my "i5" and i7 was done at 1440p, it wouldnt be apples to apples...

The "i5" benches was done A LONG time ago. Right after official launch... this is why "Data compression" isnt mentioned in the "i5" test.... game has probably changed quite a bit since then, so i should probably re-do all my tests.