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

[Prysin]

8 hours ago, themaniac said:

this is aots right? so how many cores can it use at the same time? if its only 4 then can believe that amd has a chance with zen if this is true

Ashes should scale well up to 6 cores as it fully utilizes the CPU multi threading capabilities of DX12.

 

8 hours ago, leelaa14 said:

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

yes and the US government has been installing mind controlling chips in black people to make them commit more crimes since the slavery era.

 

8 hours ago, SamStrecker said:

But how CPU bound is AOTS?

according to AMDs own slides from the past, DX12 scales well up to 6 cores/threads before diminishing returns kick in. Unlike DX11 which require a godlike amount of optimization to make even 3 cores/threads run well. Yes, DX11 struggles to make ANY real benefit of more then 3 cores/threads. If you listen to a few youtube lessons about coding and multi threading, then you will find out just how bad the CPU utilization is pre DX12 and Vulkan.

 

Sad fact is. We have had open source, publically availible multi.threading coding, that requires you to insert 8 lines of code infront of your workload to make it use any amount of threads/cores since 1994..... yes 1994.

Yet we have still not even attempted to make a proper multi threaded API before now.

 

Even sadder is. This set of codes and instructions, were developed for super-computers and HPC, for the specific task of taking normally single threaded workloads and turning them into multi-threaded workloads. You could in essence copy paste these lines of text infront of every workload in DX11 to make them run in parrallell as much as possible, without it even affecting single core CPU at all. The instruction set and compiler itself would recognize the availible resources and deal with it automatically.

 

if you dont believe me. Look up OpenMP.

[Curufinwe_wins]

6 minutes ago, Prysin said:

snip

Notable exception being that the Cryengine 3 is extremely well tuned for cpu scaling, even more so than many of the current half-ass DX12 ports.

 

But totally sad indeed.

[Thony]

On 8/10/2016 at 4:07 PM, Djole123 said:

So as you might know, Zen is just around the corner and we're waiting for some performance numbers. Some rumors in the past said that it'll have single performance similar to a Haswell chip. And boy they were right!

 

 

 

There are some Zen ES leaked benchmarks so lez have a look:

 

I can't beleive I'm saying this publicly, but, the benchmarks say that it's on par with the top tier Haswell i5 4670k.

 

But since this is an engineering sample it's not fully utilised, and I can safely say that once they rev up the clockspeeds, we can see benchmarks go to Broadwell territory.

 

Source and more details at:

http://wccftech.com/amd-zen-es-benchmarks/

Around a corner ?! If u think 6 months is around the corner then we have very different representation of the phrase. Last time i checked only small sample of Zen cpus was getting released later this year with full release early next year.

[SurvivorNVL]

So, a 4670k worth of Zen will likely be going up against Kaby-Lake, as expected, and then Skylake-E in eight-ish months.  I'm just not going to upgrade until Cannonlake for Intel's unified socket.

[laminutederire]

9 hours ago, leelaa14 said:

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

 

9 hours ago, leelaa14 said:

I completely agree with you on this, but as I said, the fact they are only using one is worrying, that's all. 

Guys and many more guys, it's a leaked benchmark, it isn't from amd themselves you know

[Prysin]

2 hours ago, Curufinwe_wins said:

Notable exception being that the Cryengine 3 is extremely well tuned for cpu scaling, even more so than many of the current half-ass DX12 ports.

 

But totally sad indeed.

well... to be honest.

Just like many other DX11 based games, the engine itself may be tuned, but the API, aka DX11, doesnt know how to utilize cores. Meaning in games, you can see high core utilization, but the scaling is terrible after al.

 

Reason for this is that currently games works like this:

Game engine -> API (DirectX / OpenGL) -> Graphics driver and compiler -> hardware

 

Issue is that the API is the bottleneck here.

 

Prime example is GTA V. Total Biscuit proved that GTA V can use up to 16 threads (with his 5960X), however looking at his FPS scores, it doesnt show up to be "leaps and bounds" faster then a normal i7. A normal i7 will have the normal 20-30% boost over an i5, but that is due to hyperthreading and overhead in CPU utilization when hyper threading is not present.

Hyper threading is technically the same as the Async Compute system that is in AMDs GCN graphics processors. 

 

The thing about DX12 games, EXCEPT ASHES, is that the "DX12" games you talk about. Hitman, RotTR, Total War: Warhammer etc is basically DX11,3. Aka the API that the XBONE uses. It has several of the features of DX12, it has low level access, but it is NOT going to be equal to proper "desktop" DX12. Because 11,3 was made for a GCN APU specifically. And whilst you can harness many of the same plugins and reuse them for DX11 on desktop (aka slapping Async compute on the engine and calling it DX12), the game engine itself is still DX11 on the CPU side. So all the benefits from CPU utilization isnt there.

 

This is why that, outside of Ashes, we do not see insane jumps of performance between a i3 and a i5, or a i5 and a i7. Or a i7 vs x99 i7.... Because the CPU core/thread scaling simply isnt there. They simply used Async compute and a few other GPU optimizations to reduce GPU overhead, because that is something devs has done for GCN in Consoles since 2012 or so. They know how to get those easy boosts in performance for GCN. They also know that Nvidia now supports it (although poorly), and thus it is a "hotfix" for performance. It does however not improve the underlying issue of poor CPU core utilization.

 

Time will show though. Will be interesting to see if AMDs prediction of 6 cores/threads "max" scaling of DX12 actually holds true.

[totjup5]

11 hours ago, SLAYR said:

I kinda want to see what it would do at 5GHz though (new 9590?), any word on the TDP.

 

and really the only way to not have a bias, and actually find out how good the IPC really is would be to run cpu benchmarks like Cinebench.

It's supposed to be 95w i'm pretty sure 

[Castdeath97]

13 hours ago, leelaa14 said:

I have already been corrected in terms of cpu performance and how it works. To answer your comment...Nvidia don't stick to one game like AMD have stuck to ashes, no. When they showed the 1080 they showed it on "multiple" games. Multiple being the key word here.

At least with Nvidia they are saying yes, it will work well on these games. AMD are showing just one, which to me is kinda weird.
I ain't about to get into a fan boy war though, as I said, I'm just saying what I see.

They keep showing Dirt Rally and probably Hitman as well. 

[Lawliet93]

On 8/10/2016 at 8:42 PM, AresKrieger said:

Yeah, didn't they say 980+ performance in the rumors on those lol, 970 performance is what we got, this is why "leaks" always need proper seasoning

AMD never said, that the rX480 will have 980 performance.... they said about R9 290x level and the card beats gtx 970´s OC @ 1,4GHz... so get your facts right kid.

 

Well... The CPU is running at 2,8GHz on all cores (3,2GHz is just a single core turbo) and it beats the fx 8350 by about 40%, the i5 4670k by about 15%+ and... at 2,8GHz. 

 

Overall... a great jump from AMD.

 

http://www.overclock3d.net/news/cpu_mainboard/benchmarks_for_amd_s_zen_es_cpus_have_been_leaked/1

[Prysin]

On 8/11/2016 at 0:20 AM, Curufinwe_wins said:

Hence transparent....

most people also know that Cinebench uses the Intel compiler and is heavily optimized on the INtel side and not on AMD side. This is no secret, its been known for years, and once you apply the same optimizations for AMD CPUs, they suddenly gain 20%+ performance.

 

However people ignore this and cry "bias bullshit". Despite the compiler in use LITTERALLY being a raw copy of Intels compiler. The same compiler intel used to cheat with during Athlon period.

 

On 8/10/2016 at 7:39 PM, Dabombinable said:

Its still too far behind though.

too far from what?

 

again, you are posting biased bullshit not based on reality, but your own twisted version of it.

Howeever you are, as always, incapable of understanding the concept of performance. Age doesnt matter one single shit. Special features are just marketing ploys for the most part. What fucking matters is PERFORMANCE.

 

Can you play on Sandy Bridge today? YES.

Is it as good as latest gen? NO

will the average consumer be able to play at 1080p 60Hz with Sandy Bridge? YES

 

So if a CPU from 5 years ago is doable today, what does that say about performance equal to what was "gr8 m8 8/8" this time last year???!

 

AMD will be ONE generation behind, unless kaby lake is truly revolutionary. The real "jump" from Intels side is Cannon lake in 2018. That is when AMD need to dig deep. In the meantime, they will be within spitting distance of most intel products.

 

 

The only thing i agree with is what @zMeul said about the promontory chipset. Honestly, that will add heat, and kill overclocking very fast. I'd rather see them do like intel and move NB on chip and SB on board or vice versa, dont remember which Intel is doing. 

[Dabombinable]

Just now, Prysin said:

too far from what?

 

again, you are posting biased bullshit not based on reality, but your own twisted version of it.

Howeever you are, as always, incapable of understanding the concept of performance. Age doesnt matter one single shit. Special features are just marketing ploys for the most part. What fucking matters is PERFORMANCE.

 

Can you play on Sandy Bridge today? YES.

Is it as good as latest gen? NO

will the average consumer be able to play at 1080p 60Hz with Sandy Bridge? YES

 

So if a CPU from 5 years ago is doable today, what does that say about performance equal to what was "gr8 m8 8/8" this time last year???!

 

AMD will be ONE generation behind, unless kaby lake is truly revolutionary. The real "jump" from Intels side is Cannon lake in 2018. That is when AMD need to dig deep. In the meantime, they will be within spitting distance of most intel products.

 

 

The only thing i agree with is what @zMeul said about the promontory chipset. Honestly, that will add heat, and kill overclocking very fast. I'd rather see them do like intel and move NB on chip and SB on board or vice versa, dont remember which Intel is doing. 

Too far behind their competitors current generation, 2 fucking generations behind like Piledriver was at launch.

[AresKrieger]

25 minutes ago, Lawliet93 said:

AMD never said, that the rX480 will have 980 performance.... they said about R9 290x level and the card beats gtx 970´s OC @ 1,4GHz... so get your facts right kid.

I never said amd said anything, I said a "leak" said that, and don't call me kid to try invalidate my statement as that just makes you look like a fool

[deviant88]

until official benchmarks its all just speculation, by the looks of it its going to be another 8150 scenario but this time with lower power consumption, an 8 core 16thread getting trashed by a 4core 8 thread is unnacceptable unless the benchmark sucks

[Majestic]

As fast as a 2013 quadcore. Boy, i'm so excited.

 

[Sauron]

Can we take a moment to realize that this is a game benchmark? I know this is where amd's cpus have been lacking in the past few years but there's much more than that to a cpu's performance...

 

On 8/11/2016 at 6:19 AM, deviant88 said:

until official benchmarks its all just speculation, by the looks of it its going to be another 8150 scenario but this time with lower power consumption, an 8 core 16thread getting trashed by a 4core 8 thread is unnacceptable unless the benchmark sucks

It's a game, games usually don't use more than 4 threads.

[Skiiwee29]

So is the official Zen unveiling the reason all the youtubers are in London right now?

[daniellearmouth]

2 minutes ago, legacy99 said:

So is the official Zen unveiling the reason all the youtubers are in London right now?

They don't have an office or anything in London (the closest is Milton Keynes), and even then, it's a bizarre place to reveal something if you're a multi-national company when all that's needed is to go to Sunnyvale in California.

[Skiiwee29]

Just now, daniellearmouth said:

They don't have an office or anything in London (the closest is Milton Keynes), and even then, it's a bizarre place to reveal something if you're a multi-national company when all that's needed is to go to Sunnyvale in California.

Weirder things have happened... 

[TidaLWaveZ]

I hope the initial Zen offering is at least the performance of a 5820k.  I'd really like to see a powerful AMD cpu that overclocks like the dickens!

[ThinkWithPortals]

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

[patrickjp93]

On 8/11/2016 at 5:56 AM, Prysin said:

most people also know that Cinebench uses the Intel compiler and is heavily optimized on the INtel side and not on AMD side. This is no secret, its been known for years, and once you apply the same optimizations for AMD CPUs, they suddenly gain 20%+ performance.

 

However people ignore this and cry "bias bullshit". Despite the compiler in use LITTERALLY being a raw copy of Intels compiler. The same compiler intel used to cheat with during Athlon period.

No it doesn't. Cinebench switched to using GCC a long time ago.

 

On 8/11/2016 at 6:05 AM, Prysin said:

too far from what?

 

again, you are posting biased bullshit not based on reality, but your own twisted version of it.

Howeever you are, as always, incapable of understanding the concept of performance. Age doesnt matter one single shit. Special features are just marketing ploys for the most part. What fucking matters is PERFORMANCE.

 

Can you play on Sandy Bridge today? YES.

Is it as good as latest gen? NO

will the average consumer be able to play at 1080p 60Hz with Sandy Bridge? YES

 

So if a CPU from 5 years ago is doable today, what does that say about performance equal to what was "gr8 m8 8/8" this time last year???!

 

AMD will be ONE generation behind, unless kaby lake is truly revolutionary. The real "jump" from Intels side is Cannon lake in 2018. That is when AMD need to dig deep. In the meantime, they will be within spitting distance of most intel products.

 

 

The only thing i agree with is what @zMeul said about the promontory chipset. Honestly, that will add heat, and kill overclocking very fast. I'd rather see them do like intel and move NB on chip and SB on board or vice versa, dont remember which Intel is doing. 

NB is on-die, SB on the motherboard.

[Bcat00]

Yawn another weak product from AMD.

 

call me when they produce something ground breaking instead of playing catchup

[patrickjp93]

On 8/11/2016 at 8:06 AM, Bcat00 said:

Yawn another weak product from AMD.

 

call me when they produce something ground breaking instead of playing catchup

Still, 40% IPC increase in a single generation is impressive. Make no mistake, Keller did a lot of good work. But, this is where it gets tough, and AMD is going to have to sink more money into improving the throughput of its chips than it has had to before just to achieve those 5-7% gains each generation. AMD's finances are going to be the thing to watch. If sales don't seriously pick up, then AMD's goose is cooked in 2019.

 

On 8/11/2016 at 0:47 AM, Prysin said:

Ashes should scale well up to 6 cores as it fully utilizes the CPU multi threading capabilities of DX12.

 

yes and the US government has been installing mind controlling chips in black people to make them commit more crimes since the slavery era.

 

according to AMDs own slides from the past, DX12 scales well up to 6 cores/threads before diminishing returns kick in. Unlike DX11 which require a godlike amount of optimization to make even 3 cores/threads run well. Yes, DX11 struggles to make ANY real benefit of more then 3 cores/threads. If you listen to a few youtube lessons about coding and multi threading, then you will find out just how bad the CPU utilization is pre DX12 and Vulkan.

 

Sad fact is. We have had open source, publically availible multi.threading coding, that requires you to insert 8 lines of code infront of your workload to make it use any amount of threads/cores since 1994..... yes 1994.

Yet we have still not even attempted to make a proper multi threaded API before now.

 

Even sadder is. This set of codes and instructions, were developed for super-computers and HPC, for the specific task of taking normally single threaded workloads and turning them into multi-threaded workloads. You could in essence copy paste these lines of text infront of every workload in DX11 to make them run in parrallell as much as possible, without it even affecting single core CPU at all. The instruction set and compiler itself would recognize the availible resources and deal with it automatically.

 

if you dont believe me. Look up OpenMP.

There is some overhead added because it's a runtime and not a static library, and OpenMP is not a perfectly scaleable solution because it runs on the fork-join model, but that doesn't change the fact we could have had multithreaded APIs as far back as 2006 when it became an open standard. Then, if performance really had been driven to its limits, we could have moved on to future unrolling (the mainstay of the HPX library).

 

Warning: not the best presenter, and color choices for slides need improvement, but the code examples are comprehensive, and he explains them very well. Also, he isn't completely transparent about comparing OpenMP's performance, and I suspect he just used static scheduling in his tests when guided scheduling was obviously a better choice.

Spoiler

 

 

[deviant88]

1 hour ago, Sauron said:

It's a game, games usually don't use more than 4 threads.

What are you talking about, games and complex program(even simple ones sometimes) use tens of threads, maybe you meant there is no asynchronous multithreading for game jobs (feeding data to gpu, physics, audio, animation, AI etc), its probably running on 1 core, as a developer if you implement async multithreading you must be retarded to just set it to hardcoded 4 cores, usually this is set to use all available cores/threads.

 

Its so boring to read the same old silly uninformed reply "bla bla dont use more than 4 cores, or 4 threads".

If your program has 100 threads they will be managed by the kernel and spread across all available hardware threads(including virutal ones, hyperthreading etc). If you program your rendering,AI physics etc on thread 1, and the rest of the threads are only doing minor work or occasionally or idling it doesnt mean the game engine or OS didnt use all the cpu execution threads. Its the work thats not properly spread/optimized by the programmers of the game engine.

[patrickjp93]

21 minutes ago, deviant88 said:

What are you talking about, games and complex program(even simple ones sometimes) use tens of threads, maybe you meant there is no asynchronous multithreading for game jobs (feeding data to gpu, physics, audio, animation, AI etc), its probably running on 1 core, as a developer if you implement async multithreading you must be retarded to just set it to hardcoded 4 cores, usually this is set to use all available cores/threads.

 

Its so boring to read the same old silly uninformed reply "bla bla dont use more than 4 cores, or 4 threads".

If your program has 100 threads they will be managed by the kernel and spread across all available hardware threads(including virutal ones, hyperthreading etc). If you program your rendering,AI physics etc on thread 1, and the rest of the threads are only doing minor work or occasionally or idling it doesnt mean the game engine or OS didnt use all the cpu execution threads. Its the work thats not properly spread/optimized by the programmers of the game engine.

Oh boy are you deluded... Most existing games really do not use more than 4 threads at a given time, and even then the execution balance is such that 1 or 2 are really bogged down and the others aren't. And part of this comes down to game devs not being even middle tier parallel programmers who know how to decompose a problem down to brass tacks. Usually after 4 threads, most engines completely give up.

 

Now, for automatic parallelism libraries that are directive-based, you actually can (and should) hard-code thread limits (OpenMP is one such example, CilkPlus and TBB being others), because unless you're a real expert, you're using fork-join parallelism anyway, and the joins have real overhead if the threads are launched often.

 

And if by asynchronous computing you mean using C++ std::async(...) this is still fork-join parallelism, and the core count is decided at compile time, not by the programmer.