First DirectX 12 game benchmarked *Update 2 More benchmarks

[Curufinwe_wins]

I was expecting to see some nonsensical rationalizing to justify preeliminar and meaningless results regardless. Nice to see I was not disappointed. 

INB4 GUYS! THE FURY X HAS ONLY BEEN RELEASED FOR TWO WEEKS LET AMD FINALIZE THE DRIVERS BECAUSE IT ISN'T A FAIR FIGHT ATM.

 

Seriously though, the biggest takeaways I see from this are nvidia ISN'T ready for DX12 (or at least not on this game.) and this game is a low thread count hog monster of a cpu stresser.

[Misanthrope]

Are you going to actually retort, or just make some wikipedia-level psych evaluation which shows you projecting more than anything else?

I did both already: preliminary results are mostly meaningless there's no need to find excuses for Nvidia, at least not yet anyway.

[Misanthrope]

INB4 GUYS! THE FURY X HAS ONLY BEEN RELEASED FOR TWO WEEKS LET AMD FINALIZE THE DRIVERS BECAUSE IT ISN'T A FAIR FIGHT ATM.

 

Seriously though, the biggest takeaways I see from this are nvidia ISN'T ready for DX12 (or at least not on this game.) and this game is a low thread count hog monster of a cpu stresser.

I did both already: preliminary results are mostly meaningless there's no need to find excuses for Nvidia, at least not yet anyway.

[Majestic]

I did both already: preliminary results are mostly meaningless there's no need to find excuses for Nvidia, at least not yet anyway.

 

I'm not excusing nvidia at all. Just saying the DX11 drivers from AMD were bad, as i've said in the past and received flak from it.

Merely adding the caveat that the huge scaling AMD does on the 390X is as much a good sign as a bad one. It means their cards are fast, but their DX11 drivers are ...well...shit.

 

And that right now, for DX11 builds, adding an AMD card into a budget build (with budget CPU) isn't a great idea. As it will have some pretty poor synergy.

[Curufinwe_wins]

 

Are you looking at the heavy results? The one that is actually CPU heavy (>20.000 drawcalls)?

http://www.pcper.com/files/imagecache/article_max_width/review/2015-08-16/ashesheavy-gtx980.png

No significant difference.

 

 

What are you talking about... 72 vs 66 fps, 67 vs 61 fps,  4 fps differences in the average test suite. That is MASSIVE for 1080p and two of the top gaming cpu's in the world.

 

Also clearly it's not single threaded. But it certainly doesn't show much if any scaling beyond 4 threads. Looks like the standard 3-4 thread game we have seen from dx11 in the past. 

 

If it was less than 3 threads the i3 should keep up fairly well, if it was more than 4 the 8370 should do much better (also the difference between the 8370 and 6300 being so small indicates it likely doesn't use more than 6 threads as the core clocks alone would make up the difference between the two.)

[Majestic]

What are you talking about... 72 vs 66 fps, 67 vs 61 fps,  4 fps differences in the average test suite. That is MASSIVE for 1080p and two of the top gaming cpu's in the world.

 

I'm looking at the DX12 portions ofcourse. As on DX11 drivers the 5960X had the same problem AMD CPU's are having. Lower IPC (3ghz vs 4.2ghz is huge) meant poorer scaling. The differences between the DX12 results of the 6700K and 5960X are insignificant.

 

The average testsuit is not good for looking at the CPU specifically. You need to look at the "heavy" results.

 

 

If it was less than 3 threads the i3 should keep up fairly well, if it was more than 4 the 8370 should do much better (also the difference between the 8370 and 6300 being so small indicates it likely doesn't use more than 6 threads as the core clocks alone would make up the difference between the two.)

 

Or the suffer from the same architectual bottleneck, since they're both made up of vishera cores.

[Curufinwe_wins]

I'm looking at the DX12 portions ofcourse. As on DX11 drivers the 5960X had the same problem AMD CPU's are having. Lower IPC (3ghz vs 4.2ghz is huge) meant poorer scaling. The differences between the DX12 results of the 6700K and 5960X are insignificant.

 

The average testsuit is not good for looking at the CPU specifically. You need to look at the "heavy" results.

I am... blue is dx12. its 72 vs 66 fps and 67 vs 61 for the 390x (which has higher fps in general so is more indicative of overall cpu usage.) The 980 shows little difference (1-4 fps) on the heavy side.

[Prysin]

 

Yep, AMD graphics cards were held back by drivers. I'm fairly sure the Fury X + intel CPU is going to be ballin'

 

 

 

Well, the bottleneck of their drivers were mostly visible with lower end CPUs (pentium, 860k, i3, 6300), which is why i suggested using a GTX 960 with those.

 

But at the higher end, while i knew it was there, the higher end GPUs performs so well in the first place that it was hard to believe they had this much left really

 

 

edit:

the difference between the 5960x and the 6700, could it be cause of better instruction sets/faster RAM?

[Curufinwe_wins]

Well, the bottleneck of their drivers were mostly visible with lower end CPUs (pentium, 860k, i3, 6300), which is why i suggested using a GTX 960 with those.

 

But at the higher end, while i knew it was there, the higher end GPUs performs so well in the first place that it was hard to believe they had this much left really

Not surprising when you compare floating point performance with Nvidia's equivalent cards. Almost a 40% difference in SP raw power.

[Misanthrope]

I'm not excusing nvidia at all. Just saying the DX11 drivers from AMD were bad, as i've said in the past and received flak from it.

Merely adding the caveat that the huge scaling AMD does on the 390X is as much a good sign as a bad one. It means their cards are fast, but their DX11 drivers are ...well...shit.

 

And that right now, for DX11 builds, adding an AMD card into a budget build (with budget CPU) isn't a great idea. As it will have some pretty poor synergy.

What I see is what we've been seeing since the release of the 390x: it's mostly between a bit and somewhat behind the 980 which is notable for a 2013 card. This game is behaving exactly the same and the numbers are where they should be, just the hardware limits of Hawaii/Grenada. This is not any kind of DX11 driver issue, this is DX12 performance gains due to processor power, hence it mostly disappears on low end processors.

[Xorbot]

Can someone please explain these 2 devices I see?  I am not familiar with them.

 

[trag1c]

If it was single threaded, the 5960X, 6700K and 4330 would've been much closer. The gap between the 4330 and 6700 says it's not single-threaded.

There multithreaded design is clearly lacking scalability . Probably a thread per subsystem, Im assuming they probably coded large parts of the engine and went "oh shit we're cpu bound" they should have easily been able to utilize every core/thread on each of those processors through queues and worker threads. That 5960x should have crushed those benchmarks if the game had a good multithreaded design. 

[Prysin]

Can someone please explain these 2 devices I see?  I am not familiar with them.

 

oh, i know what that is.

 

its called mECC RAM,

 

It works this way.

You grab this stick shaped thing, and doodle on the paper in a constructive manner. If the doodling seems incorrect, you automatically correct it. Once you are done with the doodle and has made your video, you discard the paper.

[Majestic]

There multithreaded design is clearly lacking scalability . Probably a thread per subsystem, Im assuming they probably coded large parts of the engine and went "oh shit we're cpu bound" they should have easily been able to utilize every core/thread on each of those processors through queues and worker threads. That 5960x should have crushed those benchmarks if the game had a good multithreaded design. 

 

It's not scaling to 16 threads, no. But it's not single-threaded either. My guess it's about a 6core scaling. As we also saw with the DX12 overhead test from 3Dmark.

[Misanthrope]

There multithreaded design is clearly lacking scalability . Probably a thread per subsystem, Im assuming they probably coded large parts of the engine and went "oh shit we're cpu bound" they should have easily been able to utilize every core/thread on each of those processors through queues and worker threads. That 5960x should have crushed those benchmarks if the game had a good multithreaded design.

This along with the lackluster results on AMD chips indicates that it works a little different: Single core performance it's still important over many cores, meaning that the superior clock speeds on the 6700k give it an advantage which is mostly not the case on other types of workloads that are multithreaded.

[Prysin]

It's not scaling to 16 threads, no. But it's not single-threaded either. My guess it's about a 6core scaling. As we also saw with the DX12 overhead test from 3Dmark.

which would explain AMDs slider showing actually diminishing returns with 8 cores....

 

http://techreport.com/r.x/2015_3_26_AMD_shows_off_DirectX_12_performance_with_new_3DMark_benchmark/cpuscaling.jpg

 

 

it gives this picture a bit more sense. Perhaps they knew that DX12 would only effectivly scale up to 6 cores?

[brownninja97]

Shit... I was thinking of buying a gtx 980.... But not really the time to judge yet...

nvidia drivers at least from the people i know of and myself have found the win10 nvidia drivers to be edgy at best

[Xorbot]

oh, i know what that is.

 

its called mECC RAM,

 

It works this way.

You grab this stick shaped thing, and doodle on the paper in a constructive manner. If the doodling seems incorrect, you automatically correct it. Once you are done with the doodle and has made your video, you discard the paper.

 

SCIENCE! What a wonderful time we live in...

[Mew]

This is very interesting. With mantle, AMD was able to eliminate the cpu as a bottleneck and achieve the same performance across every cpu. With DX12, that's not happening; fx is still garbage LOL

[zMeul]

How did you arrive at this conclusion?

quite simply, I presume

i3 has only 2 cores and 2HT while the AMD's FX-8370 has 8 cores

either the AMD's FX is crap or Intel's i3 is very good

[Kinda Bottlenecked]

nvidia drivers at least from the people i know of and myself have found the win10 nvidia drivers to be edgy at best

 

Yes I know about the current situation of Nvidia drivers but Boris Vorontsov iirc only develops and tests ENBs with Nvidia cards. So AMD cards are getting the shorter end of the stick apparently. But I don't really know how much of a performance difference there is between AMD and Nvidia.  Also AMD cards will have the edge in vram for high res textures and enbs with crazy AA. I need some real power for skyrim but I don't know who will give that to me.

 

Ahh... Decisions decisions... 

[brownninja97]

Yes I know about the current situation of Nvidia drivers but Boris Vorontsov iirc only develops and tests ENBs with Nvidia cards. So AMD cards are getting the shorter end of the stick apparently. But I don't really know how much of a performance difference there is between AMD and Nvidia.  Also AMD cards will have the edge in vram for high res textures and enbs with crazy AA. I need some real power for skyrim but I don't know who will give that to me.

 

Ahh... Decisions decisions... 

yeah skyrim can be a pain at times. For some reason the depth of field has just given up on me in skyrim not sure why but the game looks significantly worse without it even with my array of mods

[trag1c]

It's not scaling to 16 threads, no. But it's not single-threaded either. My guess it's about a 6core scaling. As we also saw with the DX12 overhead test from 3Dmark.

Going back to my point about a poor design. They probably implemented multhithreading into their engine after it was made and they probably implemented it based on subsystems which is the most half assed way of implementing concurrency within a game.

[Zah]

I would love to see this on more cards, but AMD's directx 12 drivers are already extremely good, and thank god.

[Faceman]

Just one benchmark, but an i3 is still beating out an FX8.