computerbaseAshes of the Singularity Beta1 DirectX 12 Benchmarks

[3DVagabond]

Made a mistake in typing which I corrected above. TheMedium enabled score was 34.9 NOT 43.9, transposition in typing ---sorry.:'(

NP. Stuff happens.

so not much difference either way for you and your setup(s). I don't know if your CPU's are making much of a difference. It seems the 4790K should be just fine. It's strange that even with the tech sites we are seeing widely varying results. Some are showing AMD killing nVidia. Others not much difference at all.

 

[Dygaza]

I think it's a bit harsh to call Hilbert from guru3d out only because of his misunderstanding of fcat and it's compatibility with dx12. Afterall, someone had to be the first to notice this programs misbehaviour. And it's good that this happened so fast in dx12 era, so we don't get many reviews tainted with this misunderstanding. Also now it's public knowledge that it ain't reliable.

Is lack of support in current driver for DirectFlip bad press for AMD. Or is lack of support for this new version bad press for nvidia? Well nobody seems to be talking about the latter, so quess this is more bad press for AMD, but things are getting sorted out anyways.

 

[Dygaza]

Made a mistake in typing which I corrected above. TheMedium enabled score was 34.9 NOT 43.9, transposition in typing ---sorry.:'(

Did you had 16.2 driver for 290's?

 

[guskline]

Did you had 16.2 driver for 290's?

Yes sir! and 361.91 for the GTX980TI SC.

 

[Bacon1]

There is a huge post over at wccftech, haven't read all of it so far:

http://wccftech.com/directx-12-async-compute-nvidia-amd/

 

[JDG1980]

Could be various reasons. One very possible reason is that people just don't want to, or find it hard, to believe that Hawaii grade GPUs are really as powerful as we're seeing.

A simple examination of the raw compute numbers indicates that AMD GPUs are more powerful than most DX11 gaming benchmarks indicate. (These numbers are available at TechPowerUp's GPU database.)

R9 390X with 5.9 TFlops is a more powerful card than GTX 980 with 4.6 TFlops. Fury X (8.6 TFlops) is more powerful than Titan X (6.1 TFlops). The mystery isn't why AotS shows these AMD cards surpassing their Nvidia counterparts. We should instead be wondering why they show such poor performance in DX11 titles. Clearly there are bottlenecks somewhere, whether in drivers, hardware, or an interaction between the two, that stops AMD from converting its extra hardware horsepower into FPS. It appears that async compute with DX12 breaks that bottleneck; the AotS benchmarks actually track the raw TFlop performance of the GPUs fairly closely.

 

[Mahigan]

A simple examination of the raw compute numbers indicates that AMD GPUs are more powerful than most DX11 gaming benchmarks indicate. (These numbers are available at TechPowerUp's GPU database.)

R9 390X with 5.9 TFlops is a more powerful card than GTX 980 with 4.6 TFlops. Fury X (8.6 TFlops) is more powerful than Titan X (6.1 TFlops). The mystery isn't why AotS shows these AMD cards surpassing their Nvidia counterparts. We should instead be wondering why they show such poor performance in DX11 titles. Clearly there are bottlenecks somewhere, whether in drivers, hardware, or an interaction between the two, that stops AMD from converting its extra hardware horsepower into FPS. It appears that async compute with DX12 breaks that bottleneck; the AotS benchmarks actually track the raw TFlop performance of the GPUs fairly closely.

Most games use about the compute equivalent of ~230 SPs per 4ROps/cycle. GCN3 is pegged at 256 SPs per 4ROps/cycle. (ATi study years ago when the Radeon HD 6970 was released).

So games haven't really tapped compute shaders much but a shift began around mid 2015. This is primarily why we've seen the R9 290x (4.3ns latency for a MADD) catch up to the 780 Ti (9-10ns latency for a MADD). The arithmetic latency on Kepler is bad. There were also utilization issues on Kepler which Maxwell resolves by mapping the 32 threads of a Warp to 32 SIMD cores per SMM quadrant with it's own Warp scheduler.

AotS pushes this further (as do console ports who have to tap into compute shaders in order to extract more performance out of the XBox One/PS4).

GCN in the consoles is placing emphasis on compute performance morso than before.

On top of that GCN is API bound on DX11. This is most likely due to the lack of a multi-threaded driver on AMDs part for translation of Command Lists into ISA as well as submission of commands to the command buffer.

DX12 alleviates these bottle necks.

 

[littleg]

On top of that GCN is API bound on DX11. This is most likely due to the lack of a multi-threaded driver on AMDs part for translation of Command Lists into ISA as well as submission of commands to the command buffer.

DX12 alleviates these bottle necks.

This is what I don't get though. If this has been hampering the performance of their GPUs in DX11 all this time and can be fixed in drivers then why haven't they done it?

 

[Mahigan]

This is what I don't get though. If this has been hampering the performance of their GPUs in DX11 all this time and can be fixed in drivers then why haven't they done it?

Probably not worth the investment. They threw everything into pushing for multi-threaded APIs and compelling developers to be responsible for multi threading.

Seems that maybe they can't afford the driver team that this would require.

 

[Paratus]

So will a lowly 7970Ghz edition catch a 780 in AoTS. I'm curious how much dx12 will help GCN1.0 cards. I haven't seen anyone test a 7970 yet.

 

[Bacon1]

Probably not worth the investment. They threw everything into pushing for multi-threaded APIs and compelling developers to be responsible for multi threading.

Seems that maybe they can't afford the driver team that this would require.

Not to mention instead of a "do it once for the whole API" approach of Mantle / DX12 / Vulkan, DX11 has to be done per game, thats where Nvidia gets its perf gains and has to do per game patches to optimize the overhead.

 

[Udgnim]

So will a lowly 7970Ghz edition catch a 780 in AoTS. I'm curious how much dx12 will help GCN1.0 cards. I haven't seen anyone test a 7970 yet.

http://www.anandtech.com/show/10067/ashes-of-the-singularity-revisited-beta/5

 

[airfathaaaaa]

Probably not worth the investment. They threw everything into pushing for multi-threaded APIs and compelling developers to be responsible for multi threading.

Seems that maybe they can't afford the driver team that this would require.

you know this is not really true

if you see the cash flow of amd there isnt any indication of them burning money away from the company which is weird considering how they manage the dx12 so far..(i think they just caught on the bandwagon of ms and intel and exploited it)

on the other hand nvidia has burned almost 60% of the 2015 revenue just like that which again is odd considering that they havent done something too

 

[flopper]

Probably not worth the investment. They threw everything into pushing for multi-threaded APIs and compelling developers to be responsible for multi threading.

Seems that maybe they can't afford the driver team that this would require.

and with dx12 coming with Polaris, make little sense to extend resources to dx11 crossfire etc. better prepare for the transition and be ready to go with zen.
I find single card works just fine with dx9/dx11 so I assume crossfire might be a tad an issue especially with titles that isnt triple AAA.

 

[TheELF]

and with dx12 coming with Polaris, make little sense to extend resources to dx11 crossfire etc. better prepare for the transition and be ready to go with zen.

Seeing how many games are still dx9 to this day it would make sense to them to improve for older versions of dx.
Even a lot of games that were announced for dx12 pulled out and went with dx11 or got pushed way back.

 

[Paratus]

http://www.anandtech.com/show/10067/ashes-of-the-singularity-revisited-beta/5

Ya I know. I was just hoping for a GHz edition vs 780. It doesn't look like the boost is enough for a regular 7970 to come close to a 780ti.

The 680 gets a nice boost which is somewhat surprising.

 

[Dribble]

and with dx12 coming with Polaris, make little sense to extend resources to dx11 crossfire etc. better prepare for the transition and be ready to go with zen.
I find single card works just fine with dx9/dx11 so I assume crossfire might be a tad an issue especially with titles that isnt triple AAA.

Makes no sense? All games out right now are DX11 or less, most games coming out are DX11 or less. Most people aren't playing the most recent releases anyway and aren't using high end cards so need all the performance they can get. Just because we have a beta or two that have DX12 doesn't mean that most people will be playing with DX12. Not optimising your driver for nearly all played games (~100% today) is one of many reasons why AMD is going bust.

The same people who accuse Nvidia of lots of evil, nerfing older cards etc are quite happy that AMD is basically nerfing all of their cards for all of today's games because some of tomorrow's games might run better. You know when DX12 really hits none of the current cards will be up to it and we'll all be recommending next gen cards anyway so basically everyone who owns an AMD card today is loosing out because AMD's driver team is not optimising for them, instead optimising for cards you haven't bought yet.

 

[Genx87]

Makes no sense? All games out right now are DX11 or less, most games coming out are DX11 or less. Most people aren't playing the most recent releases anyway and aren't using high end cards so need all the performance they can get. Just because we have a beta or two that have DX12 doesn't mean that most people will be playing with DX12. Not optimising your driver for nearly all played games (~100% today) is one of many reasons why AMD is going bust.

The same people who accuse Nvidia of lots of evil, nerfing older cards etc are quite happy that AMD is basically nerfing all of their cards for all of today's games because some of tomorrow's games might run better. You know when DX12 really hits none of the current cards will be up to it and we'll all be recommending next gen cards anyway so basically everyone who owns an AMD card today is loosing out because AMD's driver team is not optimising for them, instead optimising for cards you haven't bought yet.

Haha I had this argument about 2-3 weeks ago over the same topic. I think in this same thread. Somehow people are under the impression cards built for the DX11 era are worth arguing over DX12 performance when DX12 isnt mainstream. By the time DX12 is mainstream nobody is going to care about a 980 or Fury. But I own a 770 so according to some because an older card didnt hold up that is why I feel this way lol.

 

[PhonakV30]

So will a lowly 7970Ghz edition catch a 780 in AoTS. I'm curious how much dx12 will help GCN1.0 cards. I haven't seen anyone test a 7970 yet.

http://www.pcgameshardware.de/Far-Cry-Primal-Spiel-56176/Specials/Benchmark-Test-1187476/

Check here , surprising 280X is close to GTX 780! It's Far cry primal bench and perhaps new DX12 titles improve 280x Performance.

 

[TXBPNT]

http://www.pcgameshardware.de/Far-Cry-Primal-Spiel-56176/Specials/Benchmark-Test-1187476/

Check here , surprising 280X is close to GTX 780! It's Far cry primal bench and perhaps new DX12 titles improve 280x Performance.

R9 390 @ 1010 Mhz is 10% faster than GTX 970 @ 1300 Mhz in a Ubisoft title

 

[caswow]

R9 390 @ 1010 Mhz is 10% faster than GTX 970 @ 1300 Mhz in a Ubisoft title

no gameworks no cripple :thumbsup:

 

[MrTeal]

Did you download and install Radeon Hotfix 16.2?

Yeah, a couple times with DDU. I just did an inplace upgrade to Win10 though, so I'm not confident on anything working quite right at the moment. I hope to have time to do a clean install this coming weekend.

 

[guskline]

Yeah, a couple times with DDU. I just did an inplace upgrade to Win10 though, so I'm not confident on anything working quite right at the moment. I hope to have time to do a clean install this coming weekend.

Hopefully that helps

 

[Udgnim]

Ya I know. I was just hoping for a GHz edition vs 780. It doesn't look like the boost is enough for a regular 7970 to come close to a 780ti.

The 680 gets a nice boost which is somewhat surprising.

680 getting a boost from DX11 to DX12 likely means Nvidia not focusing driver optimizations on it

 

[Bacon1]

http://www.pcgameshardware.de/Far-Cry-Primal-Spiel-56176/Specials/Benchmark-Test-1187476/

Check here , surprising 280X is close to GTX 780! It's Far cry primal bench and perhaps new DX12 titles improve 280x Performance.

Isn't that only DX11?

Either way, shows AMD is right there with Nvidia perf wise.