Pascal now supports DX12 resource binding tier 3 with latest drivers (384.76)

Page 6 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Face2Face

Diamond Member
Jun 6, 2001
4,100
215
106
Never mind, looks like it was already done :)

Forza 6 can't run on Fermi. It needs resource binding tier 2. At the moment it's the DX12 game with the highest feature requirements. https://support.xbox.com/en-GB/games/troubleshooting/error-code-ap203

I figured as much. I was able to download it, but it would never start. Now I can't even download it because it shows my video card not supported.
Not sure how the Youtuber got it running? Could be fake...who knows?
 

TheELF

Diamond Member
Dec 22, 2012
3,973
730
126
No, its below because it was not designed to support DX12. Sure you can patch in some things through software, but its obvious that the performance is going to be dismal, since the architecture is not optimized for DX12 API, those cards are actually DX10 cards optimized.
Every game is made for consoles,anything the consoles GPUs don't support will be patched in for the windows version so the performance is going to be dismal for years to come.
On the other hand the driver overhead being reduced is a real thing that we can already see in a lot,if not all, dx12/vulkan games.
 

bononos

Diamond Member
Aug 21, 2011
3,886
156
106
Performance is below dx11 only because you are running it on a i7 where the dx11 driver thread doesn't interfere with running the game/bench try it again on 2 or 4 cores and you will see the benefit.

I seriously doubt it. The constant hitching stuttering and glitches in dx12 is horrible. I doubt Fermi's dx11 even on an i3 cpu behaves as badly or worse to make the new dx12 worthwhile.
 

Carfax83

Diamond Member
Nov 1, 2010
6,841
1,536
136
Why does every example of your's revolve around Ubisoft titles?

What's wrong with that? Ubisoft's engines, ie AnvilNext and Disrupt tend to be among the highest threaded engines out there.

OpenGL optimizations in AMD's driver have been pretty much non-existent. The reason why AMD gains so much in Doom Vulkan is the opposite of why according to you NVIDIA's driver gains so little in DX12 compared to DX11.

Yes, AMD's OpenGL drivers are even more "terribad" than their DX11 drivers, but OpenGL itself is much to blame as well. OpenGL is considerably inferior to DX11 when it comes to CPU performance, lacking true multithreading capability.

Remember my example that I posted about Doom on my own PC? About 160% increase in performance using Vulkan compared to OpenGL 4.5 in that particular area.

Where's the GPU utilization graphs?

Why would I need GPU utilization graphs? This should be axiomatic. Higher framerates mean higher utilization.

No, it was because it was a fundamentally unbalanced design. Hawaii was the most balanced high-end GCN chip.

So you're a GPU designer now? I have no idea what you consider balanced vs unbalanced, but Vega seems to have the same ratio of shaders to ROPS and TMUs that Fiji has, so the actual AMD engineers seem to not think that ratio is unbalanced.

390X 9% faster than 390, while having 10% more SPs. Fury X only 5% faster than Fury, in spite of having 14% more SPs.

Unless you have some kind of profiling software, it's impossible to say with granite certainty that this is because of the hardware design itself, or the software that's driving the hardware.

Prey is DX11, and so is Dirt 4. Do we see this same behavior under DX12 and Vulkan titles?

Sniper Elite 4
Ashes of the Singularity beta 2

In both titles, Fiji has no problem outperforming Hawaii and gains more than Hawaii with DX12 and Async, particularity in Ashes of the Singularity. It appears to me that Fiji isn't necessarily unbalanced, so much that it seems to require a low level API to achieve high utilization.

Which goes back to my original point. DX12 and Vulkan's main performance enhancement are parallel rendering and much lower CPU overhead.
 

Carfax83

Diamond Member
Nov 1, 2010
6,841
1,536
136
I think that's a bad term if AMD's driver is single threaded,nvidia in this example uses ~50% of the cpu,if it where single threaded it would be constricted to 25% max which would be less overhead not more.

No, I don't believe that's the case. When I say overhead, I mean that the workload isn't being distributed across the entire CPU, and one thread has to carry most of the burden. So while NVidia's driver might consume more of the overall CPU, the workload is more evenly distributed. Whereas with AMD, one core is slammed, resulting in lower GPU utilization and less performance.

Oh and also...

In dx11 the driver consumes ~14% of the CPU time leaving the highest division thread at~12%
in dx12 the driver is gone leaving the highest division thread running at ~17%
= profit
Dx12 is made for the little man (consoles) never forget that,don't look at test done on super high core count high Ghz CPUs...

Yes, but you're also forgetting you're getting a 16% boost in performance with DX12 vs DX11. BTW is that with The Division 1.6.1?
 

TheELF

Diamond Member
Dec 22, 2012
3,973
730
126
Yes, but you're also forgetting you're getting a 16% boost in performance with DX12 vs DX11. BTW is that with The Division 1.6.1?
That's what I'm saying,dx12 gives you a good boost if your CPU is forced to run the driver thread(s) on the same cores as the game threads,on high core count cpus the driver thread(s) don't interfere with the game threads so minimal if any gains.

It's whatever version it had on the last free weekend about a month ago.
 

Carfax83

Diamond Member
Nov 1, 2010
6,841
1,536
136
That's what I'm saying,dx12 gives you a good boost if your CPU is forced to run the driver thread(s) on the same cores as the game threads,on high core count cpus the driver thread(s) don't interfere with the game threads so minimal if any gains.

That's a bit too simplistic. It doesn't take into account the additional parallel scaling that DX12 affords over DX11, without the additional overhead. While I agree that very fast, and very high core count CPUs can mitigate a lot of the drawbacks of DX11, DX12 is on a whole other level regardless.

The reason why the gains aren't there yet, is because we are still in the transitional phase between DX11 and DX12. When games fully transition over to DX12, games will have detail levels that are much greater than anything out there now while still being performant.

It's whatever version it had on the last free weekend about a month ago.

The reason why I was asking is because The Division recently updated to version 1.6.1 which increased DX12 performance substantially. At 1440p near maxed settings, I get 8% higher framerates in DX12 compared to DX11. Might not seem like a lot, but I am not CPU bound at all in DX11, so most of that improvement is coming from things like multiengine and shader model 5.1.
 

tamz_msc

Diamond Member
Jan 5, 2017
3,770
3,590
136
What's wrong with that? Ubisoft's engines, ie AnvilNext and Disrupt tend to be among the highest threaded engines out there.
Ubisoft titles have performance issues even on consoles. They prioritized CPU performance for this generation, despite the Jaguar cores being weak for the kind of world simulation they'd aimed for, and as a result you had things like AC:Unity which was a hot mess at launch. Moreover, on the GPU front, no matter how much advancements they claim they've made, their game engines have made little progress in performing well on AMD hardware. Except perhaps R6 Siege, Ubisoft has the poorest track record when it comes to good AAA performance in this generation of consoles.
Why would I need GPU utilization graphs? This should be axiomatic. Higher framerates mean higher utilization.
Do you play CS:GO? You seem to be forgetting about the role the CPU plays at lower resolutions.
So you're a GPU designer now? I have no idea what you consider balanced vs unbalanced, but Vega seems to have the same ratio of shaders to ROPS and TMUs that Fiji has, so the actual AMD engineers seem to not think that ratio is unbalanced.
If everything is the same as Fiji, of what use is this then?
Vega%20Final%20Presentation-25.png


Why are people so eager to know if Vega has four Shader Engines, or eight(based on the die shot mock-up), for example, if everything is the same as Fiji?

Unless you have some kind of profiling software, it's impossible to say with granite certainty that this is because of the hardware design itself, or the software that's driving the hardware.

Prey is DX11, and so is Dirt 4. Do we see this same behavior under DX12 and Vulkan titles?

Sniper Elite 4
Ashes of the Singularity beta 2

In both titles, Fiji has no problem outperforming Hawaii and gains more than Hawaii with DX12 and Async, particularity in Ashes of the Singularity. It appears to me that Fiji isn't necessarily unbalanced, so much that it seems to require a low level API to achieve high utilization.

Which goes back to my original point. DX12 and Vulkan's main performance enhancement are parallel rendering and much lower CPU overhead
The drivers can't extract the same level of performance from Fury X to Fury, which they can from 390X to 390, in the most well-optimized CryEngine game at the moment, and the best performing multi-platform racing game. Do you believe that it's just a software scaling issue, or a hardware scaling issue? What you say is certainly true, that a detailed analysis is not possible by just comparing FPS numbers, but I'm inclined to believe that it's the latter. There should be no reason why a Fury X should fall behind to RX 480 levels in certain instances, if there were no problems with it in terms of hardware.

It is precisely because of the fact that you have to resort to low level APIs and architecture-specific optimizations to get the best out of Fiji that I say it's unbalanced, which is the general consensus if you lurk around at B3D. Why do you think Fiji improves in performance relative to its direct competitor, the 980Ti, as you increase the resolution? It's because it becomes less geometry bound and more pixel bound.
 

bononos

Diamond Member
Aug 21, 2011
3,886
156
106
Completely different point though.
No it goes with what I said - that dx12 on Fermi looks terrible with all the herky jerky stuttering and glitches compared to dx11. Unless you are saying that dx11 on Fermi with an i3 suffers even worse stuttering and glitches compared to dx12 so dx12 is better, which we know isn't true.
 

Carfax83

Diamond Member
Nov 1, 2010
6,841
1,536
136
No it goes with what I said - that dx12 on Fermi looks terrible with all the herky jerky stuttering and glitches compared to dx11. Unless you are saying that dx11 on Fermi with an i3 suffers even worse stuttering and glitches compared to dx12 so dx12 is better, which we know isn't true.

Might be because DX12_0 requires resource binding tier 2, and Fermi only supports resource binding tier 1.
 

Carfax83

Diamond Member
Nov 1, 2010
6,841
1,536
136
Ubisoft titles have performance issues even on consoles. They prioritized CPU performance for this generation, despite the Jaguar cores being weak for the kind of world simulation they'd aimed for, and as a result you had things like AC:Unity which was a hot mess at launch. Moreover, on the GPU front, no matter how much advancements they claim they've made, their game engines have made little progress in performing well on AMD hardware. Except perhaps R6 Siege, Ubisoft has the poorest track record when it comes to good AAA performance in this generation of consoles.

Honestly, could care less about the consoles. The CPUs in those things are anemic pieces of crap that shouldn't be anywhere near anything to do with serious gaming. That said, another reason for games like AC Unity not performing well on AMD hardware is because those games have very high levels of geometry, which is AMD's weak point compared to NVidia.

Do you play CS:GO? You seem to be forgetting about the role the CPU plays at lower resolutions.

No I don't, but the point is, a GPU's load shouldn't necessarily decrease with resolution provided that the CPU can continue to increase the framerate to very high levels.

If everything is the same as Fiji, of what use is this then?

Why are people so eager to know if Vega has four Shader Engines, or eight(based on the die shot mock-up), for example, if everything is the same as Fiji?

OK I see what you mean. I just found it coincidental that Vega has the same ratio of shaders to ROPS and TMUs that Fiji has.

The drivers can't extract the same level of performance from Fury X to Fury, which they can from 390X to 390, in the most well-optimized CryEngine game at the moment, and the best performing multi-platform racing game. Do you believe that it's just a software scaling issue, or a hardware scaling issue? What you say is certainly true, that a detailed analysis is not possible by just comparing FPS numbers, but I'm inclined to believe that it's the latter.

To me it's a software scaling issue, caused by the drivers and the DX11 API. Dirt 4 like many DX11 games seems to hammer the lead thread, while the others have much smaller loads. This sort of behavior hurts AMD more than NVidia because AMD's drivers have higher overhead. That said, the benchmarks from that website you quoted from seem to conflict with reports from other websites like PCgameshardware.de, which show a larger spread between Fury X and other AMD cards.

There should be no reason why a Fury X should fall behind to RX 480 levels in certain instances, if there were no problems with it in terms of hardware.

The only reason why Fury X would ever fall behind the RX 480 would be in very high geometry workloads. That's really what AMD's weakness is if anything, and why they underperform relative to NVidia in massive open world games such as what Ubisoft currently favors.

Assassin's Creed Origin might be DX12, so it will be interesting to see how it performs on AMD hardware.

It is precisely because of the fact that you have to resort to low level APIs and architecture-specific optimizations to get the best out of Fiji that I say it's unbalanced, which is the general consensus if you lurk around at B3D. Why do you think Fiji improves in performance relative to its direct competitor, the 980Ti, as you increase the resolution? It's because it becomes less geometry bound and more pixel bound.

I agree with you here. If anything is unbalanced about Fiji (or AMD cards in general really), its the lack of geometry prowess. Unfortunately I don't think that Vega will really remedy this issue, judging by the benchmarks that PCgameshardware.de recently did of the Vega Frontier model. In AC Syndicate (which has very high levels of geometry), the performance is around that of a GTX 1070. RX Vega will be faster, but it will likely not catch up to a GTX 1080 in these kinds of workloads.
 

tamz_msc

Diamond Member
Jan 5, 2017
3,770
3,590
136
Honestly, could care less about the consoles. The CPUs in those things are anemic pieces of crap that shouldn't be anywhere near anything to do with serious gaming. That said, another reason for games like AC Unity not performing well on AMD hardware is because those games have very high levels of geometry, which is AMD's weak point compared to NVidia.
Yet in most of the discussions you make it a point to cite Ubisoft titles first, which perform extremely poorly on consoles. Keeping the GPU aspect of the performance aside, the reason why AC Unity performs so poorly is also because it renders tons of NPCs, whose AI and animations are primarily handled by the CPU.
That said, the benchmarks from that website you quoted from seem to conflict with reports from other websites like PCgameshardware.de, which show a larger spread between Fury X and other AMD cards.
The numbers are from pclab.pl, whose results match very well with those of PCGH.de, especially in the case of Prey.