No DX12 Software is Suitable for Benchmarking *spawn*

[Scott_Arm]

Another one: https://www.overclock3d.net/reviews...d_-_a_major_boost_for_modern_graphics_cards/1

Doesn't look like a big boost, I've expected more actually. But at least it isn't slower than D3D11 anywhere.

Turing, Vega and RDNA all seem to get pretty nice bumps to me, but I guess that's just a difference of expectations. Pascal can't take advantage of the async compute changes they made.

Edit:
https://www.ubisoft.com/en-ca/game/rainbow-six/siege/news-updates/6VFn74oMO2nVQGZxvFhBb5

The team experimented with dx12 but found better utilization on the cpu with vulkan.

 

[techuse]

Sigh, I guess you are right, so we can scratch Tomb Raider off the list.

We now only have Hitman 2, Control and Sniper Elite 4, maybe Metro Exodus to some degree.


Tracking most DX12 titles in this thread, I would say AMD hardware also tend to either show no gains from DX12, or show very limited gains. There are cases where DX12 gives AMD hardware huge gains, but that's because their DX11 performance where atrocious to begin with in these cases, so DX12 only acts as the buffer that propels AMD hardware to near NVIDIA DX11 performance.


I will take the opportunity to revisit this subject again in the context of DX12, based on the latest benchmarks form Computerbase, it appears that Metro Exodus is still faster on Pascal than Vega.

https://www.computerbase.de/2019-12...cours-2020/2/#diagramm-metro-exodus-2560-1440
https://gamegpu.com/action-/-fps-/-tps/metro-exodus-test-gpu-cpu-2019

Shadow Of Tomb Raider is a tie.
https://www.computerbase.de/2019-12...#diagramm-shadow-of-the-tomb-raider-2560-1440

Control is a tie as well.
https://www.computerbase.de/2019-12/grafikkarten-testparcours-2020/2/#diagramm-control-2560-1440

Just another datapoint via HardwareUnboxed. 1060 9 GBPS vs 580.

 

[Silent_Buddha]

Just another datapoint via HardwareUnboxed. 1060 9 GBPS vs 580.

Dang. Imagine how much worse that would be if someone had been suckered into buying the 3 GB version of the 1060.

Regards,
SB

 

[Malo]

Just another datapoint via HardwareUnboxed. 1060 9 GBPS vs 580.

That's impressive longevity for Polaris on low-level APIs. Note that you won't ever sway the Nvidia marketing crew.

 

[techuse]

That's impressive longevity for Polaris on low-level APIs. Note that you won't ever sway the Nvidia marketing crew.

I just enjoy the discussion.

 

[DegustatoR]

That's impressive longevity for Polaris on low-level APIs. Note that you won't ever sway the Nvidia marketing crew.

They should've included a 4GB 570. Chances are that 1060 6G's main issue in D3D12 and VK are the devs who just don't bother with proper optimizations for an "odd" 6GB framebuffer.

 

[Scott_Arm]

They should've included a 4GB 570. Chances are that 1060 6G's main issue in D3D12 and VK are the devs who just don't bother with proper optimizations for an "odd" 6GB framebuffer.

I think the lack of async compute is probably one of the big culprits. All of the id software games disable async compute for Nvidia pascal in Vulkan. R6 Siege doesn't see big gains with Pascal using vulkan, most likely because async compute is one of the bigger gains on the gpu side.

 

[DegustatoR]

I think the lack of async compute is probably one of the big culprits. All of the id software games disable async compute for Nvidia pascal in Vulkan. R6 Siege doesn't see big gains with Pascal using vulkan, most likely because async compute is one of the bigger gains on the gpu side.

Turing has "async compute" (technically, all GPUs with D3D12 drivers have it; but I assume that you mean "async compute with the same h/w execution approach as in GCN") and doesn't fare any better than Pascal in any of such benchmarks. There's not a lot you can gain with it in an architecture which is able to saturate its processing units within the graphics context alone. So while this can explain the lack of gains on 1060 in D3D12/VK when compared against a GCN GPU, it certainly can't explain the loss of performance when compared to D3D11.

 

[Scott_Arm]

Turing has "async compute" (technically, all GPUs with D3D12 drivers have it; but I assume that you mean "async compute with the same h/w execution approach as in GCN") and doesn't fare any better than Pascal in any of such benchmarks. There's not a lot you can gain with it in an architecture which is able to saturate its processing units within the graphics context alone. So while this can explain the lack of gains on 1060 in D3D12/VK when compared against a GCN GPU, it certainly can't explain the loss of performance when compared to D3D11.

Turing GPUs get a very nice performance boost in Vulkan on R6 Siege, especially minimum frame times, where Pascal GPUs do not. The main performance enhancement they've added on the GPU side, which was not available in DX11 is async compute.

 

[DegustatoR]

Turing GPUs get a very nice performance boost in Vulkan on R6 Siege, especially minimum frame times, where Pascal GPUs do not. The main performance enhancement they've added on the GPU side, which was not available in DX11 is async compute.

That's down to implementation details in R6S specifically. Overall there is no apparent difference between Pascal and Turing in how they perform under D3D12/VK.

 

[techuse]

That's down to implementation details in R6S specifically. Overall there is no apparent difference between Pascal and Turing in how they perform under D3D12/VK.

I recall seeing AC toggle-able benchmarks that show a benefit for Turing where Pascal tends to lose a fraction of performance. Its a smaller benefit than GCN but its a few percent nonetheless.

To clarify im not referring to R6 benchmarks.

 

[Scott_Arm]

That's down to implementation details in R6S specifically. Overall there is no apparent difference between Pascal and Turing in how they perform under D3D12/VK.

But implementation is what APIs are for. The implementation is what an API allows you to do. R6 siege was able to implement async compute in Vulkan, where it could not with DX11. It is a feature of Vulkan that DX11 does not have. I'm not sure how you can compare performance if you do not allow for differences in implementation.

 

[trinibwoy]

Took advantage of the $3 deal on Division 2 and have about 8 hours of playtime.

6 crashes on DX12 so far and none on DX11.

GTX 1080, Windows 10 latest drivers.

 

[Malo]

6 crashes on DX12 so far and none on DX11.

So they still didn't fix the DX12 crashes eh?

 

[BRiT]

So they still didn't fix the DX12 crashes eh?

It's a feature of their game to make sure you don't game too long.

 

[Deleted member 2197]

Wolfenstein: Youngblood Benchmarked: GeForce & Radeon Battle It Out
February 28, 2020

But what about Wolfenstein: Youngblood benchmarked without RTX and DLSS technologies enabled? That's something I spent the better part of 2-3 days benchmarking and testing for. I added in a metric ton of graphics cards into the mix, and since there's no RTX and DLSS features here to test -- I was able to add in a bunch of AMD Radeon graphics cards into the testing pot.

https://www.tweaktown.com/articles/...arked-geforce-radeon-battle-it-out/index.html

 

[Deleted member 2197]

Basemark GPU v1.2 benchmarks with 22 GPUs
March 4, 2020
There is a refreshed benchmark in town, fabbed and founded by ex-Futuremark employees BasemarkGPU v1.2 has been released. GPU is uniquely a multi-platform benchmark. It is a product enabling direct comparisons e.g. between Android smartphones and iPhones, Windows, Linux and MacOS desktops and notebooks as well as APIs, Metal, Vulkan, DirectX, OpenGL and OpenGL ES.
...
Basemark GPU is an evaluation tool to analyze and measure graphics API (OpenGL 4.5, OpenGL ES 3.1, Vulkan 1.0, Microsoft DirectX 12 and Apple Metal 2) performance across mobile and desktop platforms. Performance is measured using an advanced, game-like workload. Basemark GPU as such Basemarks GPU targets to both Desktop and Mobile platforms by providing High Quality, Medium Quality and Simple Quality modes.

https://www.guru3d.com/articles_pages/basemarkgpu_benchmark_review_with_20_gpus,1.html

Download (free):
https://www.basemark.com/benchmarks/basemark-gpu/

 

[Silent_Buddha]

I find it amusing that it's ranked using the API most favorable to NV instead of the most used API for games or alphabetically (API).

Regards,
SB

 

[Deleted member 2197]

I find it amusing that it's ranked using the API most favorable to NV instead of the most used API for games or alphabetically (API).
Regards,
SB

If you follow the link they do have the API's charted separately. They just combined those charts in one with all API's.

 

[CarstenS]

I find it amusing that it's ranked using the API most favorable to NV instead of the most used API for games or alphabetically (API).

Considering the relative scores, that'd be OpenGL. But yes, when Basemark promises a cross-API Benchmark, they should have sorted that graph according to best result for each individual card.