VKD3D-Proton Has Been Working On Emulating D3D12 Work Graphs, But The Tech Disappoints

[phoronix]

Phoronix: VKD3D-Proton Has Been Working On Emulating D3D12 Work Graphs, But The Tech Disappoints

Back in March for GDC, Microsoft excitingly announced the official releases of Direct3D 12 Work Graphs for "enabling new types of GPU autonomy" for allowing more rendering work to be offloaded to the GPU. While this greater GPU-driven rendering with Work Graphs has been talked up by Microsoft and other parties, Valve engineers working on VKD3D-Proton for implementing D3D12 over Vulkan have found the new Work Graphs functionality to not be as nearly captivating...

https://www.phoronix.com/news/VKD3D-Proton-Work-Graphs

[Danny3]

Well, fuck DirectX 12!

Vullkan should | must be the future!

[V1tol]

Opensource developers created translation layer to be faster than proprietary native implementation. Again

[ms178]

This movie sounds familiar? Cooks cooking up new tech (e.g. Rambus, Optane), that tech fails more or less spectacularly and the cycle repeats itself until something eventually succeeds.

[Daktyl198]

Opensource developers created translation layer to be faster than proprietary native implementation. Again

It's not that the translation layer is faster, they're saying that the feature doesn't deliver any performance wins even over regular DX12. Also, Wine/Proton/DXVK are not faster than native. If you do see a better frame rate on Linux than on Windows, it's most likely due to the better memory management and (if you're using Intel or AMD gpus) drivers. Gallium Nine, to this day, is faster and more accurate than DXVK at rendering D3D9 games.

That's not to crap on the work done by DXVK and all... it's just a truth. I still say Linux drivers should natively implement D3D9-12.

[Daktyl198]

This movie sounds familiar? Cooks cooking up new tech (e.g. Rambus, Optane), that tech fails more or less spectacularly and the cycle repeats itself until something eventually succeeds.

Optane was actually a really cool idea and solid tech... but it came to the market too late. Just a couple years after it's release, the entire point for it's existence (to act as a read/write cache for spinning HDDs) was overshadowed by SSDs becoming extremely cheap.

[mobadboy]

This movie sounds familiar? Cooks cooking up new tech (e.g. Rambus, Optane), that tech fails more or less spectacularly and the cycle repeats itself until something eventually succeeds.

mantle was the biggest failure i can remember. it was the biggest "thing", which meant nobody used it, and then it disappeared. which makes sense because amd created it

[finalzone]

mantle was the biggest failure i can remember. it was the biggest "thing", which meant nobody used it, and then it disappeared. which makes sense because amd created it

Vulkan is Mantle reborn as AMD donated the foundation to Khronos

[Daktyl198]

mantle was the biggest failure i can remember. it was the biggest "thing", which meant nobody used it, and then it disappeared. which makes sense because amd created it

Dice used Mantle on Battlefield 4 (a HUGE and popular game), and it proved just how useful a low-level API like it could be, giving AMD cards a large performance uplift. Then AMD gave Mantle to Khronos who turned it into Vulkan. But yeah... "failure" lol.

[jams3223]

It's not that the translation layer is faster, they're saying that the feature doesn't deliver any performance wins even over regular DX12. Also, Wine/Proton/DXVK are not faster than native. If you do see a better frame rate on Linux than on Windows, it's most likely due to the better memory management and (if you're using Intel or AMD gpus) drivers. Gallium Nine, to this day, is faster and more accurate than DXVK at rendering D3D9 games.

That's not to crap on the work done by DXVK and all... it's just a truth. I still say Linux drivers should natively implement D3D9-12.

Where the heck did you get that information from? Wait a minute, did you think DXVK and VKD3D-Proton were emulators? lol
DXVK and VKD3D-Proton function as mappers, not emulators. While they may offer improved memory management, the mapping process does not introduce any overhead. In my experience, no working mapper has demonstrated any overhead. This scenario is similar to asserting that if I define a constant variable named 'five' with a value of 5 and subsequently assign 'five' to another variable named 'cinco' under certain conditions, it does not follow that 'five' is equivalent to 'cinco' and 'cinco is also equal to the value 5.

Take you're time when reading.

https://github.com/HansKristian-Work/vkd3d-proton/blob/workgraphs/docs/workgraphs.md