LLVMpipe Tessellation Shader Support Is Now Working - Runs Unigine Heaven

[phoronix]

Phoronix: LLVMpipe Tessellation Shader Support Is Now Working - Runs Unigine Heaven

The LLVMpipe CPU-based software rasterizer OpenGL driver within Mesa's Gallium3D now has working tessellation shader support (ARB_tessellation_shader) and can even run Unigine Heaven demo properly, just don't expect good performance...

http://www.phoronix.com/scan.php?page=news_item&px=LLVMpipe-Tessellation-Working

[baka0815]

Benchmarks or it didn't happen.

[uid313]

While LLVMpipe is a CPU-based software rastertizer for OpenGL, is there any CPU-based software rasterizer for Vulkan?

[Jabberwocky]

Tessellation is a big one!

While LLVMpipe is a CPU-based software rastertizer for OpenGL, is there any CPU-based software rasterizer for Vulkan?

https://salsa.debian.org/Kazan-team/kazan

[remenic]

Will we get to see benchmarks of LLVMpipe blasting pixels at the screen on a TR3990X soon?

[Gusar]

Will we get to see benchmarks of LLVMpipe blasting pixels at the screen on a TR3990X soon?

llvmpipe tops out at 8 (or was it 16?) threads, so you won't gain anything by running it on a core-heavy CPU. It'll quite likely run best on an i9-9900K.

[Veerappan]

llvmpipe tops out at 8 (or was it 16?) threads, so you won't gain anything by running it on a core-heavy CPU. It'll quite likely run best on an i9-9900K.

I seem to remember that being one of the selling points of SWR when it was merged. It scales to higher core counts than LLVMPipe.

[smitty3268]

I seem to remember that being one of the selling points of SWR when it was merged. It scales to higher core counts than LLVMPipe.

SWR scales better with high geometry complexity, which can be important for certain scenarios like scientific visualizations. It's not that great at pushing pixels or fragment shaders, meaning typical gaming scenarios are probably faster on llvmpipe.

[tajjada]

What exactly is stopping llvmpipe from scaling to more threads?

[chinoto]

What exactly is stopping llvmpipe from scaling to more threads?

Lack of a motivated developer and probably the hardware to verify any benefit. The few people* that would benefit from further scaling probably have a dedicated GPU anyway. Maybe further scaling could help run tests faster? Then again, you could just run tests in parallel to utilize all cores.
* Steam hardware survey for core count: https://store.steampowered.com/hwsurvey/cpus/