Intel Is Making A High-Performance Software Rasterizer For Mesa

[phoronix]

Phoronix: Intel Is Making A High-Performance Software Rasterizer For Mesa

While Mesa currently has the swrast, LLVMpipe, and Softpipe drivers as software rasterizers that run OpenGL on the CPU rather than any dedicated GPU, a team at Intel has been developing a new, high-performance software rasterizer. This Intel team hopes to upstream their new "OpenSWR" project into Mesa as offering fast, CPU-rendered graphics...

http://www.phoronix.com/scan.php?page=news_item&px=Intel-OpenSWR-Mesa

[chris200x9]

I can't wait until I can play SC on my CPU!

[valeriodean]

I can't wait until I can play SC on my CPU!

I fear you will wait forever then, it is not for games but for scientific workload. :-)

[Michael]

Been trying to get it to build for a half hour now, but keep running into LLVM errors. Tried with Ubuntu's LLVM 3.6.2 and then build LLVM 3.6.0 from source, but no go.

[caligula]

Sounds good. I've been curious to know just how fast a CPU can render stuff compared to GPUs. It's obvious that sw rasterization code has been suboptimal. Even very basic things seem slow on Mesa software renderer (on overclocked 4.0 GHz+ Intel 6700K and 32 GB DDR4) while games like Quake run just fine on Pentium MMX 133 MHz and 33? MHz RAM.

[whitecat]

it is not for games but for scientific workload. :-)

Why not using dedicated GPU ?

[swoorup]

Why not using dedicated GPU ?

I suppose this is useful for high core CPU's, when using GPU's they would be idling anyway.

[blubbaer]

Why not using dedicated GPU ?

Some scientific use cases require texture and framebuffer sizes that no current gpu supports in hardware or if it supports then with reduced performance. Medical imaging techniques are a good example. In the past FPGA were used to preprocess the data.

[GreatEmerald]

Hm, scientific workloads, and here I was hoping it would allow doing something useful with Intel's PowerVR generation hardware.

[caligula]

Some scientific use cases require texture and framebuffer sizes that no current gpu supports in hardware or if it supports then with reduced performance. Medical imaging techniques are a good example. In the past FPGA were used to preprocess the data.

I'd guess it also simplifies other issues with memory systems. Direct NUMA support, no need to swap data between GPU & CPU RAM.