LLVMpipe Still Doesn't Work For Linux Gaming

For those curious what OpenGL gaming frame-rates are like if trying to run LLVMpipe on the latest Intel Ivy Bridge processors, here are some numbers.

LLVMpipe is the Gallium3D-based software-accelerated driver fall-back that does OpenGL on the CPU. LLVMpipe leverages the LLVM compiler infrastructure to take advantage of multiple CPU cores, the latest CPU instruction sets, and other capabilities. However, running OpenGL on the CPU is still no match to a dedicated GPU; CPUs and GPUs are very different beasts and to each his own. Using LLVMpipe is good enough for a composited desktop environment (GNOME Shell, Unity, etc), but for games it's far from being sufficient. On the desktop side, LLVMpipe works surprisingly well for thin clients / multi-seats.

While OpenGL games over LLVMpipe is slow as molasses, it's also interesting to see how well LLVMpipe is doing for OpenGL gaming on the processor to measure CPU improvements, Mesa/Gallium3D architectural improvements / optimizations, and how well LLVM is doing.

I've done LLVMpipe on AMD Bulldozer and Intel Sandy Bridge, but here are some benchmarks when using LLVMpipe on Intel Ivy Bridge.

For this quick round of benchmarking is an Intel Core i7 3770K Ivy Bridge processor running Ubuntu 12.04 LTS x86_64 and then dropping in the Linux 3.4 kernel, LLVM 3.0, and Mesa 8.1-devel with LLVMpipe from Git. Being compared is the LLVMpipe performance to the Intel Mesa IVB graphics driver using the same software packages.

To no real surprise, LLVMpipe still won't work well for OpenGL gaming even with the latest high-end Intel processor. At least though it's good enough for handling modern desktops like GNOME Shell and Unity.

You can compare your system's performance to these LLVMpipe and Intel IVB numbers by simply running phoronix-test-suite benchmark 1205289-SU-LLVMPIPEI36. Additional system information and other details for this quick round of Intel Linux Ivy Bridge benchmarks can be found from the OpenBenchmarking.org result file.