LLVMpipe Performance On AMD Bulldozer, Intel Sandy Bridge

Written by Michael Larabel in Processors on 4 November 2011 at 11:52 AM EDT. Page 1 of 2. 7 Comments.

With the renewed interest in the Gallium3D LLVMpipe driver now that this software-based acceleration method is working with GNOME Shell, here are some benchmarks of this LLVM-based software driver when running some OpenGL tests on Intel Sandy Bridge and AMD Bulldozer hardware.

This article just provides two quick OpenGL game tests under Linux on LLVMpipe from two systems. These are just the unpublished results I have handy when recently comparing the FX-8150 and Core i7 2630QM hardware, since right now I am still in Orlando at the Ubuntu Developer Summit. With the two games, a variety of results were tested on each platform. I also have more LLVMpipe benchmarks from late September, with having been benchmarking this interesting Gallium3D driver for more than one year when it became to be useful (see that article for my original overview). LLVMpipe offers faster performance than other software-based OpenGL acceleration methods due in large part to its use of LLVM (the Low-Level Virtual Machine) and being able to take advantage of multiple compute cores and various instruction set extensions.

The Bulldozer setup had the AMD FX-8150 eight-core system running at 3.6GHz on the ASUS Crosshair V Formula motherboard while the Sandy Bridge hardware was a Core i7 2630QM (four core plus Hyper Threading at 2.00GHz) from a ZaReason notebook. Both systems had Ubuntu 11.10 (x86_64) running with Mesa 7.12-devel from 24 October on LLVMpipe.


Related Articles