Gallium3D's LLVMpipe Is Speeding Up

[elanthis]

Don't be fooled by that i7. It's a mobile part with about the same performance as an Athlon II X4.

People with an Athlon II X4 are just as unlikely to lack a video chipset that's unsupported by the current FOSS drivers.

So far as I can tell, there are three groups of people with an interest in llvmpipe. (1) People with newer technology the FOSS drivers haven't caught up to yet, mostly ATI customers; the ATI driver lag time is shrinking and looking fairly good right now for basic 3D accel. (2) People with ancient hardware by defunct manufacturers that don't want to pay $50 for modern-ish AGP card; screw 'em, these be computers, keep up or shut up. (3) People with oddball hardware, mostly netbooks and tablets and the like, that come with video chipsets not designed or supported by one of the Big Three; usually very weak CPUs on these puppies.

From what I've seen, the biggest group of people wanting/needing something like llvmpipe were the ATI customers with r600/r700/r800 parts. r600/r700 is pretty solid these days, and r800 is wrapping up, and will probably be ready to go around the time the 3D desktops are more than just an upcoming plan.

I am totally ready to be told I've gotten it all wrong here, though.

[smitty3268]

From what I've seen, the biggest group of people wanting/needing something like llvmpipe were the ATI customers with r600/r700/r800 parts. r600/r700 is pretty solid these days, and r800 is wrapping up, and will probably be ready to go around the time the 3D desktops are more than just an upcoming plan.

I am totally ready to be told I've gotten it all wrong here, though.

You're forgetting nouveau users. The lag time on new hardware support is unlikely to get much better while NVidia keeps the specs private, and even a lot of older hardware can be buggy. Most people who need 3D support will just go with the binary drivers if nouveau isn't cutting it, but some might want to experiment with llvmpipe instead.

[baryluk]

Nice. I would like to see a CPU usage of all cores/threads, to see how well it paralellizes now (I know some time ago there were scalability testes of llvmpipe by enabling only some cores on CPU). I would guess that 500-600% load level will mean that llvmpipe essentially uses cpu perfectly.

If you have sandybridge still on desk, how about testing llvmpipe on it?