Gallium3D LLVMpipe Driver Shows Progress

[phoronix]

Phoronix: Gallium3D LLVMpipe Driver Shows Progress

With talking recently about LLVMpipe driver improvements and having not benchmarked this Gallium3D software driver in a while, here are new benchmarks of this LLVM-based software fallback driver when using Mesa 9.1-devel Git in conjunction with LLVM 3.3 SVN code, for the very latest look at the OpenGL software acceleration possibilities...

http://www.phoronix.com/vr.php?view=MTMxNDI

[LinuxID10T]

What on earth is the problem with the 6450 in all these tests? I don't only mean THIS one but it seems that in all of them, it does horrifically.

[log0]

What on earth is the problem with the 6450 in all these tests? I don't only mean THIS one but it seems that in all of them, it does horrifically.

Weird isn't it? Maybe it is a 64bit DDR3 variant. But still...

[schmidtbag]

Weird isn't it? Maybe it is a 64bit DDR3 variant. But still...

I highly dout thats the reason, especially when you consider the CPU is also DDR3, except maybe 128bit. Im sure if that gpu had properly functioning catalyst drivers it would outperform all the others.

[smitty3268]

I highly dout thats the reason, especially when you consider the CPU is also DDR3, except maybe 128bit. Im sure if that gpu had properly functioning catalyst drivers it would outperform all the others.

It's a darn slow part. However, i wonder if it's clocked at full speed. The radeon drivers just use whatever is set as the default in the BIOS, and Michael isn't reclocking it to HIGH mode.

I think some 6450's are passively cooled, so it wouldn't surprise me if that default is the low speed.

[droidhacker]

Here is the problem with these tests of llvmpipe;

They're testing GAME FRAMERATES. That is NOT where llvmpipe is required for use.


The only important end-user application of llvmpipe is in DESKTOP COMPOSITING, and it isn't a measure of maximum attainable FPS with the CPU pegged, it is a measure of HOW LOW the CPU usage can be when dealing with an underpowered CPU, like a crappy intel-Z with a worthless powervr. Stuff without usable 3D hardware that still needs to do compositing with opengl. Does the llvmpipe driver still cripple an intel-Z processor with gnome3? If the answer to that question is anything besides "yes, and unbearably", then it hasn't improved significantly enough to be worth consideration.

[zoomblab]

Here is the problem with these tests of llvmpipe;

They're testing GAME FRAMERATES. That is NOT where llvmpipe is required for use.


The only important end-user application of llvmpipe is in DESKTOP COMPOSITING, ...

Game framerates are always interesting. If something can handle games, it will probably be able to handle desktop compositing too. It would be great however to have both desktop and game tests in such articles.

[przemoli]

Here is the problem with these tests of llvmpipe;

They're testing GAME FRAMERATES. That is NOT where llvmpipe is required for use.


The only important end-user application of llvmpipe is in DESKTOP COMPOSITING, and it isn't a measure of maximum attainable FPS with the CPU pegged, it is a measure of HOW LOW the CPU usage can be when dealing with an underpowered CPU, like a crappy intel-Z with a worthless powervr. Stuff without usable 3D hardware that still needs to do compositing with opengl. Does the llvmpipe driver still cripple an intel-Z processor with gnome3? If the answer to that question is anything besides "yes, and unbearably", then it hasn't improved significantly enough to be worth consideration.

And cheap to obtain OpenGL 3.x capable GPU!

That also is use case for llvmpipe. Someone without such can use LLVMpipe for learning OpenGL 3.x. So framerates still matter.