Hyperthreading optimizations

[Oompa Loompa]

Has anyone examined the extent to which ATI and NVIDIA have multi-threaded their drivers? I'm thinking specifically of the effects of HT, not whether individual tasks are faster with a K7 or P4.

Take 3dmark2003, for example. Regardless of whether it is a valid benchmark for comparing different video cards, it is certainly suitable for comparing the benefit of enabling HT. 3dmark is single-threaded, so any benefits from HT will come from DirectX and the video card manufacturer's drivers.

So, if you have a HT-capable P4, what % benefit in 3dmark is obtained by enabling HT? And what drivers are you using?

 

[Ante P]

I think nV has come a lot farther than ATi on multi threaded drivers, no?

 

[Tim Murray]

Zero, I would assume. Video card drivers probably don't give that huge of an overhead anyway, so that seems a little wonky.

I mean, you don't get any (or barely any) speed increase in Q3 when you use SMP (which, in theory, would be even more effective than HT), so I don't think that HT or SMP-optimized drivers would really make a difference in 3DMark, games, etc. al.

But for workstation apps... that might be a different story.

 

[Oompa Loompa]

Numbers would be more helpful than theory, in this case.

 

[Tim Murray]

I don't think anyone really has numbers at this point--it's still pretty early in the life of HT, and considering it was previously available only on one processor that was absurdly expensive, I don't think any company has done much optimization for it (too much 3DMark perhaps ).

 

[cho]

+150 points, 3.06ghz p4 ,radeon 9700 pro.

 

[Oompa Loompa]

+150 points, 3.06ghz p4 ,radeon 9700 pro.

Could you give that in percentage form? The absolute number is, in part, a function of the GPU's capability.

 

[ET]

Why would there be any big benefit for HT in drivers? What do you think can be made to efficiently use more than one thread?