Anyone seen this:-

[Frippertronix]

I'm beginning to see the logic in this sared video resource idea. At first it seemed weird because it says in the article that some degree of translation is necessary to task video resources to plugin processing, which seems very indirect to me.

However, the genius I see in this is that most people have wasted video card power much of the time because they buy expensive cards for games and unless they are doing high level animation or maybe some other kinds of video production, the 256 MB video card just sits there languishing while not being used for gaming 3D. Why not put that power to use for audio processing during what would have normally been down time, since it's barely breaking a sweat animating a DAW GUI.

[valley]

I'm about to get one of the next gen motherboards, but I don't have PCI-E right now. Is PCI-E for video cards different than the PCI-E sockets for all other cards? I just see things advertised as "PCI-E video card" so I didn't know if there is a special PCI-E bus and/or socket specifically for video.

The video card slot is physically distinct from the other PCI-E slots.

The main difference internally is that of bus rate. I can't recall off hand which PCI-E busses are which, but fundamentally there are 1x, 4x, and 16x slots. Most expansion cards would go into a 1x slot, but video cards use the 16x slot, which -- not surprisingly -- offers 16x the bandwidth of a regular PCI-E slot.

(more or less the gist of it anyway - I'm a little too sleepy to risk trying to provide details)

A dual card SLI board therefore would offer 32x regular PCI-E bandwidth, and a potential number crunching welly braodly equivalent to 4 P4 3GHz[1] processors.

[1] although comparing the processing grunt of a GPU against that of a full processor such as a P4 is a little misleading.

[james0tucson]

Why not put that power to use for audio processing during what would have normally been down time, since it's barely breaking a sweat animating a DAW GUI.

The reason is mainly that every GPU has a different driver interface, and to make it do anything besides graphics means you'd need to take a different for each graphics chip.

If you can disguise your computationally-intensive audio stuff so the graphics driver thinks it's rendering OpenGL or DX, you'd have what you need in a way that would be compatable for more than just one video card.

[Frippertronix]

Why not put that power to use for audio processing during what would have normally been down time, since it's barely breaking a sweat animating a DAW GUI.

If you can disguise your computationally-intensive audio stuff so the graphics driver thinks it's rendering OpenGL or DX, you'd have what you need in a way that would be compatable for more than just one video card.

I guess that may be what the company in the article is trying to do. If so, the fact that they dropped off the radar may be an indication that they are having a challenging time with it.

I like this idea because of the "killing two birds with one stone" aspect, but also because, while I suppose getting a dual processor system is an alternative, going the route of having two PC's working in parallel involves so much redundancy that's its prohibitively expensive for most people.

[stefancrs]

Why not put that power to use for audio processing during what would have normally been down time, since it's barely breaking a sweat animating a DAW GUI.

The reason is mainly that every GPU has a different driver interface, and to make it do anything besides graphics means you'd need to take a different for each graphics chip.

If you can disguise your computationally-intensive audio stuff so the graphics driver thinks it's rendering OpenGL or DX, you'd have what you need in a way that would be compatable for more than just one video card.

Doing generic DSP in HLSL (esp in 3.0) isn't that tricky. I'm not sure how hard it is to incorporate it into a VST plugin, but the actual DSP part isn't hard.