An heretical thought.

[kruntz]

Bob,
it seems that with a 100$ graphic card, you could have 400%-500% more audio processing power than the fastest CPU on the market...

http://www.nvidia.com/object/cuda_what_is.html

:-)

(I'm not joking...)

[Yogi]

While it may seem that there's a lot going on in terms of graphics in relative terms terms of data crunching it's pretty small. The only real time changes graphically is the meter display for each channel and that's not much of a big deal (in terms compared to multiple streams of audio data).

[ssrsound]

Yeah -- but the point is that there's this massively fast FPU co-processor that you can make use of if you're doing your programming at a low level (which bob is).... not for graphics, but for crunching audio data. At their core, GPU chips are nothing but really fast floating-point processing chips.

I think it'd be an interesting experiment... basically a switch in the app that detects if you've got a GPU that it can make use of, and for the cost of a $100 graphics card you get that much more processing power (more channels, plugins, etc).

But there's two things that might make it a waste of time:
- don't know the latency involved in using something sitting out on the PCI buss rather than directly on the processor/memory buss.
- In the time bob could make the app make use of that, Intel might double the speed of available processors, making all that work for nothing.

There is some advantage to staying generic, and just counting on the idea that hardware will catch up to you eventually.

[Brent Evans]

Yeah -- but the point is that there's this massively fast FPU co-processor that you can make use of if you're doing your programming at a low level (which bob is).... not for graphics, but for crunching audio data. At their core, GPU chips are nothing but really fast floating-point processing chips.

That might be useful if floating point math were involved. If I recall correctly, SAC and SAW use integer math.

[Trackzilla]

I remember seeing a few plugs that were designed to run in a cuda rateher than the cpu, didn't pay too much attention as I don't have such a graphic card, but the concept is being toyed with by some.

also, some of the optically based multi-touch guys are diverting their image processing for their touch systems into the gpu and achieving better response.

The point being, if you get too much going to the gpu, it'll get just as cluttered as the cpu, I think I'll vote for moore's law continuing to drive my performance up with time...especially since I'm technically 'last generation' already using an e8500 & I'm not running out of power yet.

[wascomat]

..but wouldn't all the 1's and 0's going across the screen be distracting??

-Jim

... but how can we get even more cowbell in the mix?

[RBIngraham]

There is a Impulse Response reverb I saw on KVR-VST that uses the GPU to do it's calculations. Kind of an interesting idea, but I also think this is largely a gimmick type thing. Sure if you are running a DAW with a fancy graphics card (rather than just a stock built onto the motherboard output) there is probably a lot of GPU bandwidth that is just being thrown away. But even so then you have more data going back and forth between the graphics card and the system bus. The entire point of getting a graphics card was so that the main system had to do less work, and therefore free up the CPU for other tasks.

Most graphic cards don't sit on the PCI bus these days. Most are PCIe cards, and that is a separate bus from the PCI bus. Still both the PCI and PCIe bus have to make a connection to the main system bus architecture. But PCIe has a much larger bandwidth available.

My gut tells me all this is more intended for those that like to play and/or develop video games. That is probably where 90% of the push for more speed in computers comes from anyway. Us media artists just get to have a side benefit. If audio and video artists were the only ones asking for faster and faster computers then things would move a lot slower or at least cost a heck of a lot more. So next time you see a gamer, thank them for driving the CPU race.

All this just begs the question is anyone really running out of CPU in their SAC rig? Even with now modest Core 2 Duo CPUs people seem to be getting huge channel and plug-in counts. Just curious.

[DennisC]

I think there is a real problem applying the cuda stuff to "realtime" audio processing. I am no expert, but I met one, but I believe that the cuda stuff is really best at breaking up problems into separate pieces and parallel processing them.

For instance, image processing: This fits cuda well because you can break up the full image into pieces, parallel process them, and put it back together. So processing that once took hours and be done in minutes etc.

This does not hold for realtime audio processing, as far as I see it.

Dennis

[ssrsound]

All this just begs the question is anyone really running out of CPU in their SAC rig?

As my proof-of-concept system I'm running an old shuttle 1.8 ghz Pentium-D system, and a MOTU 2408 MkII and three ADA8000s. With it, I can do 24 channels, all with EQ, compression and Aux for reverb, 6 monitor boards, each sending an mono out, and a 7th monitor board doing 24 channels of direct out for recording to an external 24-track. When i do that, I get around 40-48% CPU usage, and start to hit trouble if I push it too much farther with plugins. I've found I don't really want to push it past 50%

So yes, I'm running out of CPU. On the other hand, all I have to do is buy a newer computer/processor and I'll be fine... probably even fine to add another 2408 and three more ADA8000s.

so yeah ... Moore's Law probably wins on this one.

[Iain Westland]

and a 7th monitor board doing 24 channels of direct out for recording to an external 24-track

how? I got one of the first copys of SAC And have only just got around to opening it and looking under the skin so please accept my stupidity

iain