News - NVIDIA to bring GPU acceleration to everyday security software?

[HEXUS]

Virus scan taking a while? NVIDIA reckons CUDA acceleration could be the answer.

Read more.

[kingpotnoodle]

Might not accelerate the whole system scan due to hard drive limitations, but it could go a long way to helping cure the percieved slowness that AV causes due on on-access and real time protection, scanning items in memory, websites etc...?

[Lowe]

Beaten to it - isn't the bottle neck on a virus scan the HD?

[Methanoid]

Having seen the "huge" performance increases CUDA offers (say 0-1% on arithmetic PAR2 file checking) I wonder if this is worth the effort apart from to give nVidia free publicity.....

[Agent]

Having seen the "huge" performance increases CUDA offers (say 0-1% on arithmetic PAR2 file checking) I wonder if this is worth the effort apart from to give nVidia free publicity.....

Where is the bottleneck though?

Is it in the program itself, is it unsuited to CUDA, does CUDA have issues, is the HD being the limiting factor? and so on....

Offloading things to CUDA when appropriate and done correctly can give a huge increase in performance.

Are you using Par2cmdline? If so, remember that "This is the experimental CPU/GPU version. It may be useless to you."

Badaboom is a nice example of something that works well offloaded to CUDA.

[Biscuit]

Until i actually see this CUDA stuff benefiting the home user and being genuinely worthwhile its all just words. They seriosuly need to stop talking and slating other companies and just get on with doing things we can physically see.

[dangel]

It's not whether it can do it faster than the CPU - it's just that it's a dormant resource on the desktop for the most part - why not use it to free up the CPU for other stuff? I agree the bottleneck is the HD for completing a scan but it might be useful to keep actual processing away from the cpu (atom systems) and it's a good sell to the corporate sector perhaps. I'd certainly like to see some numbers first.

[Irien]

Wouldn't it make far more sense for AV writers (or indeed any software authors) to use DirectCompute or OpenCL or similar, since that is at least a semi-standard? I also read (perhaps wrongly) that DirectComputer 1.0 is backwards compatible to DX10 hardware and therefore could potentially be released for older OSs as well. Whilst Cuda would allow it to run on a broader range of operating systems, the lock-in to nvidia hardware must surely be more of a "downer"?

[dangel]

Wouldn't it make far more sense for AV writers (or indeed any software authors) to use DirectCompute or OpenCL or similar, since that is at least a semi-standard? I also read (perhaps wrongly) that DirectComputer 1.0 is backwards compatible to DX10 hardware and therefore could potentially be released for older OSs as well. Whilst Cuda would allow it to run on a broader range of operating systems, the lock-in to nvidia hardware must surely be more of a "downer"?

Really agree on this - nVidia would love everbody to embrace their standards for physics and compute but it's ultimately a cul-de-sac (unless MS suddenly decides *not* to go it's own way).

[Bull Dog]

Actually, CUDA won't help me with Virus scans. Why? I use ATI cards. Useless FTW. Come on NVIDIA stop pushing your proprietary standard.

[handscombmp]

Could be useful as it free ups the cpu from the anti virus and most of the time the gpu sits there not doing much.

[TheAnimus]

CUDA works well when you can offload without latency issues.

In all of the tests I've been doing (windows) and a former college (whichever BSD is in vouge) we found issues with the time it takes to build up the problem data space.

Now thats not to say for some kind of things it is the messiah, but for things like on access virus, I would say hell no. For a start off, even with DMA there is a hudge latency issue when pulling files of an SSD and getting it into CPU land. Let alone then across the PCIex16 to GPU RAM lan, have the GPU run over it, then back to the CPU.

CUDA is really quite great at some things, but this its great everywhere motto really isn't true at all, and is likely to hurt serious uptake

[kalniel]

Now thats not to say for some kind of things it is the messiah, but for things like on access virus, I would say hell no. For a start off, even with DMA there is a hudge latency issue when pulling files of an SSD and getting it into CPU land. Let alone then across the PCIex16 to GPU RAM lan, have the GPU run over it, then back to the CPU.

Isn't that what some of the fermi architecture is designed to mitigate? I can't quite remember, but thought DMA was an issue they looked at.

[watercooled]

I really can't see how this would help - my CPU is very rarely fully loaded during a virus scan, it's usually just above idle. Yeah it does spike occasionally, probably when scanning archives but I doubt this would improve scanning speed - maybe by a few seconds...

[lodore]

ive just found a link from 2007 showing the speed difference between running kaspersky antivirus on a gpu rather than a cpu. link 21x faster seems a good enough boost to me.

[watercooled]

That was scanning a big hard drive array - as mentioned already the CPU isn't the limiting factor it's the HDDs - just look at the bandwidth on that article. On a home computer I doubt you'd notice any difference at all...