Kaspersky enlists CUDA to kill viruses

[Fat.Clown]

Delivers the AV promise

For years Nvidia has been touting its Tesla GPGPU platform as the next big thing in computing. It can be used to build supercomputers for massive parallel computing tasks, dull scientific stuff and we hear it's a rather good basis for toasters, too.

However, Kaspersky decided to be a bit more sensible and utilize Tesla's number crunching prowess to improve security. We already mentioned CUDA could be used to run security software, here. The Russian security incorporated Tesla S1070 units into its security infrastructure to boost client protection. Kaspersky claims the Tesla managed to outperform a 2.6GHz Core 2 Duo in certain tasks by a factor of 360. No, it's not 360 percent faster, it's 360 times faster.

Kaspersky redeveloped its similarity algorithms and optimized them to simultaneously perform hundreds of thousands of instructions. Kaspersky boffins utilized the Nvidia CUDA SDK development environment specifically for this purpose as it allows programs to be written for the latest generations of NVIDIA graphics processors in standard programming languages.

Mind you, this doesn't mean regular users with Nvidia GPUs support stand to benefit from the technology anytime soon, but could be a sign of things to come. Both Nvidia and ATI are pushing parallel computing and offering some rudimentary GPGPU capabilities, mainly in the video encoding department. Using GPUs to power AV software is something much more useful than video transcoding and could bring GPGPU to a much wider audience.


source: fudzilla

I like.....
I also wish they specified which GPU they used and maybe how it compares to the newer i7 CPUs

 

[philbar71]

This is pretty cool, perhaps i'll try it out. I've been running with no AV for a few weeks now

 

[zedman3d]

I sent my idea of using inbuilt cuda to speed up any calculations the antivirus needed to do for customers/users to Avast. They responded with something like: "We are interested in this too and will be looking into it for future versions." I sent that like a year ago.

But anyway, cool.

 

[Puddle Jumper]

How is beating a <$100 cpu with a ~$8,000 Tesla system impressive? A Bugatti Veyron will absolutely destroy a Dodge Carravan on a race track but I don't see Bugatti making a big deal about stating that obvious fact.

 

[zedman3d]

How is beating a <$100 cpu with a ~$8,000 Tesla system impressive? A Bugatti Veyron will absolutely destroy a Dodge Carravan on a race track but I don't see Bugatti making a big deal about stating that obvious fact.

I guess because 360 x $100 = $36000.....
And they only had to pay $8000 for 360 times the performance.

 

[Puddle Jumper]

I guess because 360 x $100 = $36000.....
And they only had to pay $8000 for 360 times the performance.

As far as I can tell that $8k is only the cost of The Tesla unit itself and you still need a actual server to connect it to which would probably add another $1k or so to the total cost. For that $9k you could build a cluster of 32 1.8ghz Quad Core Opterons with 72gb of total ram which would be a far more interesting and meningeal to compare to the Tesla card. that is also assuming that developing the application for CUDA costs the same as developing it for a mulktithreaded x86 system which may not be the case.

 

[kmote]

How is beating a <$100 cpu with a ~$8,000 Tesla system impressive? A Bugatti Veyron will absolutely destroy a Dodge Carravan on a race track but I don't see Bugatti making a big deal about stating that obvious fact.

That's a very skewed way of looking at it.
I don't know the figures but we will say that the tesla system costs $8000 and performed roughly 150 times better than a system with a $100 CPU.
It's already won the value for money crown but if we say that to have value in the business, the system must perform to a certain degree which is covered by 3 tesla systems, we would optimistically be looking at 450 of those x64 systems with all the power, space, maintenance, administration and cooling that goes with it.

 

[Puddle Jumper]

That's a very skewed way of looking at it.
I don't know the figures but we will say that the tesla system costs $8000 and performed roughly 150 times better than a system with a $100 CPU.
It's already won the value for money crown but if we say that to have value in the business, the system must perform to a certain degree which is covered by 3 tesla systems, we would optimistically be looking at 450 of those x64 systems with all the power, space, maintenance, administration and cooling that goes with it.

I'm not saying the Tesla system is bad by any means but IMO comparing it to a $100 cpu is pretty meaningless. You could probably come up with a similar figure comparing Gulftwon to ARM7's but it wouldn't be any more useful.

 

[zedman3d]

I'm not saying the Tesla system is bad by any means but IMO comparing it to a $100 cpu is pretty meaningless. You could probably come up with a similar figure comparing Gulftwon to ARM7's but it wouldn't be any more useful.

Its the only thing they can compare it to. The cluster of Core 2 Duos is all they had...

 

[Puddle Jumper]

Didn't they just compare a Tesla S1070 to a single Core 2 Duo not a cluster?