Hello

Today a friend told me he is looking at building a new computer and for costs sake, rather than look into multiple CPU designs, look into using graphics cards for processing support. He asked me what I knew and if I could help build it. I know Nvidea has something out that lets you use their graphics cards for math and graphics work (called CUDA?), but I want to stay with AMD products (Im a fan and loyal). He doesnt care as much.. but anyway.

I am having a more difficult time finding exactly what I am wanting to know.. which is does AMD make graphics cards that I can use like NVidea does? What I have found so far is something called Firestream.. which from AMDs website (http://www.amd.com/us/products/serve...9370-9350.aspx) seems to be exactly what I am wanting but I want to make sure it is and if anyone has had experience with setting up something like this?

For a "normal" user there is currently no real advantage from the ability of graphics cards to accelerate some functions. There is currently only a very small number of programs that use Nvidia's CUDA or AMD's Stream (Firestream are dedicated accelerator cards, not using GPUs for calculating). Nowadays quad-core processors are real cheap, I would rather recommend to go for that instead.

I think that you are being over-optimistic about what the graphics card might do for you; essentially, a 'core' on a graphics card is pretty dumb, but there are lots of them, and they can be fast, on exactly the right kind of operation. Essentially, what TobiSGD said was correct; for most people, most of the time (under today's conditions), the graphics card doesn't/can't contribute to their computing experience. That changes if you are running specific programs that usually do a lot of the same, strictly mathematical, operations on multiple data points. Examples would be Computational Fluid Dynamics, Meteorological Simulations or simulations more generally (not that every simulation program uses graphics cards, and you'll normally have what amounts to a Supercomputer (possibly a cluster of computers, amounting to a supercomputer) for some of these, and, much as I am sure that your friend would like a Supercomputer, I doubt that he is up for the associated costs).

That said, if you are still interested, you might choose to go for a processor with an integrated GPU; I am not convinced that, at today's prices, these really represent the value that you might expect from greater integration. When I have priced up, eg, the AMD Llano processors (eg, the A8-3850 processors, or, say, a cheaper A6-3500, or whatever model from that range seems right) the prices of a processor plus motherboard compared with a non-integrated processor of a conventional type, plus a cheap (appropriate) motherboard, plus a cheap, separate, graphics card there wasn't a big difference, depending on exactly which processor you compare with which.

I am not saying that these are 'bad' processors, just that you might expect greater integration to make these an obvious bargain, but it is more case that, counting all the costs, they are just competitive, nothing more. But, if you think that eventually the software technology might make the Graphics part more usable, then you would be well placed to take advantage of that (still, that is probably more applicable if you run supercomputer programs, but there may also have been a 'folding at home' option, so, if that is the kind of thing that floats your boat, check out whether that still exists...and, anyway, that is 'sort of' a supercomputer program, distributed of thousands and thousands of computers, just not computers owned by the central authority).

Alternatively, there are Intel options, but it doesn't sound like you would want that, and while, in general, the Intel processors have better general purpose cores (depending on which exact model you buy...and your definition of better), the graphics part is less capable than the AMD alternatives. Something like a cheap Sandy Bridge Celeron or Pentium (eg, G440, 530 or the G840 series) would meet the basic requirements, but as the graphics part isn't much good, I doubt that you will ever get any real advantage from it. Again, these aren't exactly bad processors, with the exception of the G440, but they don't meet the requirement of being AMD, and the probability is strong that you'll never get the advantage that you are looking for, although the motherboards are reasonably priced.

All-in-all, I think that you should treat this as a straightforward upgrade question, and if you end up being able to do something with the graphics part, that's all to the good. I'd bet that you never actually get any real advantage from it, though, unless you have pretty specialist requirements that you haven't told us about.

Huh.. I will let him know.

So the Firestream isnt like CUDA... like using a graphics processor to handle tasks? What does it actually do then?

The counterpart for CUDA on AMD is Stream. Firestream technology is somewhat similar, but instead of using the graphics chip for computing it uses dedicated accelerator cards. A good source for knowledge about that is in fact Google, have a look at the first hit if you search for amd firestream.

=================================================================== I just built an AMD A8-3850 APU machine using Biostar TA75M-A8-3850 APU-8GB Mushkin Radioactive PC3 1333or1600or1866MHz-1TB Barracuda-500w Volta PSU-and LG SATA DVD/CD RRW for around $400.00 form a place called OutletPCdotcom. The APU has ATI video and AMD quad core CPU on the chip and it is faster than what I need now. Neat simple fast configuration for me and I thought it was inexpensive. Some folks on net say that it is about as fast as an Intel i3-2600 rig. Nvidia is most likely going to have better Linux support. Tho my system works, I have an AMD icon on my desktop reminding me that the AMD hardware is unsupported.
I had to use alternate cd for installation and then used repositories for Catalyst drivers.
GUI install cds would not display past a few seconds at the beginning for some reason...
LuCk2ya

Okay see I was looking at Google originally but found conflicting information.. thats why I asked here. I dont mean to be rude, but telling to do research when I did say I was is counter-productive.

What I am finding is confusing to me.. like this comment from the first link TobySGD suggested is

"The AMD FireStream™ GPU compute accelerators deliver all of the parallel processing power of the GPU to compute-intensive applications for scientists, engineers and consumers. "

Isnt that supposed to be the same as CUDA? And stream.. that practically says the same thing

OK, to make it clear:
CUDA -> Accelerated computing on Nvidia video cards and Nvidia's Tesla compute accelerators
Stream -> Accelerated computing on AMD video cards
FireStream -> Accelerated computing on AMD's Firestream compute accelerator cards

Ther is also OpenCL, which is the pretty much same, but manufacturer independent.

'Multipule CPU designs'? If you mean 'multiprocessors' (2 or more CPUs) its pretty much pointless for any desktop use these days. When you can buy more cores than gives you any advantages in most situations, a single multi-core CPU is all you want, let alone 'need' for the vast (97%+) of computer users.

I havent done it with a firestream card, but I have with CUDA.

That reminded me of this-

http://www.dansdata.com/gz017.htm

Very old, but on-topic.

Marketing. To quote the link I put in above, "This statement is, to use a technical expression, "a big fat pile of marketing". As in, "Look out! Don't step in the marketing!""

nVidia goes on about the 'power' of CUDA and GPGPU...because they dont have any other major products. AMD is just following suit.

I'm not sayign that CUDA, or stream, is pointless. For some users they are great technonogies. But +1 to salasi and TobiSGD, having CUDA/stream setup isnt going to turn a cheap dual-core into a octa-core beating fire-breather for any common desktop tasks...if you want to run tasks that just do floating-point it will help...but how many people run mostly/all that sort of task? Not many.

Thank you everyone for the help.