Slashdot Mirror

I guess this might be it: http://www.gpgpu.org/

I wonder if http://www.gpgpu.org/ could offload some of the Slashdot effect to their GPU?

The performance of GPU's seem to grow faster than those of CPU's. I remember someone had proposed to use GPU's to proces generic data. It would be 12 times faster than a CPU.

Go here for several examples of this -- far from simply having been proposed, it's been done a fair number of times.

The thing to keep in mind with this is that while the GPU has a lot of bandwidth and throughput, most of that is due to a high degree of parallelism. Obviously 1 GHz hasn't been a major milestone for CPUs for quite a while, but CPUs are only recently starting to do multi-core processing, while GPUs have been doing fairly seriously parallel processing for quite a while.

Along with that, the GPU has a major advantage for some tasks in having hardware support for some relatively complex operations that require a fair amount of programming on the CPU (e.g. multiplying, inverting, etc., small vectors, typically has a single instruction to find Euclidean distance between two 3D points, etc.)

That means the GPU can be quite a bit faster for some things, but it's a long ways from a panacea -- you can get spectacular results applying a single mathematical transformation to a large matrix, but if you have a process that's mostly serial in nature, it'll probably be substantially slower than on the CPU.

Along with that, development for the GPU is generally somewhat difficult compared to development on the CPU. Writing the code itself isn't too bad, as there are decent IDEs (e.g ATI's RenderMonkey) but you're working in a strange (though somewhat C-like) language. Much worse is essentially a complete lack of debugging support. Along with that, you have to take the target GPU into account in the code (to some extent). I just got a call in the middle of a meeting this morning from one of my co-workers, pointing out that some of my code works perfectly on my own machine, but not at all on any his. I haven't had a chance to figure out what's wrong yet, but I'm betting it stems from the difference in graphics controllers (my machine has an nVidia board but his has Intel "Extreme" (ly slow) graphics).

--
The universe is a figment of its own imagination.

You mean like these people are doing?

Generic GPU programming

http://www.gpgpu.org/

Multiprocessing general-purpose apps on a GPU?

What we really need are GPUs that have "public" interfaces, either HW or SW (open drivers) that others can harness for GPGPU. Let's see if that kind of competition expands the market for these GPUs, instead of just fighting ATI and nVidia for the current market.


OpenGL is a 'public' interface that effectively hides the hardware with a standard API while also offering low level programmability via it's shader language. We already have what you're asking for.

Check out the GPGPU project. It sounds like it might interest you.

Apparently that's not a straightforward question to answer. The gpgpu.org hardware features forum would probably be the best place to ask.

http://www.gpgpu.org/

Heat and poor efficiency is caused by 'serial' slow general processors. Amd is no different. Scientists are now moving to parallel/streaming processing GPU's. This kind of chip while hard to program for is still superior. The compiler tech is only getting better. Itanium is an excellent chip that you just dogged. Dump the legacy Intel ISA.

More efficient. More powerful. Great for games too !

If they sold one at the store that had 2 of these chips in them and ran XP/game and linus I would never look back at serial General purpose chips.

http://www.gpgpu.org/

Perhaps you were thinking of this?

Not sure on this but graphics processors are also streaming processors.

http://www.gpgpu.org/

http://www.newscientist.com/article.ns?id=dn7508

> http://www.gpgpu.org/

When did this site last work?

By the way, for those interested in GPGPU research/ideas, there's a pretty nice site here: http://www.gpgpu.org. It has some sample code, slides from conferences presentations, a forum, etc. It's a pretty nice site for information. I was interested in GPGPUs a few months ago and read through the material on that site heavily, but in the end I didn't have the time to try anything cool out because you'll need to learn how to use a language like Cg or steam to program your GPGPU.

The people in TFA are part of a larger group (see http://www.gpgpu.org/) that thinks about how to use graphics cards for a wide variety of math problems. Here's an abstract from one of their papers:

In our experiments we compare the execution on a midclass GPU (NVIDIA GeForce FX 5700LE) with a high-end CPU (Pentium 4 3.2GHz). The results show that to achieve high speedup with the GPU you need to: (1) format the vectors into two-dimensional arrays; (2) process large data arrays; and (3) perform a considerable amount of operations per data element.

I'll repeat a cool link posted by somebody else: http://www.cs.unc.edu/~ibr/projects/paranoia/ - this is a program, originally written in the 80s, to characterize the performance and idiosyncracies of a floating-point processor. Recent work at UNC Chapel Hill has been done on Windows platforms. (Twenty years ago, UNC Chapel Hill was one of the hotbeds of computer graphics development that eventually gave us Shrek and The Incredibles.)

This might be a good starting point:
http://www.gpgpu.org/

(before you start emailing people.
Good luck.

Now we've got loadbalancing GPUs. Which means cheap supercomputers, on a PCI LAN, in cheap P4 clients running the OS of our choice. Everyone overclocking your Pentium for more power: GPGPU is the cheapest way to get the fastest PC. First demo of a pool of parallel LAME process running on a stacked beast, let me know.

check out gpgpu or BrookGPU as starting points.

Two slight clarifications:

1) The 2 TFLOPS figure is commonly thought to be a big lie (or mistake), as a GeForce 6800 Ultra is capable of somewhere around 50-75 GFLOPS. This would mean a 50x performance increase between two generations, which seems very unlikely.

2) See www.gpgpu.org and you'll find that graphics cards can do some serious general-purpose number crunching.

Yeh, but running a few in parallel should help some. I think grandparent post is on the money with scientific visualisation, but I also think they'll be used for GPGPU-style processing on larger datasets. Nvidia seems to have similar ideas...

Look for GPGPU. They are trying to use the graphics processor for general purpose operations. It runs as any CG script would run. Just realize that it is focused more on parallel math operations then procedural. Please note that I have nothing to do with this project and haven't tried it yet.

To complement your nice Tyan motherboard, get one or two (XXX check for physical sizes) of them realizm 800 from 3dlabs. They are the only 16 lanes PCIexpress videocards I know of. Not sure about GPGPU, but at 3840 x 2400, solitaire is bound to look amazing... especially if you can get some nice 9.2Mpixel displays as well: High end videocard without a matching display, what would be the point? Check for instance the IBM T221).

Anandtech reviewed the Realizm 800 here.

Mhhh... If you wanted the machine to be a server of some sorts, then I just wasted 10 minutes typing all this!

Here - just what you're after - people using the GPU for general purpose.

http://gpgpu.org/

http://www.gpgpu.org/ says that the latest GPUs can already accelerate physics a lot. If that's so, how much is a dedicated physics chip needed?

Reading the news, than the article, it seems nobody ever saw any app which would have happened to have OpenGL-based interface [i.e. GUI]. Okay, I help, think Blender.

It is good there are people who are open minded enough to see through the cloud and recognize usabiltities and applicabilities of provided tools. But come on, don't hype it for this reason.

You can hype the gpgpu idea though, which is a very nice way to go for these kinds of applications (too). [Although using the GPU for other tasks then drawing doom3 is mostly revolutionary for the joesixpacks out there. Lotsa many conference papers dealing with the matter popped out mostly in recent 1-2 years.]

People have been doing this stuff for years all right.

Why not use the (massively parallel) processing capability of your video card to:
- mix sound buffers?
- speed up ray tracing?
- do large-scale matrix math, or other parallelizable calculations?
- iterate fractals?
etc.

If you really want to get away from video games you should also check out GPGPU. Its a site dedicated to general processing on GPUs. There is also a lot of research done in this area, such as a DB run on a graphics card as well as a lot computational geometry problems. Also, Stanford has developed a more straightforward C-like language to do GPGPU. Pretty cool stuff.

DX9 and other pieces of graphics code are useful for other type of aplications as well:
http://www.gpgpu.org/
http://graphics.stanford.edu/projects/brookgpu/sta rt.html

I think graphic cards need to step out of the cute display & gamers market a bit, i mean there is a hell of a lot processing power which can be used for someting serious, especially when u are not busy playing games with it. for instance, the massively parallel computional power can be used for better audio processing, better speech processing & all kinds of graph optimization..
in fact, it would be cool if someone came up with a way of utilizing this unused power to do something useful like google did with folding@stanford in google toolbar.

refs: http://gpgpu.org/

Check out http://www.gpgpu.org/ for cool stuff. And if I'm not mistaken, it is already possible to use SLI.

Cheers,

see gpgpu.org for more information.

The GPGPU course was at the top of my list for SIGGRAPH 2004 this year, and I was impressed with all of the presenters' material. However, in my estimation, Sh is built to more closely resemble the existing HLSL for DX and similar GLSLang from the ARB.

Brook, on the other hand, was written from a more C-like perspective, and approaches the GPU as a massively data-parallel stream processor (well, Sh does as well, but IMHO Brook achieves that aim more directly as is evidenced by things like iterator streams and similar kernels).

On the general-purpose computation on gpus website there's an interesting article about a Beowulf^H^H^HWindows XP cluster of PCs with AGP gfx cards. They mention the asymmetric bandwidth: 2 GB/s to the card and 133 MB/s back. A high-end digital stereo signal (24 bit/192 kHz) is 1 MB/s, so I'm not so sure why you need the big backwards bandwidth.

Anyway, when will nVidia start selling boards that are optimized for calculations? No VGA output and multiple processors per board. I can see it in shops. The Supercomputing Expansion Card. Improve the performance of your PC up to 20-fold!

As mentioned, the entire General Purpose GPU (GPGPU or GP2U) community has been on this a while. They recently put on a SIGGRAPH 2004 Course and there is the GPU Gems book which as some GPGPU in it.

As mentioned, the entire General Purpose GPU (GPGPU or GP2U) community has been on this a while. They recently put on a SIGGRAPH 2004 Course and there is the GPU Gems book which as some GPGPU in it.

If you already haven't, check out http://www.gpgpu.org/ for some of that :)

While it's not exactly a project to produce a holistic "offload driver", the research they do and share is probably a good starting point toward such a package...

They do a boatload of suprisingly different kinds of stuff on the GPU already!

I read some articles about using standard GPUs for matrix and mathematical operations. Does anyone know if this is being coupled with clustering? Seems that this would give you some of the power of vector processors, but thats just my $0.02.

The feature side of the equation is a lot easier to handle, as Intel has lavished all the features a techie could dream about on the new chipsets. High-Definition audio, Matrix RAID, a new bus with a bright future, and an 8GB per second bidirectional graphics slot are a few of those features that come to mind.

Try GPGPU.org - "General-Purpose Computation Using Graphics Hardware". Useful clearinghouse for this sort of thing.

Open up their specs so you can write a real-time raytracer? Why can't you use Cg or HLSL like others have done? Why do you need to write to the video card directly? You have full access to the programmability of the GPU through these languages. If not, program the damned thing in their version of assembler through the DirectX or OpenGL APIs. Unless by "tweaking OpenGL or DirectX" you mean "programming the GPU", your statement seems flat-out wrong.

Don't believe you can do it? Here's a link some projects that do real-time raytacing, radiosity, photon mapping, and subsurface scattering, all on GPUs. These GPUs are programmable without them opening up their specs.

(The desire for them to open up their specs is for other reasons, not because they are hiding some functionality from you.)

Two points:

First, Why? Most people don't even make movies that are raytraced.

Second, they already are doing raytracing on the GPU. Purcell had one working in 2002. There was a presentation on it, in a course at SIGGRAPH 2003. The GPU is maybe a little faster than the CPU, right now, for raytracing.

"Tweaking OpenGL" is kind of like saying "tweaking the CPU", any more. It's fairly close to a generalized stream processor. And their specs already are open enough to have figured this out. Look at GPGPU and read some more about how people are doing amazing stuff on the GPU today. No need to wait for ATI and NVidia to open up any specs - they already did. Cg and GLSlang are fully up to the task.

And, photon mapping and similar techniques are much more sophisticated than raw raytracing.

Is this compatible with Brook and other general-purpose GPU programming techniques? The use I see for it is this:

Imagine an openmosix cluster of dual-processor machines that run bioinformatic calculations and simulations. Lots of matrix math and such - pretty fast (and definitely a lot faster than a single researcher's machine).

Now imagine the same cluster but each machine has 2 or 4 dual-head graphics cards and each algorithm that can be created in Brook or similar is. That gives each machine up to 2 CPU's and maybe 8 GPU's that may be used for processing. The machines are clustered so a group of ~12 commodity machines (1 rack) could have 24 CPU's and 96 GPU's. Now that would be some serious computing power - and relatively cheap too (since 1-generation old dual-head cards are ~$100-$150).

By the way, does anyone know if there is any work going on to create toolkits for Octave and/or MatLab which would utilize the processing power of a GPU for matrix math or other common calculations?

General-Purpose Computation Using Graphics Hardware. Anyone interested in this topic should check that site out.

As for organizations beating slashdot to the punch on this one, that's true... but it's good to see this getting even more exposure. :)

GPGPU (General-Purpose computation on GPUs) was a hot topic at various conferences in 2003; a number of papers were published on the subject. At SIGGRAPH 2004 there will be a full-day course on GPGPU given by eight of the experts in the field (including myself).

Mark Harris of NVIDIA maintains a website dedicated to GPGPU topics, including discussion forums and news postings. Well worth a browse if you're interested in GPGPU topics.

I look forward to seeing some of you at SIGGRAPH! :)

--Cliff

As for organizations beating slashdot to the punch on this one, that's true... but it's good to see this getting even more exposure. :)

GPGPU (General-Purpose computation on GPUs) was a hot topic at various conferences in 2003; a number of papers were published on the subject. At SIGGRAPH 2004 there will be a full-day course on GPGPU given by eight of the experts in the field (including myself).

Mark Harris of NVIDIA maintains a website dedicated to GPGPU topics, including discussion forums and news postings. Well worth a browse if you're interested in GPGPU topics.

I look forward to seeing some of you at SIGGRAPH! :)

--Cliff

here's a link gpgpu.org

you can do that by typing <a href="http://gpgpu.org">gpgpu.org</a>

Seems worth checking out: GPGPU.ORG - "General-Purpose Computation Using Graphics Hardware"

> AGP does a lot better taking data in, but it's still pretty
> costly sending data back to the CPU.
I've heard that mentioned a few times, is it true?

From the AGP 3.0 spec:
The AGP3.0 interface is designed to support several platform generations based upon 0.25m (and
smaller) component silicon technology, spanning several technology generations. As with AGP2.0, the
physical interface is designed to operate at a common clock frequency of 66 MHz. Its source
synchronous data strobe operation, however, is octal-clocked and transfers eight double words
(Dwords) of data within the span of time consumed by a single common clock cycle. The AGP3.0 data
bus provides a peak theoretical bandwidth of 2.1 GB/s (32 bits per transfer at 533 MT/s). Both the
common clock and source synchronous data strobe operation and protocols are similar to those
employed by AGP2.0.11

Later on Page 96:
Traditional AGP devices can demand up to the maximum bandwidth available over the AGP ports.
However, the AGP system does not guarantee to deliver the requested bandwidth, nor does it guarantee
transfers will take place within some clearly specified request/transfer latency time. ...
This is done by the system guaranteeing to process a specified number (N) of read or write transactions of a specified size (Y) during each isochronous time period (T). An AGP3.0 device can divide this bandwidth between read and write traffic as appropriate. Further, the system transfers isochronous data over the AGP3.0 Port within a specified latency (L).
(emphasis mine)

I'm no expert, just asking if the "low upsream bandwidth" assumption is true. If it is, there could still some applications (eg: simple data compression) that could use it. Also, maybe output from VGA/DVI ports could be tapped.

some one else posted this...

www.gpgpu.org

Website on this topic (Score:0)
by Anonymous Coward on Sunday May 09, @01:57AM (#9098550)
General-purpose computation using graphics hardware has been a significant topic of study for the last few years. Pointers to a lot of papers and discussion on the subject are available at: www.gpgpu.org [gpgpu.org]

http://www.gpgpu.org/ is a great resource for general purpose graphics processor usage.

General-purpose computation using graphics hardware has been a significant topic of study for the last few years. Pointers to a lot of papers and discussion on the subject are available at: www.gpgpu.org