Scientists make eight-GPU 'desktop supercomputer'

[Claymonkey]

Tech Report Story

Fastra's Site

This looks like a VERY good use of these new GPU's and I'm all for medical advances on the cheap. This is the kind of thing that will bring 3D rendering for surgery and diagnosis to regions that could in no way afford it.

At around $500 a pop, Nvidia's dual-GPU GeForce 9800 GX2 graphics card may be a little too pricey for most gamers to afford. However, scientists at the University of Antwerp in Belgium think it's a pretty good building block for desktop supercomputing. The ASTRA research group at the University's Vision Lab has built a desktop system with four GeForce 9800 GX2 graphics cards, which it uses for tomography computations.

In ASTRA's words, tomography "is a technique used in medical scanners to create three-dimensional images of the internal organs of patients, based on a large number of X-ray photos that are acquired over a range of angles." With the computing power of four 9800 GX2s, ASTRA says it can perform tomography calculations at the same rate as 350 modern microprocessor cores all working together. That can cut computing times from weeks on a regular PC to just a few hours. Building the machine cost a total of less than €4,000 ($6,200), which is at least a couple orders of magnitude cheaper than a conventional server cluster.

Take this and the RepRap project RepRap and we could come a long way very shortly. There is huge interest in 3D printers in medicine since you can 'print' a copy of the organ you're going to operate and study it in the round before a single cut is made.

[Dr Mordrid]

Interesting from all fronts - AMD is scheduled to come out with a 12 core CPU where each core has its own high speed GPU capable of multichannel stream processing (the AMD/ATI name for NVidia's CUDA, known generically as GPGPU).

[Nowhere]

While both technologies indeed fall under GPGPU...they are a bit different in concept (from what I've heard AMD "Close to metal" is, well...closer to metal, while CUDA is higher level library depending on drivers; and who knows what Intel will cook up)

Another "format war"?... :/

[Dr Mordrid]

Unfortunately, quite likely.

[Nowhere]

Seems like we have a standard in the making...

http://en.wikipedia.org/wiki/OpenCL

(and supposedly AMD is adopting it)

[Dr Mordrid]

Well, well....now if only they can get Intel, NVIDIA etc. to go along.

I'm not holding my breath

That said: anyone else catch that NVIDIA's Tesla-10 GPGPU tech is coming?

Link...

>
The Tesla-10 Series chip ships with 240 processing cores – up from 128 cores in the previous product. Although, these are not the beefy cores associated with general purpose chips made by Intel, AMD and others. Instead, they’re little babies that have previously just handled graphics jobs.

Overall, the new chip boasts 1.4bn transistors and 1 Teraflop of computing muscle.

That 1 Teraflop figure is up from half a Teraflop with the older Tesla 8 chip. In addition, the new Tesla chip kicks memory support up to 4GB from 1.5GB, and that’s again a key leap forward for placating the HPC crowd.

The base unit inside of a Tesla chip has been dubbed a Thread Processor Array (TPA). The TPA consists of eight cores, which all have access to a shared memory bank. Nvidia then combines 30 of the TPAs to make a full Tesla 10 chip.

Those customers looking to get into the Tesla game have a couple of system options. Nvidia has rolled out the S1070 box, which is a 1U unit that contains 4 of the Tesla 10 chips. So, that’s 960 cores running at 1.5GHz, reaching 4 Teraflops of performance. The system also holds 16GB of memory, has peak memory bandwidth of 408GB/sec and consumes 700 watts.

You’ll need to connect the S1070 to a host server with a general purpose CPU via a pair of PCIe Gen2 cables.

If an entire box isn’t your thing, then Nvidia offers up the C1060, which is a cigarette carton-sized device that plugs into the PCIe slot on a motherboard. This puppy holds a single Tesla 10 chip clocked at 1.33GHz, has 4GB of memory and eats up 160 watts.
>
On an algorithm having to do with Dynamics of Black Holes, for example, the new unit runs four times faster than its predecessor and a whopping 84 times faster than a CPU. And, if you’re into Cholesky Factorization, then the Tesla 10 unit shows much more dramatic scaling than the Tesla 8 units while also coming close to quadrupling performance.

Now, Nvidia CEO Jen-Hsun Huang might have said that the processor is dead, but the company really seems to see regular CPUs living alongside these GPGPU systems. Nvidia talked to us an awful lot about heterogeneous computing where the CPUs handle some tasks, and GPUs take on those specialized, parallel tasks that can map well onto the weird silicon.
>

[Evildead666]

yup, saw that and thought "thats a sh*tload of memory"
and then thought that the memory might have to replicate, which means 4GB really useable...there's no memory sharing yet AFAIK...even with cuda.

[Nowhere]

Doc, I just realised...perhaps Intel is already sort-of onboard. Since Apple initiated OpenCL it's reasonable to assume they plan to put it in some serious use...meanwhile most Macs ship with Intel GFX and I don't see how is that going to change, especially with Intel Larabee.

[Dr Mordrid]

Possible, but in this environment I'll believe it when I see it.