Santa Clara (CA) – PC vendors such as Dell will soon begin selling desktop supercomputer built around Nvidia’s Tesla floating point accelerators. Priced around $10,000, these PCs may look innocent, but they pack serious horsepower – a theoretical maximum of 4 TFlops of single-precision processing.   

Deskside supercomputing isn’t a new idea. A few years back, there was a company called Orion Multisystems, which tried to establish an overweight $100,000 desktop tower with 96 Transmeta Efficeon processors as a potent workstation, but failed. Nvidia tried to sell a $7500 Tesla (D870) deskside case with two C870 gen 1 cards (518 GFlops each) in late 2006, but canceled the device with the new, gen 2 Tesla family, because of the limited success of the D870.

However, that Nvidia deskside supercomputer is back, albeit in a revised and potentially much more useful form. The company has come up with a reference platform of what it imagines to be a capable Tesla desktop and convinced vendors such as Dell, Asus, Lenovo, Scan and Boxx to configure and sell such PCs under their brand names. Of course, the focus of these desktops is on Tesla cards, four cards to be exact. But these systems are fully functional PCs and do not need a separate PC to be operated, as it was the case with the original D870. Besides four C1060 cards, Nvidia recommends a quad-core CPU as well as 16 GB of memory, which, according to the company, should result in PCs with a price tag of slightly below $10,000.

If you are into numbers, such a PC would have 960 graphics cores and a combined performance rating of about 3.6 to 4 TFlops in single-precision and about 400 GFlops in double-precision applications. The C1060 GPUs (T10P processor) are clocked at 1.33 GHz and rated at a performance of about 900 GFlops in a single-unit configuration.

Depending on the application performed, Nvidia claims that a Tesla PC will be about 250 times faster than a regular desktop PC. However, we need to be fair and mention that this number also depends on the system the PC is compared to. For example, a PC equipped with an Nvidia SLI system or two of ATI’s Radeon HD 4870 X2 cards can compete with a Tesla PC in the Teraflops department. In fact, ATI says its tow 4870 X2 cards provide about 4.8 TFlops in single-precision and will actually outperform the four Tesla cards in double-precision applications with an estimated maximum performance of about 480 GFlops.  

However, the Tesla system outshines regular graphics cards in memory availability. 4 GB of GDDR3 800 memory is integrated in each GPU, which translates into a total of 16 GB of graphics memory for a Tesla PC. The combined memory bandwidth is 408 GB/s. All that processing power translates into substantial power consumption as well. The C1060 cards are rated at 160 watts each, which means that four cards and a capable processor can top 750 watts easily.     

The Tesla supercomputer seems to be a natural expansion of Nvidia’s Tesla product line and an opportunity for the manufacturer to sell more Tesla cards. We have no doubt that there are plenty of research organizations and corporations that could take advantage of a $10,000 “supercomputer”. 

I think you meant gpu not cpu?

Nov 18, 2008 02:16     

I think you mean GPU not CPU here? "4 GB of GDDR3 800 memory is integrated in each (CPU)"
BlueSohip report abuse vote down vote up Votes: -3

So..whats the point?

Nov 18, 2008 08:58     

Whats the point in this kind of machine? What kind of work would be done with it?
shinobi904 report abuse vote down vote up Votes: +0

not exactly a "Tesla PC"

Nov 18, 2008 17:31     

It is not exactly a "Capable Tesla PC" if it requires a X86 CPU to manage it. Now if it had an OS that ran on it without the need of a X86 CPU, THEN, it would stand alone as a "Tesla PC". Otherwise it is just a very powerful add in card. Sound more like marketing mumbo jumbo to get attention.
tomt report abuse vote down vote up Votes: +0

Still a Tesla PC...

Nov 19, 2008 03:19     

It's still a Tesla PC because the Tesla cards do all the real work. Who cares if a CPU distributes work to the cards? A CPU capable of distributing work costs very little, so its essentially a commodity. Not so when you try to go cheap on the GPU on this system. Saying it's not a tesla PC because a CPU distributes work is like saying your PC isn't a folding PC (folding@home) because it still relies on the folding master servers and your ISP to distribute the work to you.

ninja999 report abuse vote down vote up Votes: +0

ahhh...

Jan 16, 2009 08:53     

i just played with my junk.
jonny rocket report abuse vote down vote up Votes: -1