tesla computing questions

ive been interested in the c1060 tesla card. but unsure of its function… what i was wondering is this card cpu based or software based?? in other words will it automatically off load the cpu when the load is high or will it only function on certain softwares that enable the card? i also was kinda confused on the CUDA thing. it seems that what i’ve read you actually have to program the card to function if anyone can shed some light on these things it would be great.

to get to the point i want to build a personal supercomputer with this card but not sure exactly its functions… i was planning a high end i7 with this card and a pair of ati’s but then i heard that this card isn’t compadable with ati’s due to drivers unable to coexist in the same os… is this also true? if so i was going to use two 280 gtx’s and was wondering if i could sli the three?? once again plz help me figure out if i could build my dream

The offload happens on graphics operations encapsulated in DirectX calls, OpenGL calls, and such. Your graphics intensive programs (games, for instance) will detect the card and know how to use it if you installed it and its drivers properly.

CUDA gives you a way to program the card and use it to accelerate your own applications. If you’re not into these kind of nerdy things, then don’t worry about CUDA.

There are some threads here that deal with ATi/NVIDIA compatibility, but my advice is to avoid mixing the vendors. Also, the Tesla are professional cards, why not use a compatible GeForce? Your dream machine would be cheaper.

[quote name=‘liv’ date=‘Dec 9 2008, 10:41 AM’ post=‘474480’]

The offload happens on graphics operations encapsulated in DirectX calls, OpenGL calls, and such. Your graphics intensive programs (games, for instance) will detect the card and know how to use it if you installed it and its drivers properly.

would this also include os’s that highly graphical such as vista 64… thx for replying

also i 'm not to worryed about the cheapness as so much the quality… if you had any sudjestions i would appreaciate any

I’m not sure how Vista takes advantage of a graphics card, but the benefits of a powerful one are visible in high resolution, many-frames-per-second apps, like intensive games. The number one upgrade to make Vista happier is in the amount of RAM.

Perhaps the best GPU out there is ATI’s RV770. They make a board with 2 of them (I think it’s the Radeon HD 4870 X2). Recently Nvidia has announced a board with 2 GT200, see http://forums.nvidia.com/index.php?showtopic=83628. I know the GT200 used to need more power than ATI’s competitor, but that may change with the upcoming 55 nm generation.

I highly doubt that OpenGL calls get offloaded to a C1060. PhysX might be able to use the card in games, but apart from that I don’t think it will have any value. As far as I understand, only applications using CUDA can make use of the C1060.

You don’t need the C1060, you won’t be able to use the C1060.

Just buy regular video cards. If you want to use PhysX or CUDA, buy NVIDIA cards. (CUDA will run on a GTX280 just fine, actually even better.)

so seriously there is no performance increase in haveing the tesla in your system??
then why are they advertising personal super pc’s with this card on nvidia site?

To be honest because it makes nvidia look cool.

In fact, the whole Tesla brand exists, as much as for any other reason, to reassure companies who are apprehensive about using plebeian video cards to do serious work. It’s a mindgame :rolleye:. As are most things.

Internally, Teslas use identical chips as GeForces (but with more RAM and, supposedly, more thorough testing). They also have the video connector sawed off (again, to make them look more serious).

Because for applications written to take advantage of CUDA, it is a personal supercomputer. A system with 4 Teslas running HOOMD is as fast at molecular dynamics as a cluster with more than 128 processor cores. Many other CUDA applications get similar speedups.

As others have said, it won’t make windows or general applications any faster. Applications must be written to take advantage of it.

But even then, you could put four GTX280s in your PC and get the same benefit. For a quarter of the price.

i don’t think you can even have 4 gtx280’s in your system at once.

But thanks for the heads up guys it would be to big a pain to have to write code for everything im planing on doing

It is difficult, but 4 card workstations have been built:

http://fastra.ua.ac.be/en/index.html

http://www.nvidia.com/object/tesla_supercomputer_wtb.html

Sorry for misleading you truthsouless, I had no idea a Tesla C1060 only works in CUDA. Tesla series and the GeFroce series share the same chips, so I don;t understand why NVidia went that route. Anyway, better stick to geforce then.

As I said, to make it clear that the c1060 is “for profecionals.” Look how many people in the other thread wish it had a vga connector.