Sign in with
Sign up | Sign in
Your question

Questions about Nvidia Tesla cards? Look here first

Last response: in Graphics & Displays
Share
December 6, 2010 6:48:45 AM

Hey Everyone,

I've had a lot of questions lately about the Tesla C1060/2050 parallel computing cards from our clients and figured I'd chime in to help out. I'll try and keep it short and answer the simple questions.

Overclocking: The Tesla cards are not meant/designed to overclock as they are most stable at the stock speeds. If you are lacking the speed you need, there may be a couple reasons. You need to have sufficient RAM and an adequate processor to compliment the Tesla cards. If on of these are slow, it will slow down your Tesla. Don't use cheap parts.

SLI: These cards do not need an SLI bridge between them. As long as the are plugged into a PCI Express 2.0 lane they should run fine.

Cooling: This seems to be a huge topic lately. Yes, these cards get pretty damn hot when under heavy loads. The way I combat this heating is by choosing a large case and high CFM fans. We have had 4 Tesla cards in a single system and this sky rocketed the heat. We ended up placing 4 ~220 CFM fans within the system to allow circulation. We also ONLY choose full tower cases as our experience has showed us that mid tower cases don't allow the hot air to move around sufficiently. With a full tower, it allows for more room for air circulation. In total, most of our systems ship with 4-6 fans in them usually with high CFM ratings (depending on the build).

Why these cards?: These cards are designed for single and paralleled computation only. They are meant to calculate at incredibly high speeds handling lots of information of data. Don't use these for gaming as there are no display ports anyways! What you really want to be looking at within these cards' specifications is the amount of memory bandwidth, peak performance and the amount of "CUDA" cores (processor cores).

If you have any other questions, fire away. I'll share as much of my experience through our builds as I have. Thanks!
May 20, 2011 5:10:17 AM

rmcbiosciences said:
Hey Everyone,

I've had a lot of questions lately about the Tesla C1060/2050 parallel computing cards from our clients and figured I'd chime in to help out. I'll try and keep it short and answer the simple questions.

...

If you have any other questions, fire away. I'll share as much of my experience through our builds as I have. Thanks!



I'm having trouble getting a Tesla C2050 to coexist with a GeForce video card. When they are both installed, CUDA enabled software bypasses the Tesla and uses the GeForce card's GPUs. The only workaround I've found is to use an ATI Radeon card for video and the Tesla works fine for computation.

I'd love to find a way for the Tesla to work WITH a GeForce card and have them BOTH crunch numbers for me. Any ideas?

Same problem in both WinXP 32bit and Win7 64bit.

Thanks,
Chris
m
0
l
May 28, 2011 5:30:34 PM

Hey Chris,

So this is sort of a tricky solution that involves the code of the program your using. Run Device Query from your SDK. It should come up with some device ID's such as Tesla (0). The Tesla Card needs to be changed from it's default state to Device ID (1). The only rough part of this is that is has to be done by programming it into the code of the software that you are using. Usually there is some documentation on the CUDA program you are using and will sometimes have how to change device ID's. In any case, using multiple devices typically has to be enabled by re-working some of the code.

Hope this helps!
m
0
l
!