newbie questions

I want to research whether it is feasible to use GPU for my company goals.
Couple questions:

From learning perspective are there differences of Tesla vs game cards (GEFORCE/Quadro)? Is it reasonable assumption early development phases can be done on cheaper system and then moved into advanced (Tesla-based) platforms?
What Compute Capability at this page http://www.nvidia.com/object/cuda_gpus.html means - performance or some advanced (new) functionality?

Thank you

Yes, the GeForce cards are excellent and inexpensive ways to learn CUDA. The high end GeForce cards (GTX 470/480) are comparable in performance to the Tesla cards (with one caveat), and even slightly faster for some things.

The current Tesla series (C2050 and C2070) differ from the GeForce GTX 400-series cards in a few ways, listed here:

http://forums.nvidia.com/index.php?showtop…amp;pid=1073830

The key differences are more GPU memory, 4x faster double precision performance (but similar single precision performance to GeForce), option to use ECC with GPU memory and better quality assurance testing.

For learning/evaluating CUDA, these differences usually are not important. If your code is limited by double precision performance, then you’ll have to keep in mind the 4x difference between GeForce GTX 470/480 and Tesla C2050/2070.

Every time (well, almost) NVIDIA adds a new set of hardware features to the architecture, they increase the compute capability number. These new features sometimes increase performance of existing code, and sometimes they just open up the possibility of doing a new class of computation efficiently on the GPU. For the most part, compute capabilities build on each other, so that each contains all the features of previous compute capabilities.

To give you an idea of how this works, a brief and non-exhausive summary of the compute capability history looks something like this:

Compute capability 1.0: Original CUDA architecture

1.1: Atomic operations in global memory

1.2: Smarter memory controller with relaxed rules for good memory bandwidth, atomic operations in shared memory

1.3: Native double precision

2.0: L1/L2 on-chip cache, concurrent kernel execution, major changes to eventually support all of C++ in device code

2.1: Rebalancing of CUDA cores per multiprocessor and changes in how instructions are scheduled (unlike the previous updates, this update seems to have been made to reduce the costs of the compute capability 2.0 design and does not appear to improve performance at all)

Yes, the GeForce cards are excellent and inexpensive ways to learn CUDA. The high end GeForce cards (GTX 470/480) are comparable in performance to the Tesla cards (with one caveat), and even slightly faster for some things.

The current Tesla series (C2050 and C2070) differ from the GeForce GTX 400-series cards in a few ways, listed here:

http://forums.nvidia.com/index.php?showtop…amp;pid=1073830

The key differences are more GPU memory, 4x faster double precision performance (but similar single precision performance to GeForce), option to use ECC with GPU memory and better quality assurance testing.

For learning/evaluating CUDA, these differences usually are not important. If your code is limited by double precision performance, then you’ll have to keep in mind the 4x difference between GeForce GTX 470/480 and Tesla C2050/2070.

Every time (well, almost) NVIDIA adds a new set of hardware features to the architecture, they increase the compute capability number. These new features sometimes increase performance of existing code, and sometimes they just open up the possibility of doing a new class of computation efficiently on the GPU. For the most part, compute capabilities build on each other, so that each contains all the features of previous compute capabilities.

To give you an idea of how this works, a brief and non-exhausive summary of the compute capability history looks something like this:

Compute capability 1.0: Original CUDA architecture

1.1: Atomic operations in global memory

1.2: Smarter memory controller with relaxed rules for good memory bandwidth, atomic operations in shared memory

1.3: Native double precision

2.0: L1/L2 on-chip cache, concurrent kernel execution, major changes to eventually support all of C++ in device code

2.1: Rebalancing of CUDA cores per multiprocessor and changes in how instructions are scheduled (unlike the previous updates, this update seems to have been made to reduce the costs of the compute capability 2.0 design and does not appear to improve performance at all)

Thank you!

I guess I can $300 expenses covered than. BTW - Geforce does not have Compute capability data - does 470/480 support 2.0?

Thank you

Thank you!

I guess I can $300 expenses covered than. BTW - Geforce does not have Compute capability data - does 470/480 support 2.0?

Thank you

GTX 470/480 support CC 2.0.

GTX 460 supports CC 2.1.

GTX 470/480 support CC 2.0.

GTX 460 supports CC 2.1.

And, at least for the time being, I would avoid the GTX 460 until the performance oddities with CC 2.1 are better understood. (The hardware scheduler relies on instruction-level parallelism in your CUDA code to a greater extent than previous compute capabilities, leading to some performance regressions.) The GTX 470 hits a pretty nice performance/price point and gives you access to all the interesting new features of the Fermi GPU.

And, at least for the time being, I would avoid the GTX 460 until the performance oddities with CC 2.1 are better understood. (The hardware scheduler relies on instruction-level parallelism in your CUDA code to a greater extent than previous compute capabilities, leading to some performance regressions.) The GTX 470 hits a pretty nice performance/price point and gives you access to all the interesting new features of the Fermi GPU.

Thank you.

Not sure my next question is in the right forum, but if you know where to direct me…

Looks like (silicon-based forms of) life changed a lot since my last DIY computer.

I thought I can put video card into me workstation - but just before ordering I noticed there is a power supply requirement (550W) on the 470 card. Mine have 305W (Dell precision).

Is there anything else I need to know prior to ordering (special connectors, voltage)?

Again, sorry for bringing hardware questions into this forum, but I’m not sure what is the better place to ask these questions.

Thank you.

Not sure my next question is in the right forum, but if you know where to direct me…

Looks like (silicon-based forms of) life changed a lot since my last DIY computer.

I thought I can put video card into me workstation - but just before ordering I noticed there is a power supply requirement (550W) on the 470 card. Mine have 305W (Dell precision).

Is there anything else I need to know prior to ordering (special connectors, voltage)?

Again, sorry for bringing hardware questions into this forum, but I’m not sure what is the better place to ask these questions.

The two things you have to worry about with high end graphics card are power and space. The GTX 470 requires two 6-pin PCI-Express power connectors directly from the power supply to the card itself. Pretty much any power supply of sufficient capacity to power a GTX 470 should include(at least) two such power cables.

The second problem is space. The GTX 470 is a nearly full length and double-wide PCI-Express card. The card will take up two adjacent slots and extend across the motherboard. Computers built from standard parts usually don’t have a problem with this, but highly integrated system, like those from Dell and HP, often cut corners that make installing a high-end GeForce impossible. It can be very hard to tell from the web whether or not a high-end GeForce will work in such a system, unless they offer it as a build option.

Your best bet is to find a vendor that will ship the computer with the GTX 470 already installed, or to find a vendor where you can talk to real people and explain what you need. Tell any competent computer builder you want to install a GTX 470, and they should know what to do.

The two things you have to worry about with high end graphics card are power and space. The GTX 470 requires two 6-pin PCI-Express power connectors directly from the power supply to the card itself. Pretty much any power supply of sufficient capacity to power a GTX 470 should include(at least) two such power cables.

The second problem is space. The GTX 470 is a nearly full length and double-wide PCI-Express card. The card will take up two adjacent slots and extend across the motherboard. Computers built from standard parts usually don’t have a problem with this, but highly integrated system, like those from Dell and HP, often cut corners that make installing a high-end GeForce impossible. It can be very hard to tell from the web whether or not a high-end GeForce will work in such a system, unless they offer it as a build option.

Your best bet is to find a vendor that will ship the computer with the GTX 470 already installed, or to find a vendor where you can talk to real people and explain what you need. Tell any competent computer builder you want to install a GTX 470, and they should know what to do.

Thank you. Looks like I should buy a gaming system.

Thank you. Looks like I should buy a gaming system.