Contact us!
Find Us On Facebook

Free Ground Shipping

Pay by VISA, Mastercard, Discover, AMEX, Paypal or Bank Transfer

AMD Solution Provider
NVIDIA Partnerforce Program Member

NVIDIA's "Fermi" GPU Now Shipping

Original Article Date: 2010-04-06

By now pretty much everyone in the field of computing will have heard of the use of GPUs (Graphics Processing Units) in solving complex scientific and numeric calculations. Their main offering is the ability to perform massively-parallel calculations many times faster than conventional CPUs. This is because GPUs are comprised of hundreds of computing cores (instead of 2,4,6,8 or 12 cores in conventional CPUs), making them adept at performing large numbers of repeated calculations quickly using hundreds of parallel threads of execution.

The use of GPUs for general purpose computing applications (as opposed to graphics) has exploded in growth over the last 2 years through NVIDIA's previous GPU architectures. But the keynote of this new release is that Fermi is the first GPU to be designed specifically with general purpose computing in mind. The fact that it is also an extremely capable graphics processor is incidental!

Here's a summary of Fermi's new features:

  • 480 cores - double the number of the previous GPU generation.
  • Greatly improved double-precision performance - 7x the previous chip - 515Gigaflops per GPU in double-precision.
  • Full cache hierarchy - L1 and L2 caches are present. Each Streaming Processor Block (a set of GPU cores) has its own L1 cache, whilst a larger L2 cache is available for the entire GPU. These caches greatly improve performance on applications such as fluid simulation, physics solvers, sparse matrix multiplication, etc.
  • ECC Support, a necessity for reliable GPU clusters, is now fully supported on all core registers and both GPU cache levels, protecting computations from "soft" RAM errors induced by cosmic-ray disruption.
  • "Gigathread" intelligent program thread management - provides greatly improved thread flow, including a 10x improvement in context switching, concurrent kernel execution and improved thread-block scheduling.
  • Improved development tools - C, C++, OpenCL, DirectCompute and Fortran compilers are now available, as well as NVIDIA Parallel Nsight for Visual Studio.

The New Card Line-Up

At release, NVIDIA have just two graphics cards sporting the new Fermi core. These are:

  • GeForce GTX470 - 448-cores, 600MHz RAM Clock, 1200MHz GPU Clock, 320-bit 1280MB GDDR5 RAM, 215W Maximum Power Draw
  • GeForce GTX480 - 480-cores, 700MHz RAM Clock, 1400MHz GPU Clock, 384-bit 1536MB GDDR5 RAM, 250W Maximum Power Draw

There will be a set of "Tesla" cards which will have practically identical specs to the above GeForce Units, except they will come with quadruple the RAM (i.e. 6GB per GPU). These are expected to be released in Q3 2010.

The GeForce cards are now available for integration as either graphics or GPU cards, in any of our workstations, and also, specifically in our pair of GPU-optimized systems.


Fermi represents another leap forward in this important and rapidly growing segment of the computing market. So if you're already familiar with GPU Computing, then now is the time to upgrade. And if you're wondering whether GPU Computing could help your business or educational institution, then feel free to ask - we'll be glad to advise!

Best regards,

Ben Ranson
Chief Systems Engineer
Electronics Nexus