Fermi or not Fermi that is the Question
As the room held its breath, Nvidia CEO, Jen-Hsun Huang gave the audience what it had come to see today at the GTC event in San Jose, announcing he was "incredibly proud" to show off the firm's latest GPU architecture, Fermi.
"Fermi is the soul of a supercomputer in the body of a GPU," said Jen-Hsun as he showed off the gleaming GPU for the first time. "It's an absolute powerhouse," he gushed, telling the audience Fermi had three billion transistors and sported 512 cores.
Fermi, said Jen-Hsun, has six 64-bit memory partitions, for a 384-bit memory interface, supporting up to six GB of GDDR5 DRAM memory as well as boasting native support for C++ and support for ECC.
Nvidia's CEO explained it had taken "massive investment and risk" to build the GPU from the ground up, but that the results were well worth the struggle.
"We've learned that we want to make our GPU applicable and useful to a larger footprint of applications and uses," he said, plugging Cuda to the developers in the crowd and emphasising: "you need real tools, real programming tools.
"Many of you said you wanted more double precision," he said, revealing that Fermi actually speeds up double precision five times, and that's just on first silicon. It should be able to reach a speed-up of a factor of eight in the near future, he said.
