Nvidia Tesla scores IBM GPGPU server win

May 20, 2010 // 2:37 p.m.

Tags: #computing #cuda #fermi #gpgpu #hpc #ibm #nvidia #parallel #server #tesla

It's been a tough few months for Nvidia, with Fermi facing tough competition upon launch and one staunch allies such as BFG dropping out of the graphics card market.

There is some good news for the green team though; Nvidia recently scored a design win with IBM's iDataPlex servers, and Nvidia Tesla 2050/2070 GPU computing modules, which use its latest Fermi GF100 core.

HPC, workstation and server products typically carry higher margins than consumer products, enabling companies to make more money - something AMD and Intel have benefitted from for years with their Opteron and Xeon CPUs. Prior to the launch of Fermi, Nvidia talked about its desire to focus on GPGPU, and getting IBM on board is a big sign that Nvidia's strategy might pay off. IBM claims it can fit "38,000 processing cores in a single rack", with each 2U server contains two Tesla cards and two Intel Xeon 5600 CPUs.

What's interesting is that despite the adoption of Tesla by IBM - and the fact it has promo videos on it site (which you can see below), the official specs page for its dx360 M3 server make no mention of Nvidia GPGPUs until you get right down into the features and benefits tab, where it's an additional "solution integration" rather than a main selling point. The point is further enforced in IBMs own video demo.

Do you think Nvidia will make it big with Tesla? Will its focus change away from PC gaming or is it always going to support it? Let us know your thoughts in the forums.