by Anton Shilov
05/27/2010 | 02:30 PM
A high-ranking executive Silicon Graphics Inc. (SGI) said in an interview that the latest Nvidia Corp.’s Tesla T20 computing processors based on Fermi architecture dissipate so much heat that SGI was looking forward alternative products that could substitute Nvidia’s chips inside the company’s high-performance computing (HPC) servers.
“AMD is catching up very quickly with its ATI graphics chips. ATI's high end GPGPU cards are only barely nosed out by Nvidia's Fermi based Tesla boards in some applications. There will be even performance capability between AMD/ATI and Nvidia within the next 18 months,” said Bill Mannel, SGI's senior director of server product marketing, in an interview with The Inquirer web-site.
Mr. Mannel said that computing processors with very high heat dissipation – such as Nvidia Tesla C2050/M2050 that have 225W thermal design power – create additional amount of work since those modules require more advanced cooling solutions. The increased temperatures also result in “worse failure profile” as well as increased costs due to necessity to spend more on design, manufacturing and maintenance. Finally, in certain cases overheating graphics cards force end-users to scale up the cooling infrastructure" to meet the higher ambient temperature demands in HPC data centers.
According to Mannel, when SGI first evaluated GPGPUs, Nvidia appeared to be a better solution, but now things are looking rather different. The report claims that SGI is looking for alternatives for Nvidia Tesla and AMD’s FireStream seems to be the most obvious choice. Intel Corp.’s Larrabee-derivative for HPC seems to be far away, as a result, FireStream has a chance.
In fact, graphics processors developed by ATI offer higher raw power in terms of TFLOPS compared to Nvidia’s chips. However, Nvidia has been aggressively promoting its CUDA programming platform, which allows to create GPGPU software for many applications, including those that are used for financial analysis, oil and gas explorations and so on. It will take a lot of time for AMD to catch up in terms of tools for its FireStream computing solutions, until then, SGI and others will not have an easy-to-use alternative for Nvidia Tesla.