NVIDIA Tesla M2090 Releases the Most Powerful Universal Computing Card

On the same day that the GeForce GTX 560 was released for the gaming market, NVIDIA's Tesla product line for general-purpose computing and high-performance computing has also been updated for the first time in nearly a year. For large users such as scientific computing and supercomputers, NVIDIA has released the Tesla M2090, the most powerful universal computing card on the market.

The Fermi architecture Tesla was released as early as 2009, but due to limitations in power supply, cooling, etc., the Tesla C2070/C2050 for workstations and the Tesla M2070/M2050 for data centers and supercomputers all restricted the GPU. Two sets of SM only enable 448 CUDA cores. Nowadays, the GeForce GTX 580 in the game market has already opened a complete 512 CUDA cores by using the GF110 core. The Tesla M2090 is its corresponding high-performance computing version.

The Tesla M2090 offers 512 parallel processing cores, double-precision floating-point peak performance of 665 GFlops, and single-precision floating-point peak performance of 1331 GFlops. With 6GB GDDR5 cache, the memory bandwidth under ECC check is 177GB/s. Because it is for data centers, supercomputers and other users, like the M2090 and the M2070/M2050, there is no fan in the heat dissipation design, and only the heat pipe heat sink relies on the fans in the rack to dissipate heat. Of course, the display output interface will certainly not exist on such a general-purpose computing card.

According to NVIDIA's official tests, four Tesla M2090 plus four CPU systems can perform 69 nanoseconds of simulations per day in the AMBER 11 molecular simulation. For a supercomputer that used only a CPU, the maximum record was only 46 nanoseconds per day.

At present, Tesla M2090 has been installed in HP's HP ProLiant SL390 G7 high-performance computing 4U server, which can carry up to 8 Tesla M2090 and two CPUs.