Nvidia Corp. along with its partners Appro and Supermicro on Tuesday said they had initiated shipments of servers designed for high-performance computing featuring the new generation of Tesla computing boards featuring Tesla T20 processors. The new chips offer very high double-precision computing performance.
The Appro 1U Tetra GPU server is offered in two models; the Appro 1426G4 server hosted by the latest Intel Xeon processor supporting up to 96GB of memory and the Appro 1326G4 server hosted by the latest AMD Opteron processor supporting up to 128GB of memory. Both models feature four Nvidia Tesla C2050 that offer 12GB of ECC GDDR5 memory support with a total of 1792 stream processors (448 stream processors per card) delivering double precision performance of over 2TFlops of computing power per server.
“The launch of the Appro 1U Tetra GPU Server with 4 Tesla 20-series GPUs illustrates Appro’s innovation and leadership in high density computing systems. The Appro 1U Tetra GPU server is the first platform that provides two CPUs and four GPUs in a 1U server package offering a terrific value proposition to HPC customers who need high density and supercomputing performance at lower price points,” said Andy Keane, general manager of the Tesla business unit at Nvidia.
Supermicro's new GPU-based HPC product line features both Intel Xeon 5600-series (Westmere) processor as well as AMD Opteron 6100-series (socket G34) processors, along with advanced remote monitoring and management capabilities that data center IT staff require and are also optimized for a wide range of graphics and computationally intensive applications. Supermicro's 6016GT-TF-FM205 1U Tesla-based server features two Nvidia Tesla C2050 GPUs via two Gen2 PCI-Express x16 connections. Additionally, Supermicro offers 2U twin-GPU server with hot-swappable devices and onboard QDR InfiniBand for 40 Gb/second high-bandwidth connectivity. The total double precision computing performance of Supermicro’s new HPC-oriented servers is around 1TFlops.
"Supermicro systems optimized for the new NVIDIA Tesla M2050 GPU computing module enable seamless integration of GPUs into large, scale-out deployments in commercial and scientific computing data centers," said Mr. Keane.
Although commercial launch of actual products based on Nvidia Tesla T20-series computing processors happens around eight months after the formal announcement of the Fermi architecture, which is optimized for computing, it is important to note that the HPC-oriented servers from Appro and Supermicro are off-the-shelf products ready for deployment. Previously, companies and institutions interested in using GPGPU technologies (general purpose computing on graphics processing units) should acquire ATI FireStream or Nvidia Tesla processing boards and install them into actual servers themselves.
The Tesla C2050 and C2070 computing processor cards will retail for $2499 and $3999. At present Nvidia only ships the model C2050 with 3GB of memory, whereas the C2070 with 6GB of memory should be available later.