Appro International and Super Micro Computer, or Supermicro, are unveiling new and enhanced servers that incorporate Nvidia's Tesla 20 series graphics technology.
These servers are the latest to bring traditional CPUs from Intel and Advanced Micro Devices together with GPU technologies to offer businesses and HPC (high-performance computing) environments greater compute density and energy efficiency.
Nvidia and AMD, through its ATI graphics business, have aggressively pushed GPUs for mainstream computing. That is a direction being embraced among customers, particularly in the HPC space, according to John Lee, vice president of advanced technology solutions at Appro.
The Tesla 20 series is based on Nvidia's new "Fermi" architecture, which is designed to offer reliability features normally found in traditional CPUs.
"We really believe the GPUs are here and are going to make a big splash," Lee said in an interview. "They will be here for a long time."
Appro has been active in developing systems that bring together CPUs and GPUs. The company in 2009 unveiled its HyperPower Cluster, which includes Intel- and Nvidia-based systems.
On May 4, Appro officials announced an expansion of the company's GPU-based efforts in the form of the Appro Tetra server, a 1U (1.75-inch) system powered by two CPUs and four Nvidia Tesla M2050 GPUs. It includes such features as six hot-swappable 2.5-inch hard disk drives, one PCIe expansion slot, integrated IPMI 2.0 remote server management capabilities, high-speed I/O options and a Microsoft Windows or Linux operating system. Energy-saving software also can power down pairs of GPUs when needed.
The Tetra comes in two models: the 1426G4 server with Intel Xeon 5600 chips and the 1326G4, with AMD Opteron 6100 processors. Both models offer 12GB of memory, 1,792 GPU cores and more than 2 teraflops of computing power.
Lee said the Tetra will be good for businesses interested in trying out GPU computing for the first time, or those interested in expanding their GPU environments.
In addition, Appro rolled out a GPU expansion blade also powered by the Tesla M2050 GPU as a configuration option in its GreenBlade System.
Appro's GreenBlade System is designed for high performance and power efficiency. The 5U (8.75-inch) system supports up to 10 server blades. When configured for a CPU-GPU combination, the GreenBlade System can run five two-socket CPU server blades and five GPU expansion blades that hold up to 4,480 cores, which Appro officials said translates into 5 teraflops of performance. The gB222X server is hosted by Intel Xeon chips, the gB322H by AMD's Opteron processors.
With the GreenBlade System, Lee said, IT administrators can dynamically power down the GPUs when running CPU-based applications, rather than having to shut the system down and take out the GPUs as had to be done previously.
Supermicro is rolling out its second-generation GPU servers, powered by Tesla M2050 GPUs. In addition, they feature Intel's Xeon 5600 and AMD's Opteron 6100 CPUs, as well as remote monitoring and management capabilities.
One of Supermicro's offerings, the 6016GT-TF-FM205, is 1U and supports two Xeon 5600 series chips and two Tesla M2050 GPUs.
The 7046GT-TRF-FC405 is housed in Supermicro's new 4U (7-inch) SC747TQ-R1400 rack mount chassis. Supermicro also is rolling out the 2U (3.5-inch) Twin GPU server.
Nvidia is continuing its mainstream GPU drumbeat. On April 29 in a column on Forbes.com, Bill Dally, Nvidia chief scientist and vice president, argued that Moore's Law was reaching the end of its life on CPUs, and that the only way to keep it going was through the adoption of parallel processing offered in such technologies as GPUs.
AMD is looking to put full computing and graphics capabilities on a single chip in its Fusion strategy, while Intel is expected to continue to expand the graphics capabilities of its CPUs.