This week Nvidia announced the shipment of two Tesla products: the C1060 GPU for Dell's Precision R5400, T5500, and T7500 workstations, and its ready-to-deploy preconfigured clusters featuring the Tesla S1070 1U GPU. Not to be confused with the American heavy metal band (although they're Tuan's favorite so we hear), Tesla is the company's first dedicated general purpose GPU, allowing it to serve alongside the main GPU and help carry some of the processing burden. However, unlike mainstream Nvidia GPUs that can also serve as general purpose GPUs, the Tesla processors are more on the high end, built for high computational performance rather than awesomely cool graphics in Crysis.
As the months wear on, consumers will begin to hear more about "supercomputing." Just yesterday, Asus released its "supercomputer" motherboard, the P6T7 WS, equipped with two Nvidia nForce 200 chips and seven PCIe slots. The board was built for CUDA processing, and the company even said that four CUDA cards could actually break four teraflops of performance; end users would need at least one Quadro graphics card. Of course, this doesn't come close to Nvidia's Tesla Preconfigured Clusters, with configurations starting at 16 teraflops of performance.
“There are 15 to 20 million engineers, scientists and researchers around the world struggling for time on supercomputers, which has led to a huge pent-up demand for computation,” said Andy Keane, general manager of the Tesla business at NVIDIA. “With the launch of the Tesla Preconfigured Cluster, every one of them can easily deploy a GPU-powered supercomputing cluster that dramatically reduces their power consumption while still advancing the pace of their work.”
So what exactly is this cluster thing? It's actually an off-the-shelf, ready-to-deploy IT infrastructure system consisting four or more four-Teraflop 1U systems. This package enables IT managers and researchers to add GPU-based computing capabilities to existing data center systems. According to the company, these "preconfigured clusters" dish out 30 times the performance than CPU-only solutions, and actually consumes less power, saving the end users wads of cash. Nvidia also said that the Clusters comprise of x86 CPU servers coupled with Tesla S1070 1U (1.75-inch) GPU systems. With the basic Cluster starting at 16 Teraflops (four Tesla S1070 1U systems), all systems include host servers, infiniband switches, cabling and are fully customizable.
As for the Dell Precision Workstations, end users will have their own "personal supercomputer" with the equivalent computing power of a cluster, but without the super hefty price tag. The R5400, T5500, and T7500 workstations utilize the Tesla C1060 GPU clocking in at 1.3 GHz, and features up to 4 GB of GDDR3 memory, a 512-bit memory interface, 240 streams, and a single precision floating performance peak of 933. In comparison, the Tesla S1070, used in Nvidia's Tesla Preconfigured Clusters, clocks in at 1.296 to 1.44 GHz, offers 16 GB of total dedicated memory, uses a 512-bit memory interface, and has a single precision floating performance peak of 3.73 to 4.14 Teraflops.
Needless to say, there's a lot of power going on whether it's a Tesla-charged Dell Precision workstation, or a Tesla Preconfigured Cluster from Nvidia. For IT managers looking to expand, head here to find out where to purchase one of Nvidia's Clusters. For smaller Dell-based workstation, head here to view Dell's Precision section.