Announcing our VDAC 8-16 GPU Systems
RenderStream is pleased to announce our VDAC (Visual and Data Analysis Cluster) which uses the power of up to 16 GPUs to breeze through computational problems like those found in GPU based rendering, oil and gas, medical research, augmented reality and mobile visualization enhancement. (Application allowing)
Since 2006 RenderStream’s founders have been involved in GPU-based high performance computing. We recognized numerous markets that this acceleration could be of serious benefit. These markets are characterized by their need for high-performance computing appliances that are small, with sizes that are workstation or 1u to 20u in a rackmount system, powerful, four to fourteen teraflop (TFLOP) and relatively low cost for the system and relatively low cost to operate. The fastest dual multi-core CPU (MPU) can deliver around 0.2 TFLOP necessitating forty to hundreds of them to meet a given application’s needs and even more if the memory must be shared amongst individual computers in the cluster. The cost in space, resource and equipment restricts implementation of traditional computer clusters to all but the largest ventures which thus conservatively suppresses 60% to 70% of the potential total available market (TAM) from developing. Without HPC:
· It limits the ability of a small clinic or doctor’s office to diagnose life threatening diseases without sending out to large hospitals for expensive procedures that with the right appliance could be done sooner, cheaper, with less anxiety and stress from the wait and travel.
· It limits the ability of the small digital studios from effectively competing in the multi-billion dollar digital media industry.
· It restricts scientific data gathering and analysis.
· It hampers disease analysis and safe cure development and drug discovery.
· It stifles real-time analysis during oil and gas exploration.
· It hampers at the municipal level critical security applications and related data processing that enhanced image analysis and data mining can provide.
· It impacts financial and economic analysis.
To generally enable these types of results to a broader community, a number of things had to occur. We needed: hardware tuned to run the programs. Development of programming language tools and libraries that mimic capability used in programming MPUs and that are already used by non-graphic programmers to enable migration to the new technology. Finally and most importantly, tasks that needed to be and could be parallelized.
Both Nvidia and ATI (AMD) have been developing chips to work in a genral purpose GPU (GPGPU) environment with Nvidia placing an emphasis of GPU only solution and AMD working on a more integrated approach. Giving Nvidia solutions the edge is that they have the most highly developed programming language called CUDA. Prior to CUDA early pioneers in the GPU-Acceleration field were directly programming shaders making the problem being solved to be expressed in terms of the graphic language. CUDA greatly simplifies the process providing access to many groups that otherwise would not have attempted it. CUDA applications have been in development for a few years now and we’re starting to see various GPU based applications that tackle a variety of problems from high end rendering and animation, to medical research applications like tomography, solving electromagnetic field problems and protein folding to name just a few. But CUDA is not the only language and we look forward to OpenCL and other languages to mature and join the fray.
At RenderStream we are providing solutions with two to eight GPU’s, and we just introduced a sixteen GPU development system (Linux only). Our solutions are made using Nvidia Geforce, Quadro and Tesla cards. For even more speed we announced at the March 2010 ACM Siggraph meeting in Austin our VDACTr8 with eight single GPU cards and VDACTr8x2 solution using eight dual GPU GTX295 cards.
This latter system with each card having access to PCI-e 2.0 X16 lanes provides 10 to a theoretical 14.2 TFLOPs of single precision compute power. This system is ideal for speed and heterogeneous GPU-MPU computing environments where memory on the GPU card is not important.
If graphics memory and mission criticality dominate, we offer VDACTr8 with eight Tesla or Quadro cards as the application demands or a VDACTr8x4 Cluster using the VDACTr8 server and four Tesla S1070 or Quadro® Plex 2200 S4. The type of card or card-mix chosen depends on the application. For advanced visualization, augmented reality and immersive environments choose Quadro and Tesla. For instance with eight FX5800 cards on a single computer the VDACTr8 offers the most powerful single computer solution for powering eight HD projectors that are used for building advanced immersive training CAVES. But using eight Tesla C1060, the new C2050 or C2070 this becomes the most powerful 8U HPC compute node on the planet.
In any case building the right tool for the job is within our grasp. Please visit the website and contact us to see how we can help you attain your goals.