40 Tflops 3U rackmount using eight Tesla K40s
Built by Oscar on 03/26/15Share:
Asus System ESC8000 G3 3U Server
2x Intel Xeon E5-2660 V3 2.6GHz Ten Core 25MB 105W
8x Crucial DDR4-2133 16GB ECC Reg. (128GB Total)
Samsung 850 Pro 1TB SATA 6Gb/s 2.5inch SSD (Primary)
4x Samsung 850 Pro 1TB SATA 6Gb/s 2.5inch SSD (Storage)
Ubuntu 14.04 LTS Installation w/ CD (64-bit)
8x NVIDIA Tesla K40 PCI-E 12GB (Passive)
This 3U, 8 x NVIDIA Tesla K40 system has an incredibly high compute density. With a peak performance of over 13 TFLOP/s double precision and over 40 TFLOP/s single precision this system is powerful enough to have been ranked on the top 500 SuperComputer list as recently as 2007. In comparison, the 2007 supercomputer it would have beat had over 200 nodes occupying at least 5 full racks and cost over 1 million dollars!
With today's technology, you don't need nearly the budget or physical space to get this amount of compute power. This particular system is intended for use in GIS (Geospatial Image System) analysis, but there is a long and growing list of real-world applications that can benefit from NVIDIA GPU acceleration. While not a complete list by any means, these applications include:
Scientific -- Molecular Dynamics, Quantum Chemistry, Bioinformatics, Numerical analysis, Physics
Machine Learning -- Artificial Neural Networks, DNN, Bayesian analysis, "Deep Learning"
Engineering -- CFD (Computational Fluid Dynamics) FE (Finite Element) Structural Mechanics, Electronic Design
Image analysis -- GIS, Video analytics, Signal analysis
Geoanalysis -- Weather prediction, Seismic data processing
Computational Finance -- Analytics, pricing, risk analysis, derivative pricing etc..
Manufacturing -- 3D Rendering, Design, Ray Tracing
Media and Entertainment -- Rendering and Animation, Image manipulation, Compositing and Visual effects, Encoding
Let us know what you need and we'll send you a quote for the right PC for you.