It took some time to try to line up the desired and compatible components. Many specs and discussion forums were consulted as homework - now to see if all will fit! All the parts have been ordered and we're waiting for components to arrive. Already have the motherboard and CPU (and of course the Tesla).
What a beautiful mobo (= motherboard :) : ASUS P6T7
From i4u.com:
Asus launched the ASUS P6T7 WS SuperComputer motherboard for CUDA Parallel Programming. The Asus P6T7 is equipped with 7 PCI-E Gen2 x16 slots supporting up to three NVIDIA Tesla cards and one NVIDIA Quadro card.
It is certified for NVIDIA Tesla GPU computing, based on the NVIDIA CUDA parallel computing architecture that supports up to three NVIDIA Tesla cards and one NVIDIA Quadro card. Each GPU delivers 240 CUDA parallel processing cores, producing 1 teraflop of processing power simultaneously. This adds up to a total of 960 parallel processing cores, delivering 4 teraflops of processing power.
Capable of supercomputer performance 250 times faster than that of a standard PC, the P6T7 WS SuperComputer is an efficient solution for highly computational applications in the fields of life science, engineering science, financial modeling, electronic design automation and virtualization.
http://www.i4u.com/28315/new-asus-p6t7-cuda-motherboard-gives-you-4-teraflops
From bit-tech.net:
ASUS, the leading producer and innovator of motherboards, today has launched the most revolutionary and powerful Parallel Computing solution available to date: the ASUS P6T7 WS Supercomputer motherboard. Equipped with 7 PCI-E Gen2 x16 slots, the P6T7 WS Supercomputer has been certified for NVIDIA® Tesla™ GPU computing which is based on the NVIDIA® CUDA™ parallel computing architecture, supporting up to three NVIDIA Tesla cards and one NVIDIA Quadro card. This amounts to a total of 960 parallel processing cores providing a massive 4 teraflops of processing power—delivering 250* times faster computing and graphics processing performance. The ASUS P6T7 WS Supercomputer also provides users with flexible high speed I/O, graphics and computing options.
http://www.bit-tech.net/news/industry/2009/05/14/asus-p6t7-ws-supercomputer-workstation-moth/1
More Asus P6T7 revierws:
http://www.motherboards.org/reviews/motherboards/1913_1.html
http://hothardware.com/News/Asus-Debuts-P6T7-WS-SuperComputer-Motherboard/
http://www.tomshardware.com/news/Asus-supercomputer-motherboard-P6T7,7720.html
Trivia: Asus makes laptops for Apple, Dell, Falcon Northwest and Hewlett-Packard. This OEM manufacturing may be taken over by Pegatron in the companies restructuring - In January 2007, ASUS split into three distinct operational units: ASUS, Pegatron and Unihan).
More about the CPU chip in the next post...
- Acknowledgements:
- This equipment is funded by research grants from the Carnegie Foundation and South Africa's National Research Foundation (NRF) .......................................... For more on high performance computing in SA, visit www.chpc.ac.za or School of computational and applied math (CAM) @ wits: http://www.wits.ac.za/academic/science/cam/6327/staff.html . . A light introduction to parallelisation was presented in a couple of seminars. Here are some slides: http://www.dianewilcox.net/GPUFinance_Wilcox_April2011.pdf
Tuesday, November 23, 2010
Thursday, November 18, 2010
The first component
First component: Nvidia Tesla C2050 capable of up to 1 teraflops. The first teraflop machine was built by Cray in 1995. By 1997 this was most cheaply achievable with a beowolf cluster at a cost of 30 000 USD per gigaflop.
http://www.nvidia.com/object/product_tesla_C2050_C2070_us.htmlFLOPS
[Wikipedia.com] In computing, FLOPS (or flops or flop/s) is an acronym meaning FLoating point OPerations per Second. The FLOPS is a measure of a computer's performance. The first exaflops machines are being planned for 2018. In 2010 the fastest six-core processor PC has a theoretical peak performance of 107.55 GFLOPS (Intel Core i7 980 XE) in double precision calculations.
The current worlds fastest computer [China] :
The Tianhe-1A supercomputer uses 7,168 Fermi GPUs to churn through over 2.5 petaflops.
The current worlds fastest computer [China] :
The Tianhe-1A supercomputer uses 7,168 Fermi GPUs to churn through over 2.5 petaflops.
yottaFLOPS | 1024 |
zettaFLOPS | 1021 |
exaFLOPS | 1018 |
petaFLOPS | 1015 |
teraFLOPS | 1012 |
gigaFLOPS | 109 |
megaFLOPS | 106 |
kiloFLOPS | 103 |
Subscribe to:
Posts (Atom)