First thanks for sharing your opinion. However, it isnot only for backtesting. It is also for optimization. If i am going to test on the tick " bid/ask" data " not minute or subminute" and my ATS containing huge math. So, i would prefer to cut days of backtesting into hours or minutes. However, Optimization is also my target. If i could do multi parallel optimization at once. This would be great to finish it in very less time than usual. Also, to rent the power for someone who wish to do some calculations and he cannot .. " Just an idea popped up right now in my head".. If someone wanted to optimize his system, he or she can get an access to the system to do what he wants. I think there are many applications for this power in the filed of trading. I wish i am a programmer to get more deeper. Thanks McGene
I agree that backtesting is not that crucial for forcasting any performance. However, i think filling factor could be added also relatively into your backtesting setting. The prespective of having a cheap supercomputer is different than to have a powerful server. I still feel i need a cheep supercomputer.
I want a cheap supercomputer just because they are there!! Wouldn't I sleep better knowing that it was there if I ever found a use for it? Just kidding, but it would be fun to build one, get the highest end Intel mobo with all the processing power and ram it could be stuffed with, use the Raid SSI drive thing or ramdisk or whatever actually works...
IMHO, before going whole-hog on a PSC, you should get a NVidia GTX560 card like this first: http://www.newegg.com/Product/Product.aspx?Item=N82E16814261109 What's nice is that it has double the memory, 2GB, compared to other GTX560's and memory is what's most important when it comes to CUDA. For $234 you can find out if CUDA is going to accomplish what you're intending to do. If CUDA is workable then you can upgrade to the Tesla series, which are true supercomputer GPUs, and these folks have a pretty flexible workstation with up to 3 Teslas. http://www.avadirect.com/workstation-pc-configurator.asp?PRID=14032 Depending on what you build it could run you $4,000 to close to $40,000.
May I suggest you buy a cooling system for the one you already have?. A baseball cap worn sideways should do the trick.
My budget for this project is under 3000$. This is the challenge. http://www.informationweek.com/news/201803520 http://www.llamma.com/xbox/beowulf.htm http://www.xyroth-enterprises.co.uk/bclust.htm http://searchdatacenter.techtarget.com/news/1112598/Supercomputers-come-off-the-shelf-skills-dont http://www.chessbase.com/newsdetail.asp?newsid=1013
http://www.hanpau.com/index.php?page=help_build_supercomputer http://www.topix.com/forum/city/rome-ny/T42JLB59MG8RG5CBQ http://www.zdnet.com/news/build-your-own-supercomputer/116815 http://fastra.ua.ac.be/en/index.html http://fastra2.ua.ac.be/
I may be wrong but it looks like that beowulf cluster only runs in linux. Does your back-testing/optimizing software run under linux?
You can run Windows apps on top of Linux using a hypervisor platform like KVM or Virtualbox. The real question is would running Windows in a hypervisor be able to utilize the cluster in any useful way? You could run multiple virtual Windows machines simultaneously I suppose, with each backtesting different scenarios. Of course it would be better to do this in a native Linux app but I haven't found such a platform yet.