Multicore PC " supercomputer' VS GPU supercomputer In terms of using for trading, optimization, modelling, and backtesting. Could you list which side you would prefer? and Why?. Also, could tell us what is specs in your mind for any pick and under what budget you can build it?
Could we have a hybrid between GPU and CPU? So, if we have some calculations require sequential dependency so we target it to CPU, But if we have other set of calculations that require multithreading so we target for GPU. Is my question valid. Sorry for my illiteracy in this topic. I am here to understand.
Realistically, that kind of money will only get you a decent workstation with two GTX590's. Here's a public wish list on NewEgg that, for $3K (excluding taxes and shipping) will give you (NewEgg currently limits the GTX590 to 1 per customer, so you need to order a second one at an additional $750): http://secure.newegg.com/WishList/PublicWishDetail.aspx?WishListNumber=17928906 2 GTX390's, these are dual-GPU cards with 1.5GB per GPU. For $150 less per card you could go with an MSI GTX580 that comes with a single GPU but has 3GB. An i7-2600K and Asus P8P67 WS motherboard. What's nice about this board is it comes with an NF200, which will give you a full 16x per video card. 8 GB of DDR3-1866 CAS 9 RAM. 2 60GB SSDs, for a 120GB RAID0 system drive. 2 750GB SATAIII hard disks for a 1.5GB RAID0 data drive. An Enermax 1350w P/S since the two GTX590s will consume about 700w at load. A DVD RW, Noctua heatsink and Cooler Master Nvida mid-tower. The i7-2600K can overclock to 4.5GHz without breaking a sweat. At that speed it's comparable in performance to a $1k i7-990x for a lot of computing tasks. Tom's hardware did a decent comparison a few months back. All told, this is a pretty damned nice CUDA workstation and almost exactly what I'll be ordering myself in the next week or two.