Poker-AI.org

Another interesting service: http://www.picloud.com/
I didn't check the prices and how they compare to other services such as Amazon EC2.

I wonder how well something like piCloud would work for CFRM. It would be incredibly slow to update your regrets, but you could run many many iterations in parallel.

what sort of number crunching do you have in mind? simulating all atoms in the universe?

evaluating possible hands?

what is the number of computations you want per second?

Have you considered impact of montecarlo on your planned computation? is 0.01% accuracy reasonable trade-off for the reward of 500ms calculation compared to 20min calculation?

raw cpu power if you need bulk is cheaper online (via hosting providers, either cloud or arrays of dedicated), than actual hardware you may want to purchase.

what is the lifespan intended use of this number cruncher? perhaps 'hosted' solution is more expensive if you plan to run cruncher for 12 years..?

average i5 should be enough for any crunching you need in less than 1000ms, when it comes to poker evaluations.

I don't agree. Depending on your needs and the time requirements, online can be quite a bit more expensive than purchasing a server/workstation outright.

especially when you can build a good machine as cheap as this

I ended up paying just under ~£950 for an i7 3820 and Asus P9X79 with 32GB of 2133mhz RAM, upgradable to 64G if it looks like my plan is working.

Kickass machine, over-clocked to just over 4.5ghz. Massively faster than my AMD machines at FPU calcs.

I basically run it 24/7 at just under 100% CPU. It wouldn't take long for that to add up if I was paying for it remotely.

Maybe you need this ?
http://www.techhive.com/article/2043985 ... rency.html
Don't forget the swimming pool

I ran across this while poking around:
http://www.nallatech.com/Latest-News/na ... -2008.html

It looks like you can write a custom algorithm into a FPGA and interface it directly with a Xeon socket. Wouldn't that be fun for CFRM.

Although, from the looks of it, the memory interface is kind of slow:
http://www.nallatech.com/Intel-Xeon-FSB ... stems.html

By comparison, DDR2-800 offers 12.8 GB/s efficiency. Not sure how badly that would affect performance.

what do you think about this ?

http://www.hardware-attitude.com/fiche- ... 92-go.html

192Go ram !!

Not to forget you have to learn a hardware describing language like VHDL or Verilog. It's actually not that much fun ...