Hi, i am planning to dedicate hardware for doing just optimization. Do you have any suggestions regarding the most suitable specs. Suitability means cheap and efficient thanks
On what platform you want to optimize ? Is the software able to use multi-processor ? Are you able to outsource your opti's to GPU ? I'm running optimizations in two different software solutions on two different hardware configurations. 1st .. on Matlab via GPU => CUDA on an old Intel E6700 processor system and 2nd .. via Multicharts on a Dual Xeon X5670 system. This performs a lot. Regards. Mike
Is optimization a bottleneck in your process? If it is, then the higher cost of "faster" hardware will also be offset by an increase in your "efficiency" to some extent (i.e. you will be able to crunch through more optimizations in a shorter time, and get more good systems into the market quicker). Another thought ... perhaps several cheaper machines (each optimizing a different system, or the same system on a different set of instruments) is better than a single more powerful machine? I guess it is down to your own "management bandwidth" in being able to juggle the processing of the results coming from each machine with any other tasks you need to do ...
With reference to the above, I'm curious as to whether you have made any decision, or moved forward (as I have to make a similar decision soon, too). Have you? What have you decided? Any thoughts, or feedback? Thanks.
I recently bought a machine specifically intended for huge parameter-sweep backtesting with Trading Blox. A typical "job" runs for 2 days and tests hundreds of thousands of parameter combinations. I ended up having my local computer shop assemble these components into a system: CPU = intel core i7-930 mobo = ASUS P6T for X58 i7 RAM = DDR3, 12 Gigabytes (3 channels x 4GB per channel) HD = 2 Western Digital 1TB @ 3GB/s, configured in RAID-1 OS = Win7 Ultimate edition, 64b PS = Corsair 750W plus the usual other stuff: chassis, network, video, etc. Trading Blox is a Windows application so I took the easy route: install Win7 as the native OS. It's pretty much a dedicated Blox machine, why fool with VM or dual boot or other complexity. Choosing a CPU + mobo that supports 3 channels of RAM really helped my application because Blox generates a lot of memory traffic. I generally run two copies at a time (working in different zones of the solution space), because 2 parallel copies seems to be sweet spot for highest throughput, i.e. max # answers per hour of wall clock time. I guess Windows finds a way to squander the other two cores The shop gave me all the packaging for the various components; I found it hilarious that the ASUS mobo box says "Best Quality Power Components, 100% Japan-made." I guess they're trying to reassure customers that this mobo won't have Capacitor Plague.
Personally I feel the best and most suitable hardware is matched specifically to the application it needs to run. I had a custom application written that checks many hundreds of thousands of chart decisions on each unique run. It currently runs in MultiCharts using eSignal data but I'm going to migrate it over to my own custom proprietary financial engineering platform so I can better utilize a 64bit, multi core and mega-ram environment. I found a refurbed HP ProLiant 3.0Mhz Quad Xeon server (8 cores) and beefed it with to 2-256GB Solid State HD's and 32GB of DDR2 1333mhz RAM. I'm running Win 7 Ultimate 64bit and added a dual video card all for under $2,500.
Could you plz give some more info regarding your GPU/CUDA system of optimization. It is really interesting.. If you have any documentation, this would be nice. Thankx
I use a MAC G5, dead quite, amazing speed, 8 cores....running windows 7 on dedicate flash drive, fastest pc i ever had..or dare i say mac!