<r>It's been mentioned before but Simon Peyton-Jones worked on some QF haskell stuff.Here's a recent talk:<URL url="http://ulf.wiger.net/weblog/2008/02/29/simon-peyton-jones-composing-contracts-an-adventure-in-financial-engineering/My"><LINK_TEXT text="http://ulf.wiger.net/weblog/2008/02/29/ ... nee...

A little off topic, but have you guys tried the Intel Thread Building Blocks library..The for loop in your code should parallelize nicely across cores..nd

<t>>>I'll get the fftw code compiled and running and then see what I can get out of the NVidia processors.These are two mutually exclusive acts right??-- get fftw running on something (AMD,Intel)-- run cudafft on the GPU. 'Cus there ain't no fftw version of cuda. So, I see about 37 GFlops/s on a 1-d...

<t>A CM machine is the Connection Machine supercomputer from Thinking Machines (circa early 90's). The CM's bare a striking resemblence to the modern GPU (massively parallel, very very SIMD -- single instruction multiple data). >>>like wicked. What are the typical speed up factors in your experience...

<r>I don't know if anyone else has seen this, but Jack Dongarra atUniv. of Tenn has an interesting paper about use 32 bit floating pointarithmetic for 64 bit accuracy. I found this paperin the context of the IBM cell processor,where calculationsusing float are much much faster than calcs using doubl...

<t>Oddly enough the g80 (and the cuda api) don't necessarily map thatwell (as well as previous GPUs anyway) to visualization applications. Reallyit's a compute engine . We've had a coupla of these things in our lab for 4 or 5 months and they are wick'd fast (for the right app.. over 150 Gflopssustai...

<r>Why are they using a database? In the MD codes I've seen (ok MD as applied tomaterial science) most of the output data is written to a flat file (on a per node basis).I know for this code <URL url="http://sc05.supercomputing.org/schedule/pdf/pap122.pdfthat%27s"><LINK_TEXT text="http://sc05.superc...

<r>Just a word of warning. I've used vtk and paraview extensively and they really aren't for the faint of heart. Most of Paraview design (well most of the parallel rendering) was done by a guy on myteam and even he has problems using it sometimes <E>:-(</E>). I've spent a better part of a year addin...

<t>Fabulous, I'll tell the radiation transport and Sn transport guys at workthat they can stop working. Perhaps you could turn your mighty clusterto the problem of turbulence. ok ok, start with something simple likethe knapsack problem, then move on to turbulence.>>Here's my take:>>Anyone who uses a...

<t>What's really fascinating about all this multicore stuff isit completely changes the way you think about problems.Imagine a world where the compute you do is free and the getting the data to the processor is what cost.How would change what/how are you doing things.As a point of reference we have ...

At the end of the day parallel/distributed programming (for non-trivial apps) is black magic.

<t>I dont't see what the value add is for the boost.mpi. All they are doing is wrappingsmpi calls. Complexity for the sake of complexity doesn't make a lotof sense. Also how much additional latency are you adding to the systemby doing all this abstraction. As an example, openmpi runningover a bi-dir...

<t>w>>oul you mind telling us about the hardware and setup you used?>>Cell seems to be a nice project to play around with.>>Cheers>>MichaelMichael,We have an 2 blade cell system (1 cell per blade) that lives in aIBM blade center. Each blade runs a copy of Fedore Core 4 witha few patches (which I thi...

<t>I thought I'd share any interesting result. Just for grins I ported a reallysimple linear congruence RNG(when I say simple I mean simple)to the IBM Cell. And ran a coupla of test... The results are (in humbleopinion) pretty interesting.To generate a 1e6 random numbers on a 64 bit Opteron takes .1...

<t>>>... coprocessors such as Clearspeed CSX600 could significantly speed up. The csx600 is about>>8k sterling and gives you 50GFlops peak if you can program it. Or IBM's cell blades (but again its rather tricky toThe Clearspeeds don't have any ECC; so you'll have to write your own which will slow d...