Hi,
Do you know anything about nvamg? My understanding is that it is an Nvidia project that has not been advertised much and was used in some special apps like Ansys Fluent.
I was at a talk by an NVIDIA guy in May. It's an aggregation-based algebraic multigrid scheme, using similar components like what is implemented in CUSP. It will be a commercial product from NVIDIA (maybe free or cheap academic licenses) using MPI on a small number of nodes, primarily targeting widespread engineering software. The team consists of about 5-10 members and is headed by Robert Strzodka if I remember correctly, so they need to generate enough revenue to 'feed' them :-)
Thanks for sharing your experiences. I found the paper below earlier this evening when I revisited the ViennaCL web site and looked at the benchmark section.
Ah, I see. :-) Even though I spend some time on improving MIC performance since these benchmarks were run, the overall trend remained unchanged. I'm still hoping for a better OpenCL SDK for MIC from Intel... Best regards, Karli