4 ms·
Very nice read! How does this work compare to work on java 8 to bring GPU compute to the jvm by AMD, ARM and Oracle? Do you see this as a competition or do you
by jerven 13y ago
Very nice read! How does this work compare to work on java 8 to bring GPU compute to the jvm by AMD, ARM and Oracle? Do you see this as a competition or do you feel you are in a separate niche?
- Hydraulix989 13y agoYes, we are in our own separate niche. Rather than optimizing the computation of individual loops and the like, we're attacking the problem from a "Big Data" perspective where we seek to optimize processing large data sets. We're talking about special filesystems a la FB's Haystack to optimize disk sector layout coupled with RAID schemes to bring data to the GPU as efficiently as possible to overcome I/O bottlenecks. We are initially interop'ing with Hadoop, and we are going to extend this to support Pig and Hive queries (along with database support). I really do think this is an "apples and oranges" comparison, and it's great that they are taking the C++ AMP route of parallelizing loops (perhaps with some parallel primitives mixed in), but at the same time, there's a lot more to the problem we are trying to solve than just parallel compute.