Last author is involved with PhysX (partner?)
Comments came mostly in the form of questions that were not addressed in the article. Everyone wanted to see more information. Such as:
- performance comparisons between Rigel and NVidia G80 and possibly Larrabee
- coding comparisons
- how is it easier to code?
- code examples please!
- distinction between API and LPI, does API code achieve good performance? how easy is LPI to code?
- how much work was req'd to adapt benchmarks?
- the architectural ideas aren't new, but if they get it fab'ed, it will be something new?
- the processor has high bandwidth, how well is it used?
- what are the details of this 'multi-stage crossbar'?
- what are the scalability issues?
- will the old code run if the number of cores/cluster is increased?
- how sensitive is performance to in-core-cache out-of-core-cache issues? does the programmer have to effectively treat the cluster cache as a scratch pad (being careful to adjust his algorithms to stay in and not spill) while not having the control a scratch pad offers?
- how large is the MRI image size?
Peter – please email the author and ask for code samples! TY!