Xeon Phi basics
I tried searching, but there are amazingly few threads about the Xeon Phi on this forum.
I'm looking to gather some information on these boards. Intel is running a crazy promo right now where you can get the 31S1P for $200 (normally $1700). My guess is Intel poured a crazy amount of R&D into these products and the market just isn't picking them up, so they're trying to toss out the cheaper models for pennies to get people interested in the bigger models.
So, does anybody have any details on them? How do they compare to GPUs? Are proprietary compilers required? Part of the promo is a free 6-month 5-seat license for Intel Parallel Studio XE, but how necessary is this? Would gcc work? What about other proprietary compilers? For example, Portland Group has their standard "Workstation" license, and a separate "Accelerator" license which includes hooks for CUDA. Would either of these be able to build code for the Xeon Phi or is it in a whole other league?
I'm highly tempted to push the boss to purchase one of these 31S1P promo boards just to drop in one of our existing servers and play around with the possibilities, but I'd like to know more about the requirements to actually utilize it (I know optimizing a program for this could require significant code modifications to increase vectorization, etc. I'm not talking about code optimizations here, just logistical requirements, and various compilers' abilities to take advantage of the Phi's vector processing and parallelization, provided the code is written as necessary).
Any input, especially from those who have some Phi/GPU experience would be appreciated. This is a very different ballgame than my typical big powerful brute-force Xeon systems, so I'm a little out of my depth.
Last edited by suicidaleggroll; 12-04-2014 at 10:57 PM.
|