Thursday, January 22, 2009
Back at work
I am back at work in NJ. I am working primarily on two things. The first is the wrap up of my polyphase filter bank code. This will wind up in GnuRadio, DttSP, HPSDR, and more. Next, with Tom Rondeau, we are trying to get a killer app together for the Cell processor in general, but the PS3 in particular. If we do not succeed, I plan on giving up on the Cell altogether (just as IBM has done in my opinion).
The problem with the Cell is two-fold. First, it is too difficult to program and it is not commodity priced. The former has led to the latter. Graphics cards and GPU's are here to stay. They will be here long after everyone forgets the name Cell BE and they are going to be fast. It is a shame that IBM had so little real interest in pushing this forward and each step of the way I consider major mistakes to have been made that prevented it from catching on. My guess is that the Cell BE 2 will be at best a small make over of the Cell BE and at worst, tossed over the side and handed off to Sony to utterly ruin.
I think the best option overall, for serious high performance software radio computing is the new AMD Phenom, Intel I7, and Nvidia GPU cards sitting in PCIe-16x slots. These will be programmable, with the fast hardware usable by compiling in function calls and/or intrinsics. This will not be optimal, but it will be immediate. It takes over a month just to get linux installed, Cell SDK installed, learn enough to do anything, and then you have to deal with DMA, memory that is too damn small, and very expensive pricing for anything but broken versions (6 SPE versions of Cell in PS3) and their horrible GigE support because we can't have DRM/Video/BlueRay thingy's stolen now can we? BlueRay will also fail to be a big money maker like DVD's. Streaming media is going to nail it to the floor.
Enough prognostication. Back to work!