How about a Einstein client optimized for ATI GPU's like Folding has? I not only have a X1950Pro (RV570XT GPU) to play with but I just got a shiny new X2900XT as well. I would love to turn the R600's 700 million transistors loose on this project!
Current GPUs and the PS3* are only capable of 32 bit (single precision) floating point operation. Einstien is heavily dependent on the extra precision of 64bit (double precision) floating point operations. A port is unlikely anytime soon.
* The PS3 has a single G5 type general purpose CPU with double floating point suppport, but all the extra cores that produce t he huge number of flops on the specsheet are single precision only.
Current GPUs and the PS3* are only capable of 32 bit (single precision) floating point operation. Einstien is heavily dependent on the extra precision of 64bit (double precision) floating point operations. A port is unlikely anytime soon.
* The PS3 has a single G5 type general purpose CPU with double floating point suppport, but all the extra cores that produce t he huge number of flops on the specsheet are single precision only.
The SPEs (aka the "extra cores") are capable of IEEE 754-compliant double precision mathematics. However, their double precsion is nowhere near as fast as the single precision floating point capabilities (which do not conform to IEEE 754 because they are optimized for game and multimedia performance where accuracy is not so important, so are useless scientifically anyways). Nevertheless, their double precsion capabilities are great for scientific computation even though they are optimized more for games and multimedia according to one research paper I have read and have unfortunately forgotten the name and authors of.
The PS3 cell doesn't need to IEEE754 compatible at all! How did old applications work long-long ago without any compatability or even with errors in FPU (Pentium FDIV bug)? We need only an IEEE754 compatible compilator and that's all. When I had need to write a proggy with accurate computations that days, I turned off FPU compiler in Borland Pascal and used a software math library instead. And that was working fine. So, we need now only to try PS3 with IEEE754 compatible compiler.
Implemementing floating point using integer math is SLOW. Really, really, really S L O W! You'd still only have one core doing work, and it'd be far slower than using its double precision FPU instead.
The XBOX 360 would also be a nice platform, it has a triple core PowerPC based CPU with an impressive floating point throughput. And the XBox is cheaper than the PS3 . There have been reports that Rosetta@Home is in talks with Microsoft about a Rosetta@Home version for the XBox. Now this could be interesting. Unlike Folding@Home which runs on the PS3, Rosetta is a BOINC project. So could there be a BOINC version for the XBOX 360 ??? Could other projects bring their projects to the XBOX as well?
One problem, however, is that earlier editions of XBox 360 must have had a rather poor cooling, barely sufficient for average gaming. Running BOINC projects on those early boxes might send even more XBoxes to console heaven.
How about a Einstein client
)
How about a Einstein client optimized for ATI GPU's like Folding has? I not only have a X1950Pro (RV570XT GPU) to play with but I just got a shiny new X2900XT as well. I would love to turn the R600's 700 million transistors loose on this project!
Horribleron
Current GPUs and the PS3* are
)
Current GPUs and the PS3* are only capable of 32 bit (single precision) floating point operation. Einstien is heavily dependent on the extra precision of 64bit (double precision) floating point operations. A port is unlikely anytime soon.
* The PS3 has a single G5 type general purpose CPU with double floating point suppport, but all the extra cores that produce t he huge number of flops on the specsheet are single precision only.
RE: Current GPUs and the
)
The SPEs (aka the "extra cores") are capable of IEEE 754-compliant double precision mathematics. However, their double precsion is nowhere near as fast as the single precision floating point capabilities (which do not conform to IEEE 754 because they are optimized for game and multimedia performance where accuracy is not so important, so are useless scientifically anyways). Nevertheless, their double precsion capabilities are great for scientific computation even though they are optimized more for games and multimedia according to one research paper I have read and have unfortunately forgotten the name and authors of.
The PS3 cell doesn't need to
)
The PS3 cell doesn't need to IEEE754 compatible at all! How did old applications work long-long ago without any compatability or even with errors in FPU (Pentium FDIV bug)? We need only an IEEE754 compatible compilator and that's all. When I had need to write a proggy with accurate computations that days, I turned off FPU compiler in Borland Pascal and used a software math library instead. And that was working fine. So, we need now only to try PS3 with IEEE754 compatible compiler.
Implemementing floating point
)
Implemementing floating point using integer math is SLOW. Really, really, really S L O W! You'd still only have one core doing work, and it'd be far slower than using its double precision FPU instead.
The XBOX 360 would also be a
)
The XBOX 360 would also be a nice platform, it has a triple core PowerPC based CPU with an impressive floating point throughput. And the XBox is cheaper than the PS3 . There have been reports that Rosetta@Home is in talks with Microsoft about a Rosetta@Home version for the XBox. Now this could be interesting. Unlike Folding@Home which runs on the PS3, Rosetta is a BOINC project. So could there be a BOINC version for the XBOX 360 ??? Could other projects bring their projects to the XBOX as well?
One problem, however, is that earlier editions of XBox 360 must have had a rather poor cooling, barely sufficient for average gaming. Running BOINC projects on those early boxes might send even more XBoxes to console heaven.
CU
BRM