[rant] The PS3 should be able to do this project, but the porting will be hell as its' going to run the cell processor (which if I read correctly and it was true, will require a lot of the software to do scheduling and what not), which might be nice but only 1 processor could possibly run this project, because of the other not-so-whole processors (Einstein could probably only run on the central POWER PC arch cpu (others only do single floats and I think this project does doubles, correct me if I am wrong)), but still a POWER PC should yeild some decent results. The albert ap could possibly run on a PS3 (recent stats say that the PS3 will have 256 MB ram, which will kick some butt). Now if it where possible to the entire cell that would be insane .. 9 einstein WU running (assuming it was possible to use the enable the extra SPE (synergy is such a stupid word but oh well)). However, if Akosf can make a 'dual' core (software wise), then using his techniques, it might be possible to make the other SPE's do some work.
As far as the PS2 ... the emotion engine will probably be a problem lol [/rant]
Need to petition Microsoft to support BOINC on the Live network, like a game. Then "normal" people could let their machines run with out having to hack their machines. They can take credit for the calculations as far as I am concerned.
Are you seriously saying that we would be able to use the ps3 in 32-bit mode? Also can you make the optimized app PPC based, so it will run on the Cell? Which means we would get 218GFlops/s+ per machine instead of those little 25-30GFlops in 64-bit.
Assuming the 50% double precision's a hard number, a naive port would have a theoretical level of ~50 gflops is possible (25 + 25). A tailored port could potentially have faster run times by replacing a step done in double with a lengthier single version that could none the less run in the same time.
If the doubles are only needed to maintain a given level of precision and not to keep the calculation from becoming unstable*, the validator could be loosened somewhat to allow a greater use of single ops.
This sounds interesting, Making a PS3 into a cruncher for Einstein, I assume that this Would run under some sort of Boinc program? In any case this interesting, I might get one eventually, Besides My PS2 is an older model and could use a replacement.
[rant] The PS3 should be able
)
[rant] The PS3 should be able to do this project, but the porting will be hell as its' going to run the cell processor (which if I read correctly and it was true, will require a lot of the software to do scheduling and what not), which might be nice but only 1 processor could possibly run this project, because of the other not-so-whole processors (Einstein could probably only run on the central POWER PC arch cpu (others only do single floats and I think this project does doubles, correct me if I am wrong)), but still a POWER PC should yeild some decent results. The albert ap could possibly run on a PS3 (recent stats say that the PS3 will have 256 MB ram, which will kick some butt). Now if it where possible to the entire cell that would be insane .. 9 einstein WU running (assuming it was possible to use the enable the extra SPE (synergy is such a stupid word but oh well)). However, if Akosf can make a 'dual' core (software wise), then using his techniques, it might be possible to make the other SPE's do some work.
As far as the PS2 ... the emotion engine will probably be a problem lol [/rant]
"But it's turtles all the way down!"
Need to petition Microsoft to
)
Need to petition Microsoft to support BOINC on the Live network, like a game. Then "normal" people could let their machines run with out having to hack their machines. They can take credit for the calculations as far as I am concerned.
single precision's good
)
single precision's good enough for einstien, akos's apps make heavy use of sse or 3dnow, both of which are single precision ops.
RE: single precision's good
)
about the half of the calculations needs double precision ( but it is subtitutable )
Are you seriously saying that
)
Are you seriously saying that we would be able to use the ps3 in 32-bit mode? Also can you make the optimized app PPC based, so it will run on the Cell? Which means we would get 218GFlops/s+ per machine instead of those little 25-30GFlops in 64-bit.
Assuming the 50% double
)
Assuming the 50% double precision's a hard number, a naive port would have a theoretical level of ~50 gflops is possible (25 + 25). A tailored port could potentially have faster run times by replacing a step done in double with a lengthier single version that could none the less run in the same time.
If the doubles are only needed to maintain a given level of precision and not to keep the calculation from becoming unstable*, the validator could be loosened somewhat to allow a greater use of single ops.
* see http://einsteinathome.org/goto/comment/27992
This sounds interesting,
)
This sounds interesting, Making a PS3 into a cruncher for Einstein, I assume that this Would run under some sort of Boinc program? In any case this interesting, I might get one eventually, Besides My PS2 is an older model and could use a replacement.
http://contract.rebuildthedream.com/
http://tinyurl.com/Pass-HJ47-NO-Time-Limit-On-ERA