The server that was running the PIES client/server system, the GFN/PIES/LG range reservations (and my ECMNET server, Paul Leyland's ECMNET server, and goodness...
... Also the records of what has been reserved? ... elsewhere as I ... sieving will ... I presume you remember that 24576_k3.abc and _k5 can be had from my web...
... I have a backup from a month or so ago. Google has more recent reservations cached though, so reconstructing shouldn't be too hard. ... Not as far as I...
I've been playing with djbfft and overseis and believe I may have extended it by a bit... up to 65536 limbs or Phi(196608,k*b^2) or close to 800K digits! I'm...
... I already tried, and the slowdown at 16384 limbs was awful. There was already a slowdown at 8192 limbs, which no-one apart from me seemed to notice. ...
On my machines at any rate, k5 seems to be feasible up to 800000. Since I'm running out of what I've already reserved and I believe no one has any reservations...
... It's the speed ... What were ... I gave up another hour of sleep and got the code to compile on the xeons. Here's the timing for 8192 and 16384: ...
... This is a little surprising. Does it use SSE2 for FPU operations? The Sparc code is optimised for 16 FPU registers, and I didn't think that the 64 bit...
... the ... I'm running them in EM64T mode (x86_64 linux) so they DO have 16 FPU registers. However I'm only using -march=c3 -mcpu=nocona as when I first...
OK. Fortunately the xeons are fast machines, so I can compile the djbfft code multiple times. I ran the speed test with many different options. First I...
http://www.cs.berkeley.edu/~jonah/pies/superseis-0.0.tgz gzip compressed tar file. known to compile on a couple of linux 2.6 boxes. this is pre-alpha, NOT...
... Whivh version of whose DJBFFT were you running, and which overseis was this? Anything > 8192 limbs won't work with my version of DJBFFT for any size over ...
... what seems off? are they slower? faster? a bit? a lot? did you use sparc or ppro or auto for djbft/conf-opt? ... (679541bits) Bus error ... Oops. A...
... I do 10 times as many runs to get a more accurate figure. That should be obvious just from diffing speed.c. ... This is just a modification to overseis....
It seems that supercherries already bust the cache on the VIA C3 processor, because superpeaches are not that much slower. Here are some supercherry timings: ...
... Maybe I'm comparing apples to oranges, but these are way off. The first is from Phil's. The second is Jeff's. I would expect these to be relatively...
... Just diff speed.c. ... There's not really much merging that needs to take place, as I made the exact same changes about 3 years ago (I've not looked at...
... Keep ... the exact ... I'm assuming ... functions), and ... caches have ... sense now. I got it to scale, yes, but how "well" is an open question. As Phil...
It seems I've unwittingly usurped part of someone else's range. I think it might have been Bob. Phil sent me a follow-up on this a day or so ago, but I seem...
I am currently processing 24576_k5_680000-740000, and am about 50% through it. You say that you've searched up to 712561? I'll not report any primes in that...
Ahs anyone else taken reservations for k3_ supercherries? I'm in need of some more allocation and I don't want to step on ayone as I did wth k5. Phil?...
... "Sector Read Error" :-( As far as I remember, no-one had taken a k3 reservation. Google's cache doesn't work, alas. The tail end of k5 is free though,...
Having problems with message search?
Fill out this form to ensure your group is one of the first to be migrated to the new message search system.
