2x Xeon x5450 QuadCore Workstation for factorization?

[Dubslow]

Originally Posted by fivemack

That's not true in theory, which is why it's not true in practice. It's not a matter of alternating issuing a line of micro-ops from each thread; what happens is that each thread pushes micro-ops into a separate queue, and the issue unit can pull micro-ops out of either queue to fill the execution units.

If the code were sufficiently superbly optimised that every execution unit could be filled at every cycle by either thread, then this would come out the same as alternation; but there's very little code (though prime95 is probably an example that comes close) which is that optimised. Generally there are opportunities for things from thread 2 to fill holes in thread 1 and vice versa.

Ah, okay. Prime95 forms the core of my experiences with high-speed computing :-)

So despite my monitor showing full CPU load, that doesn't necessarily mean that each cycle is being used to execute the program? (Currently running GMP-ECM via YAFU, but it appears the same (fully loaded) whether it's Msieve, GMP-ECM, YAFU/SIQS, or lasieve/ that's running.)

[fivemack]

It's not so much a question of every cycle being used, as of every sub-slot within every cycle being used, and I don't think performance monitors generally monitor at that level.

For ecm on the i7/960:

1 copy does a 1e7 curve on (102^128+1)/7443446978046721 in 85.9s
6 copies take average 89.4s
12 copies take average 169.7s

so ecm gains only about 5% from HT - still not to be sniffed at

When I wrote a program whose purpose in life was entirely to stall on memory access, I get a 60% gain from HT - six copies run in 16 seconds each, twelve copies run in 20 seconds each

[R.D. Silverman]

Originally Posted by fivemack

No, for GNFS sieving you want to run eight threads on either a dual quad-Core2-Xeon or a single quad-{Sandy,Ivy}Bridge with hyperthreading.

Running 8 GNFS threads requires a LOT of memory........

[fivemack]

Originally Posted by R.D. Silverman

Running 8 GNFS threads requires a LOT of memory........

Only gnfs-lasieve4I16e uses more than 1GB per thread, and even 16GB is a reasonable configuration for a desktop bought nowadays - order of $100.

[LaurV]

Originally Posted by fivemack

Only gnfs-lasieve4I16e uses more than 1GB per thread, and even 16GB is a reasonable configuration for a desktop bought nowadays - order of $100.

sieving with 16 threads with 4I13e, for a C127 of aliquot 4290.i2321 in this right moment, it takes 29MB per thread (I may be again parallel with the subject...)