Bah! You had to ruin my dreams with your fancy trans-warp engine didn't you!!! :bow:

I'm only at 13.3M...

Jeff.

<off-topic>
*secretly informs the [url=http://stardestroyer.net/Empire]Empire[/url] of the fleet's location* :devil:
</off-topic>

[QUOTE=ixfd64]<off-topic>
*secretly informs the [url=http://stardestroyer.net/Empire]Empire[/url] of the fleet's location* :devil:
</off-topic>[/QUOTE]

Oh good grief, PLEASE don't start a Star Trek vs. Star Wars who-would-win-in-a-fight off-topic subthread. There was a thread like that over at the TrekBBS; it degenerated pretty quickly into people yelling "You suck!" at each other and flaming people who didn't agree with their opinion that the little bumps on top of the star destroyers were shield generators (the other side of the debate contended they were sensor dishes).

I've run into traffic on the multi-CPU system I've been using, and my latest build (I stupidly overwrote the old one) no longer works on an alternate dual-CPU system I had in reserve, which has a slightly older version of the OS and no compiler (i.e. I have to build a static binary on the system with the good compiler and hope it works on the others). So my conservative estimate is sometime next week for my run to finish. Still faster than my original single-threaded estimate, but annoying nonetheless.

While playing around to see if I could get back to a working multi-threaded build on the dual-CPU Itanium system, I ran into something odd (both figuratively and literally, as I shall explain). While trying various settings on the 2-CPU machine I couldn't get 2-threaded mode to run, no matter what I tried. At some point I cut and pasted a test-run input set I'd used on a 4-CPU machine, and suddenly it worked! I looked carefully and noticed that I'd specified 3 threads, and for some reason the 2-CPU system handles that just fine.

Even weirder: previously (with the first version of the working multithreaded binary I built a week ago, which is long gone) I'd gotten ~.07 seconds per iteration in 2-threaded mode, which is not as efficient as I'd like (only ~30% faster than the .095 sec/iter I get in single-CPU mode on that system) because only part of the FFT-based multiply was parallelized. Still with only that partial parallelism but running in 3-threaded mode (and I verified that the 3 threads were indeed dividing their time among the 2 CPUs), the timing drops to .05 sec/iter, i.e. roughly what I expected to be able to achieve from optimal, full multithreading of all parts of the algorithm.

Strange stuff, this.

Long story short - I have 2 separate systems on which I get similar timings, so at the rate of .05 seconds per iteration I expect to finish in roughly a week's time.

Tony, please leave just a teensy bit of the Borg cube for me to shoot at when I get there... ;)

[QUOTE=ewmayer]please leave just a teensy bit of the Borg cube for me to shoot at when I get there... ;)[/QUOTE]

Don't worry, we'll have it carved up into nice little car-sized chunks perfect for target practice. :wink:

Sorry for not posting this before. :redface: I've been catching up on my e-mail and noticed this in one the the CNET newsletters I get:

[url]http://news.com.com/2100-7343-5215542.html?part=dht&tag=ntop[/url]

Whoa! I don't visit the forum for 11 days and look what happens! I guess i should stop by more often :innocent: .

hey, maybe i could visit less often to stimulate the amount of mersenne reports :rolleyes: .

21450000 ( 89.24%) (0.018 sec/iter)
 

[QUOTE=Jeff Gilchrist]Bah! You had to ruin my dreams with your fancy trans-warp engine didn't you!!! :bow:

I'm only at 13.3M...

Jeff.[/QUOTE]

Sorry Jeff. It should be done tomorrow.
Guillermo and I need to make still some more improvements to Glucas in order to reach the maximum on a NovaScale 16x ia64 machine :
0.009 sec/iter some day.
Tony

okay people ... you can't hide the exponent forever :smile:

Whooops! Cat is out of the bag.