Seti@Home just killed ver 1.06

[Rags]

Rattledagger,

nope, I don't think it's a speed bug. I got good times in win9x, win2k, and not too shabby times in Linux (<4.5 hr)

Rags

[Rattledagger]

Rags, that's your times with the cmd-3-clients, compared to your old 1.06-results?

[Rags]

Roughly a 50% increase. I still get better times using the GUI interface compared to the cmd line versions (by roughly 10-20%). Of course it's cmd line all the way still in linux, but you know what I mean

Rags

[Heiney]

I really don't understand you guys. V3 does over TWICE as many calculations than 2.x and below and it pulls it all off in about the same time. It is MUCH more effective for the SCIENCE of the project.

Remember, the whole point of this project is SCIENCE!



--------------------
ABIT BF6, Pentium III Katmai 450-&gt;663MHz, GW VOS32 Cooler, 256MB SDRAM, Toshiba 6X DVD, Quantum Fireball Plus LM, Matrox Millenium G400 DH @ 160/200, Creative SBLive, Enlight 7237 Case, 300watt PCPC TurboCool PS, and some fun with a Dremel!

[Rags]

Don't even get me started on the "Science" of this project. There have been many articles on this very subject. SETI is purposely puting a performance cap on the programs because they don't want it to run any faster. Intel has made an SSE optimized version, a 3DNow! promotional group has an optimized version, yet SETI won't release it because too many units are being processed as it is, there just aren't enough work units to fill the void. The original version does the same basic thing, the new version just gives more detailed information, but isn't really necessary. They are adding extra calculations to slow things down, and then as a side benefit get better rundowns of the information.

Rags

[Heiney]

Perhaps. I doubt that adding pulse and triplet calculations are that subtle of improvements. The new FFT routine is supposedly ~50-60% more efficient on low-cache CPU's.

My bet is that if a CPU is slower it is just not as efficient at the triplet/pulse code as other CPU's. I doubt the new FFT code slows anyone down.

You do realize that these guys don't have the cash to get the servers they need to handle their current input/output, let alone handle more throughput.

But, yeah, the fact that it is quite a bit slower on some machines is a downside. My P133MMX system used to pull off a work unit in 23-26 hours, now with 3.0 I haven't dropped below 31 hours.

My P3 on the other hand is actually slighly faster overall. Athlons and Durons are bunches faster. P2's are slightly faster. Don't know for sure about K6's but a friend said he was seeing somewhat better times.

And, if they had a client with 3DNow/SSE, what do you think would happen to those
CPU's without that? Either they'd have to make a separate batch of larger work units for those faster clients, or people would have to live with their MIPS/Sun/IBM/PPC/Alpha chips being lots slower than a PC.



--------------------
ABIT BF6, Pentium III Katmai 450-&gt;663MHz, GW VOS32 Cooler, 256MB SDRAM, Toshiba 6X DVD, Quantum Fireball Plus LM, Matrox Millenium G400 DH @ 160/200, Creative SBLive, Enlight 7237 Case, 300watt PCPC TurboCool PS, and some fun with a Dremel!

[Brian R.]

Fast, slow, who cares? Run WUs until boredom sets in... Then do something else

[Assimilator1]

I've found & heard that v3.0 CLi is faster on cpu's with on die cache eg Celerons ,but slower with cpu's with off die cache ,eg PII's ,my PII 280 went from around 10.5hrs/WU to about 12 hrs/WU.
Of course the exception from the above are Athlons ,I would guess that the smaller FFT fits much better into the 128k L1 cache.

nehalmistry

What are your WU times with CLi v2.4?

BTW SETI will be killing off CLi v2.4 on Nov 18 ,the GUI will still run for a short while afterwards though.
Anyway ,like some others said ,it's 'the search' thats important ,not so much the WU time

Rags

>>>>The original version does the same basic thing, the new version just gives more detailed information, but isn't really necessary. They are adding extra calculations to slow things down, and then as a side benefit get better rundowns of the information.<<<<

You old synic!:P (spelling??)
You don't really believe they did it ONLY to slow down the client do you?
Anyway ,it speeds it up on more PC's than it slows it down


------------------
I guess I won't recruit for my team here Team Anandtech