Linux S5R2 App 4.35 available for Beta test
Bernd Machenschalk
Moderator
Administrator
Joined: 15 Oct 04
Posts: 4349
Credit: 253660616
RAC: 35156
27 Jul 2007 15:56:58 UTC
Topic 193003
(moderation:
A new Linux App is available from our Beta Test page.
This App is part of a new generation of Apps that follow our attempts to track down and fix the remaining problems of the new software we're using since S5R2. It has the same codebase as the 4.33 Windows App, thus basically the same features.
For the "SIN/COS LUT" it uses the original modf() variant which showed to be fastest (no surprise - the original code was tuned on Linux).
BM
BM
Linux S5R2 App 4.35 available for Beta test
)
You have forgotten to update the beta page :-)
Direct link
http://einstein.phys.uwm.edu/app_test/linux/einstein_S5R2_4.35_i686-pc-linux-gnu.tar.gz
I have crunched 33% of a big
)
I have crunched 33% of a big WU using App. 4.31. Can I switch to 4.35 or it's better for me to wait until it's finished?
Tullio
RE: I have crunched 33% of
)
Hi Tullio!
You can switch at any time, no need to wait until the WU is finished.
CU
BRM
RE: You have forgotten to
)
Fixed, thanks. The md5sum was correct, though.
BM
BM
RE: Hi Tullio! You can
)
Thanks Bikeman. 4.35 certainly looks faster than 4.31, which was painfully slow.
Tullio
RE: RE: Hi Tullio! You
)
Absolutely. The speed difference will vary depending on the CPU type, but between 15% and 25 % seems to be usual speedup.
It' so amazing what a single line of code can do, it shows how "hot" the hot-loop in E@H is (and how tedious it is to get the code optimized across different combinations of CPUs, OS & compilers).
CU
BRM
My host (sorry I'm lazy and
)
My host (sorry I'm lazy and don't feel like linking many results) has completed 8 WU with 4.35, 4 of them fully with the latest beta, and the other 4 were split between the 4.35 and 4.31 which is reflected in crunch time.
4.35 is much faster than the other betas, but is still slower than the official app.
Data pack frequency 464 :
Approximate times
Official app - 42,600s
4.31 beta - 50,200s
4.35 beta - 45,800s
There are 10^11 stars in the galaxy. That used to be a huge number. But it's only a hundred billion. It's less than the national deficit! We used to call them astronomical numbers. Now we should call them economical numbers. - Richard Feynman
I can confirm what Dave is
)
I can confirm what Dave is reporting. This is the results list for an overclocked Athlon XP 2400+ running around 3200+ speed. It was running 500.70 frequency data in 90.4K secs with the 4.21 official app. This slowed to 104.4K secs under 4.31 and the first result, largely but not fully under 4.35 is now in. The time is back down to 94.1K and should drop a little more with the next result which will be fully 4.35 crunched. Unfortunately it will still be a little slower than 90.4K secs.
For me, the biggest plus in all of this (for Linux) is the almost complete lack of validation problems. That's really encouraging, thanks.
Cheers,
Gary.
My own results do not show
)
My own results do not show any noticeable difference in speed between 4.35 and 4.21 apps, they are very close. 4.35 seems to be even slightly faster, with a completion time of 11856 seconds, against 12900 seconds for 4.21
I m testing 4.35 on a single machine (AMD Duron 1600 Applebred) and with only one WU already completely computed with this app (correctly validated against a Windows box).
This WU has the same data packet frequency (473) than the ones I completed three weeks ago with 4.21 (already deleted by the server). The granted credit is lower now, but in a very small amount: 394.67 against a previous value of 394.69
If the size of WU's has really changed a bit, then perhaps I could say that 4.35 is not really faster than 4.21. But I can say also that it is not slower. At least on my machine.
But I suppose that a small difference in speed could be possible, because 4.35 has much more debug code.
I have cunched two WUs with
)
I have cunched two WUs with the 4.35 application on an Intel Core 2 @ 2.13 GHz. Computing time here is about 7 to 15% more than with the official 4.21.