Thanks

I will try when I finished the uts I currently because I do not want to lose them ...

---

Alliance Francophone

hello

With this file app_info, is that if there were more units 6.10, 6.08 and I received the 6.09?

cordially

---

Alliance Francophone

[EDIT] Well heck, then they worked super-fast.

And now something completely off topic and different , numbers of post's aren't updated............

I think I'm going mad

---

The new Lunatics Installer 0.37 - Fermi compatible - is out!

Windows Installer 0.37 Release

Helli

---

A loooong time ago: First Credits after SETI@home Restart

I've just crunched a few, 2 at a time, MB WU's on a GTX470, 18 minutes they took,
0.04CPU + 0.5GPU, looking OK.
Have to wait till they're UPLoaded and look at the results.
Using BOINC 6.10.58, 64BIT. OS is WIN XP64 and using the 0.36(?) installer, only
set GPU to crunch 2 at the time.

---

The new Lunatics Installer 0.37 - Fermi compatible - is out!

Windows Installer 0.37 Release

Helli

Any reason for non Fermi users installing v.37?

IMHO? No. I can't see any Performance Gain on my GTX280/GTX295.

Helli

---

A loooong time ago: First Credits after SETI@home Restart

The new Lunatics Installer 0.37 - Fermi compatible - is out!

Windows Installer 0.37 Release

Helli

Any reason for non Fermi users installing v.37?

Yes, if you're doing CUDA work.

reduced susceptibility to '-12' errors over all prior releases (stock & opt). They still occur, but fairly rarely.

But read the release announcement on the front page carefully, driver version has a new minimum, etc.

                                                                Joe

IMHO? No. I can't see any Performance Gain on my GTX280/GTX295.

Helli

They've added some improvements to help prevent the -12 errors. Some still sneak by but the number has been greatly reduced. I didn't really have the problem but many others have.

You're too fast for me Joe!! :-)

---

PROUD MEMBER OF Team Starfire World BOINC

The new Lunatics Installer 0.37 - Fermi compatible - is out!

Windows Installer 0.37 Release

Helli

Any reason for non Fermi users installing v.37?

While it's not any faster than V12, it's more reliable, producing less -12's,

Claggy

Think we could get a mod to move these last few posts to a new thread? Seems to me this new installer might be important enough to warrant it's own thread. Plus add the warnings and driver version change.


I started a quick thread on it. Maybe others could add a bit more about it.

---

PROUD MEMBER OF Team Starfire World BOINC

Nice comparison, between app.'s, All CUDA, btw.
By the way, running 2 MB tasks, together. Makes no difference in run time.

---

Hi Unified Installer (Optimized SETI) v .37 is out and works great so far. Thank you Jason and team for your hard work.

One of the first results of CUDA FERMI.
First one validated.
Looks like it works OK. Even at lower AR, (0.016).

---

Think we could get a mod to move these last few posts to a new thread? Seems to me this new installer might be important enough to warrant it's own thread. Plus add the warnings and driver version change.

Have to agree, with PerryJay, this thread has quickly gathered so much messages
it affects page loading, even with (A)DSL-Cabel connection (1MByte D'Load, 50KByte UPLoad), but people on other, mostly slower connections, like Dial-In, or a wireless connection, compairable with a cell phone, it takes 'ages'

Two CUDA cards, GTX295 and GTX470 and a result at an AR of 0.39, both using optimized app.'s

MB WU.
The 470 does 2 tasks at a time. (0.04CPU+ 0.5GPU)
Probably the 480 will be as fast or even faster then the 295, at this 0.4AR, though.
Also on low(er) AR, f.i. 0.010AR, the 470 has no trouble in crunching them, whitout error's, that is.
This is sure a big leap forward in GPU processing, Richard, Jason, Raistmer and Joe Segur, in no paticular order, hope I didn't forget anybody,
you all did a marvelous job, thanks from a happy and satisfied cruncher ;-)

---

yes thank you for your work ....

---

Alliance Francophone

(...)
Two CUDA cards, GTX295 and GTX470 and a result at an AR of 0.39, both using optimized app.'s

MB WU.
The 470 does 2 tasks at a time. (0.04CPU+ 0.5GPU)
Probably the 480 will be as fast or even faster then the 295, at this 0.4AR, though.
Also on low(er) AR, f.i. 0.010AR, the 470 has no trouble in crunching them, whitout error's, that is.
(...)

Hmm.. If I look to your 0.39x AR result, your GTX470 need ~ 1,435 secs.
If 2 WUs/GPU, so ~ 1,435 secs. - 2 WUs.

My OCed GTX260-216 (EVGA SSC) with stock MB_6.09_cuda23 make the same AR in ~ 620 secs.
After ~ 1,240 secs. the GPU made also 2 WUs.


How long would need a VLAR WU on your GTX470?

---

Too bad I do not understand much about the English because I would have wanted to understand, because you are talking about two wu both ?!....

---

Alliance Francophone

I try to explain with my poor english.. ;-)

A GTX4xx graphic card could run 2+ WUs simultaneously.
Because the currently available CUDA applications don't use the whole GTX4xx graphic cards (Fermi chip).

You need to edit your app_info.xml file entry (which is in your setiathome.berkeley.edu project folder):

<coproc>
       <type>CUDA</type>
       <count>1</count>
</coproc>

..to..

<coproc>
       <type>CUDA</type>
       <count>0.5</count>
</coproc>

for 2 WUs/GPU

..or to..

<coproc>
       <type>CUDA</type>
       <count>0.33</count>
</coproc>

for 3 WUs/GPU

You could test the best settings, if 2 or 3 WUs simultaneously would be better.

You could look for example with help of GPU-Z, how much your GTX4xx graphic card is used of one CUDA application.
And then you could look how much usage with 2 WUs simultaneously.
If you GTX4xx is maxed out continuously at 100 % with 3 WUs simultaneously, maybe it would be better to let run only 2 WUs simultaneously.
You could compare over the calculation times.
But you must use same AR WU calculation times.


So if Fred have after ~ 1,435 secs. 2 WUs* simultaneously finished.
And I have also 2 WUs* (one behind the other (with stock/original MB_6.09_cuda23_app)) after ~ 1,240 secs. finished, then a GTX4xx or the new CUDA build is not so fast. ;-)

[* AR 0.39x]

---