I have it running with no problems. Just had to back off on the OC a little as the 480 was getting a tad warm at 104 deg.

Andy

---

wow 104 degrees if i am not mistaken is 1 degree shy of the 105 degree thermal limit on that card!! really make sure to back it off enough to have a 5-10 degree cushion!

that is if you meant 104 degrees CELSIUS!

---

I have it running with no problems. Just had to back off on the OC a little as the 480 was getting a tad warm at 104 deg.

Andy

There are aftermarket coolers that will drop your 480's temps by 30C. I use the Arctic Cooling line myself and they work wonderful. They are very quiet. The only issue is they are big. They are not too hard to install if you follow the directions and take your time while being gentle.

---

I'm seeing a 10-15% speed-up on my 460. 1 wu at a time was taking around 11 mins, now less than 10. Shorties finish in <3 mins, but I try to move them to the cpu, which usually needs more work anyways.

Also, the processing lull when the next wu starts was 16 seconds on my system with x32f, now it is just 8 seconds with x38g.

And the 'penalty' for all these improvements is the card's temp has increased 1 or maybe 2 degrees. :)

Great Job! all around.

Thank You Lunatics team!!

Martin

...one thing I notice that differs from running other apps. The progress indicator isn't updating smoothly, but in jumps of 0.9 % exactly. Maybe that's by design though.

I know that a normal(if there is such an animal) AP, takes about 21-23 hours on my CPU, now it's going to be interesting to see how much faster the GPU can crunch an AP.

Astropulse processing is in 111 sections, so the jumps are actually 0.9009009009... %. The progress code is essentially unchanged from stock so wouldn't be linear with the optimizations in any case, but also Raistmer hates to waste host time doing essentially cosmetic stuff more often than necessary.

                                                                Joe

On my rig, E5400 dual with a GTS 450 running three at a time, On the GPU I'm completing the shorties in about 8 - 10 minutes. the mid range work is taking about 36 - 48 minutes each. This is about the same as what I was doing with the old 0.32 app but I was only running two at a time then. I'm also seeing some new work dated 30 April 2011 taking about an hour give or take a couple of minutes.

For my rig this is fantastic. A very nice speed up. Before this little hickup I reached a RAC of over 9300. That's the highest I've ever been. Should be interesting to see what the jump will be when we get going again.

---

PROUD MEMBER OF Team Starfire World BOINC

Another heads up for you crunching fanatics out there. And I just now remembered this.....

For those of you using the eFMer priority program to run your Cuda work at high priority....
Remember to change the program name in eFmer to the new app name after doing the new Lunatics install....otherwise it is no longer changing the priority.

Meow!

---

"Time is simply the mechanism that keeps everything from happening all at once."

Say I just upgraded to x38g and now It's loading cuda_fermi wu's, I don't have a single fermi in this PC, Is this ok to use with My GTX295 cards?

???
Where do you see anything about fermi in the WU description?

---

"Time is simply the mechanism that keeps everything from happening all at once."

I believe this is because the
<plan_class>
tag in your app_info.xml file has "cuda_fermi" for version 6.10.

-Dave

---

I put x38g_cuda32 and 275.33 WHQLs on my GTX216 last night (my time) and I've noticed a definite speed up :) The only 'issue' I'm having with it is that whenever it's loading the WU into the GPU (that first 15-20seconds or so), if I'm watching a video on youtube or dvd, playing a game or whatever, the video and sometimes entire system hangs until it starts crunching. Doesn't seem to happen if I'm -not- watching a video/playing a game, though it could just be that I don't notice it.

Other than that, I've noticed temps have increased ~2C, nothing major, and GPU load sits at a constant 80%+, mem controller load 40%+ and it never dips below these values. Previously, I'd see spiky dips anywhere between 70-85%.. so that would explain the speed up, if nothing else :)

Thanks for the release!

Say I just upgraded to x38g and now It's loading cuda_fermi wu's, I don't have a single fermi in this PC, Is this ok to use with My GTX295 cards?

All the WUs are the same, so yes it is OK. And the servers don't much care what the plan_class is, as long as it contains "cuda" x38g tasks will be considered "SETI@home Enhanced (anonymous platform, nvidia GPU)".

The reason for the plan classes included in the app_info.xml which the installer sets up is to match the 3 stock classes. Then tasks aren't deleted for those upgrading from stock. The 6.10 cuda_fermi class is listed first, so new tasks are simply assigned to that app_version.

                                                                Joe

Do the new CUDA apps use more constant cpu time now? It seems when monitoring the core usage in Task Manager, the CUDA apps are using up to 3% of the processor the entire crunch time. I thought with the previous version, the cpu portion was mainly starting and ending the tasks, with essentially no cpu use during crunching. Not a real big deal just making sure this is the way it is supposed to be.
Thanks
Dan

---

Is the file "libfftw3f-3-1-1a_upx.dll" no longer needed fr CUDA x38g? It wasn't included in the installer, and it is no longer referenced in the app_info.

It was in X32f though. What is the "libfftw3f-3-1-1a_upx.dll?, and why isn't it needed now?

I'm planning to upgrade the CUDA on my ION PC, that's why another batch of questions :-)

Correct. It is no longer needed with any of the Lunatics applications, and, if you are only using the Lunatics applications packaged in the 0.38 installer, there is no need to refer to this file in app_info.xml

The file is a 'library' of computer routines for "Fast Fourier Transforms", a mathematical technique heavily used in our sort of signal processing. For further information (more than you probably needed!) see http://www.fftw.org/.

The stock (Berkeley) applications all use this external library for their FFT needs. The Lunatics applications have the necessary mathematical tools built in to the application itself, so the external dependency has been removed.

Installed the new version late afternoon yesterday and left my drivers alone. Sure seems like th work crunches noticeably faster. I haveb noticed like a few others that the temps went up 2 degrees C. across CPU and GPU.

Ill run these drivers untill we get rid of all these shortys So i can get a good read on times.

Thanks to all of you at lunatics for a great job.

---

[/quote]

Old James

Wow, this is much better compared to the 0.37 release. On my GT240 (driver 270.61) it is about 15 percent faster. The 0.37 release was the first optimized version I used and it did not seem to improve much compared to the 'stock' app, but 0.38 certainly makes a big difference.

Please keep up the good work !

Tom

if I'm watching a video on youtube or dvd, playing a game or whatever, the video and sometimes entire system hangs until it starts crunching

Best to pause while you're doing other video-intensive tasks to maintain the quality of the science.

If somebody gets a valid test on the 266.58 stable driver VS newer or current versions using the new apps, the kitties are curious whether it is worth risking the upgrade.
And this would be, I guess, a separate question for 200 series cards and Fermi class cards.

Meow.

---

"Time is simply the mechanism that keeps everything from happening all at once."

If somebody gets a valid test on the 266.58 stable driver VS newer or current versions using the new apps, the kitties are curious whether it is worth risking the upgrade.
And this would be, I guess, a separate question for 200 series cards and Fermi class cards.

Meow.

I was using driver 266.58 with the F app and left it the same for the G app on host 5530010. However I am using an 8500GT so I'm not sure if that will help you in any way.
I didn't do any benchmarking like Sutaru did, but the time estimates in BOINC were going down quickly after changing the app.

---

SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[

If somebody gets a valid test on the 266.58 stable driver VS newer or current versions using the new apps, the kitties are curious whether it is worth risking the upgrade.
And this would be, I guess, a separate question for 200 series cards and Fermi class cards.

Meow.

Let's see:
GT 240 (running at x16 v1.1 on an old motherboard)v266.58 Both AR=2.72
Old - http://setiathome.berkeley.edu/result.php?resultid=1940437112
New - http://setiathome.berkeley.edu/result.php?resultid=1951420494

GTS 450 Both v267.59 AR=.369 (running 2 at a time)
Old - http://setiathome.berkeley.edu/result.php?resultid=1927383287
New - http://setiathome.berkeley.edu/result.php?resultid=1951863015

The GT 240 shows 103 seconds faster on a VHAR and I posted earlier that the GTS 450 was approx. 100 seconds faster on VHARs

The GTS 450 shows 516 seconds faster with the new app.

As always YMMV

---

</Tazz>

And as perryjay mentioned you can run 3 instances on 450 as well.

---

With each crime and every kindness we birth our future.