OK, I was wrong.

http://boinc2.ssl.berkeley.edu/beta/download/setiathome_6.09_windows_intelx86__cuda_fermi.exe

(double underscore) has just done the business for me, and that was starting from the directory Jason posted - I was able to see the file listing this time (so '403 forbidden' has been lifted since I posted, either by admin intervention or simply post-outage).

File size is 2,476 KB - make sure you get the whole thing (browser downloads have a habit of dropping out early, leaving a short file). Because I've only done the download manually, I haven't (yet) worked out if the fftw lib is needed, or any app_version settings that might be needed.

Edit - the .exe has the following dependencies. You should be able to get them from the same directory (untested).

LIBFFTW3F-3-1-1A_UPX.DLL
CUDART32_30_14.DLL
CUFFT32_30_14.DLL
NVCUDA.DLL

That last one is a new one on me.

Edit2 - None of the three CUDA DLLs named is present in the Beta download directory. I still think we need someone to attach and report back on what they are assigned by the server.

Salute!

Where I can get V13 at least? As far as I see, it works with Fermi, rite?

And bcs I too cant access the above folder. it would be nice if someone who has the stuff needed, pack it all at put it for DL, if possible. Tnx.

I hate something being idle.

---

This is curious. If I click on my own link (which worked yesterday), I get "The service is unavailable".

But if I click on the copy of the exact same link which I posted in the Beta forum a few minutes later, the app downloads OK.

Basically, I think that's saying: "This is still a Beta application. Nodbody knows whether it works or not. Don't be greedy and jump the gun, trying to run it on the main project before it's passed quality assurance."

Wait for the testers to catch up. Better yet, become testers yourselves, and put something back into the project that way. Sorry, I'm not flush enough with money to spend £310 GBP to do the testing for you.

I didn't have an issue downloading the file myself at your link later in the day. I didn't get around to installing it til this morning but it is working.

I had to reschedule tasks to the GPU's so that it didn't get clogged by WU's that don't perform well under the GPU's. And I also had to change the amount of tasks running on the gpu's themselves (before I was running 2 on each card resulting in 6 concurrent tasks - but at that time I wasn't running anything on the CPU's) I have seen tasks complete in as little as 2m43s with this app.

Quite amazing thus far.

---

I didn't have an issue downloading the file myself at your link later in the day. I didn't get around to installing it til this morning but it is working.

I had to reschedule tasks to the GPU's so that it didn't get clogged by WU's that don't perform well under the GPU's. And I also had to change the amount of tasks running on the gpu's themselves (before I was running 2 on each card resulting in 6 concurrent tasks - but at that time I wasn't running anything on the CPU's) I have seen tasks complete in as little as 2m43s with this app.

Quite amazing thus far.

I am extremly interested it what you are learning. It will take a maybe a month or so to get my two Fermis installed, and water cooled. (Or I would have to sleep with the twin turbines) I feel so much better about going forward with my build knowing the results of your, Raistmer's, and Jason's work. I will ask more when the time is right, but a very sincere thank you for now. If anyone wishes it, I would be happy to test anything you can think of once I get these cards in and silenced...

Steve

---

Warning, addicted to SETI crunching!
Crunching as a member of GPU Users Group.
GPUUG Website

Just a note - Fermi cards don't like heat! Once they hit 80c they throttle way back. And of course by their nature they do run hot anyway and make sure if do remove them from an active system that you let them cool down since the heatsink is part of the exposed area on the card. I'm already exploring water cooling these because at full duty they will need it.
Todd

---

This is curious. If I click on my own link (which worked yesterday), I get "The service is unavailable".

But if I click on the copy of the exact same link which I posted in the Beta forum a few minutes later, the app downloads OK.

Just as a reminder, at present situation with .13 server we can post modified URLs like instead of
http://boinc2.ssl.berkeley.edu/beta/download/setiathome_6.09_windows_intelx86__cuda_fermi.exe

working 'hand-made' link is
http://208.68.240.18/beta/download/setiathome_6.09_windows_intelx86__cuda_fermi.exe

and it will work without 'The service is not available' ;)

I thought a little push from side 2 to 1 would be a well idea..

GTX4xx user, how work this app for you?

---

Todd Hebert, it's look like your 3x GTX480 machine is running the stock 6.09_cuda23, right?
Or you have DLed/installed the stock 6.09_fermi app from SETI@home BETA?

---

I am running an app that I got from Rastimer - it reports 6.08 but is using the CUDA 3.0 DLL's.
Working good - but of course not optimized - I only see about 75 max load on the GPU's and I am feeding them with a 1/3 of a processor core (Intel X5470 3.8Ghz) So one full core for the three GPU's. I have played with doing .5 and 1 core each and have seen really no difference.

Also, loading 2 tasks per GPU really slows all the tasks to a crawl. Average completion time for each task with 1/3 of a core is 8m45s.
Todd

---

I am running an app that I got from Rastimer - it reports 6.08 but is using the CUDA 3.0 DLL's.
Working good - but of course not optimized - I only see about 75 max load on the GPU's and I am feeding them with a 1/3 of a processor core (Intel X5470 3.8Ghz) So one full core for the three GPU's. I have played with doing .5 and 1 core each and have seen really no difference.

Also, loading 2 tasks per GPU really slows all the tasks to a crawl. Average completion time for each task with 1/3 of a core is 8m45s.
Todd

I am quickly depleting my cache, and have not done any CUDA work for about 6 days. My RAC is increasing to 13,619.99 totally on my 6 core CPU 980 running at 4.18 GHz. Without the GTX 295 blanking the screen and creating errors, I think I can get quite a bit more out of the processor. I had it up to 4.33 GHz for a couple of hours before the screen blanked, and sent me to the BSOD.

I'm getting all the hardware set up first for the two 480's, then I'm going to stress them with Furmark to see how hot they get with the water cooling. Once everything is set and in place, the crunching begins. Once I see where all this is going, and my rig is stable, I am willing to test new configurations, or what ever I can do to help out. The first thing I want to see though is a rising RAC. I can hardly wait for this weekend, when I will have time to get everything ready. This is really fun!

Steve

---

Warning, addicted to SETI crunching!
Crunching as a member of GPU Users Group.
GPUUG Website

Oooh - Im getting quite excited!
the seller will be taking back my 295 in a week for testing and when I get my refund, Im debating whether to put the extra in and buy a 480 - are we really that close to having them working?
Can someone using 480's post back and give an update on how well they are working, heat issues etc etc....

Goes off to check bank account and do some sums :-)

---

Oooh - Im getting quite excited!
the seller will be taking back my 295 in a week for testing and when I get my refund, Im debating whether to put the extra in and buy a 480 - are we really that close to having them working?
Can someone using 480's post back and give an update on how well they are working, heat issues etc etc....

Goes off to check bank account and do some sums :-)

edit

Just a note - Fermi cards don't like heat! Once they hit 80c they throttle way back.

I thought they were designed to run at 95C with a thermal limit of 105C???
This could cause issues as not everyone wants to watercool...

---

Go with an EVGA GTX480 Hydro Copper FTW*.
High OCed with water block.
Only 649.99 US$

I guess there are also other manufacturer out there which have OCed and water block inclusively.

[* ForTheWin, highest OCed series of EVGA]

---

Go with an EVGA GTX480 Hydro Copper FTW*.

I like how their design reduces the card to a single slot.

---

Note the hoses are at the top of the card - above where most cards end (the top of the slot).

I'm not suggesting cramming 4 of these in a single box, just that it's nice it doesn't cover up a perfectly good PCI slot like all the other ones do.

---

It looks the same with this card, but the advantage I have with the Danger Den cards is the nozzles are reversable.

Steve

---

Warning, addicted to SETI crunching!
Crunching as a member of GPU Users Group.
GPUUG Website

Edit - the .exe has the following dependencies. You should be able to get them from the same directory (untested).

LIBFFTW3F-3-1-1A_UPX.DLL
CUDART32_30_14.DLL
CUFFT32_30_14.DLL
NVCUDA.DLL

That last one is a new one on me.

Edit2 - None of the three CUDA DLLs named is present in the Beta download directory. I still think we need someone to attach and report back on what they are assigned by the server.

Sorry, it's taken me a while to get an answer to that. From official sources:
LIBFFTW3F-3-1-1A_UPX.DLL, CUDART32_30_14.DLL,
CUFFT32_30_14.DLL are part of the CUDA 3.0 Toolkit, NVCUDA.DLL is part of the driver.

I assume that the Fermi driver is meant.

As for Compute capability and such:
Fermi has compute capability 2.0. Note that this is not directly related to CUDA versions. To take full advantage of new Fermi features you need to upgrade to CUDA 3.0 (new features will be even added with newer CUDA versions). However there is no problem in running pre 3.0 apps on Fermi. The reason being is an HW independent intermediate assembly language, PTX, that CUDA apps are shipped with and that can be compiled to GPU native code - even to ones that have not existed when the PTX code was created. So if you run an app that has no Fermi specific machine code shipped with it, the driver will JIT compile the PTX version of the code to Fermi native the first time you run the app.

---

Yep, Richard Haselgrove discovered that a couple of days ago when a task he was predicting to fail, then completed and validated O.K,

Errors with Fermi & AP

Strangely, none of the GTX 470 / 480 Crunchers seemed to have picked up on this, yet.

Claggy