What help it, if you announce it here, because only a very small number of SETI@home members (Lunatics privileged) could test this app?

---

@Lunatics beta testers,
'x32f' build is added as an attachment in the beta thread. It's the same as 'x32e' with some cosmetic stderr tidyups and some minor changes *hoped* to address GTS250 & GT 240 issues (needs testing all around, but especially on those cards) ... If it works on those, it will replace the original x32 build in beta downloads Due to the apparent enhanced reliability we're seeing from x32e. If it doesn't fix issues on those cards ... then I'll keep looking ;)

Cheers, Jason

Do you mean the GTS250 and GTS240 (these are older G92b GPUs), the GT240 is a different beast based on the GT215 GPU, and not released until Nov 2009.

Do you mean the GTS250 and GTS240 (these are older G92b GPUs), the GT240 is a different beast based on the GT215 GPU, and not released until Nov 2009.

*I believe* the issue with particular earlier test builds was spotted on GTS 250 & GT 240 (without the S). But I would appreciate tests of the 'x32f' build on ANY cuda capable cards with sufficient driver (series 256+) & Video RAM (256meg probably won't cut it, at least 384meg on card runs OK, don't know exact figure).

Major goal in this test build is decreased susceptibility to '-12 errors' (though not expected to be totally eliminated ... yet) ... and *hoping* to fix errors spotted on those particular cards. Not intended as a 'final public release' build, but mostly a 'sanity check' ... which I suppose is an odd thing for Lunatics to be doing.

Thanks for asking :)
Jason

---

"Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions.

I am running a GT240, 1 gig RAM, for about 2 months, and so far only had a few -12 errors, fairly recent driver but can't check actual version until I get home, hopefully tomorrow.

I connect to Lunatics when I get home and look at the instructions etc.

I am running a GT240, 1 gig RAM, for about 2 months, and so far only had a few -12 errors, fairly recent driver but can't check actual version until I get home, hopefully tomorrow.

I connect to Lunatics when I get home and look at the instructions etc.

Cheers!.

---

"Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions.

I am running a GT240, 1 gig RAM, for about 2 months, and so far only had a few -12 errors, fairly recent driver but can't check actual version until I get home, hopefully tomorrow.

I connect to Lunatics when I get home and look at the instructions etc.

Cheers!.

Managed to contact son, he says last Nvidia driver on the network drive is 257.16 d/loaded 16/06/2010, I think that is driver in use.

(he said don't contact me now, "I'm cooking my dinner".)

Managed to contact son, he says last Nvidia driver on the network drive is 257.16 d/loaded 16/06/2010, I think that is driver in use.

(he said don't contact me now, "I'm cooking my dinner".)

LoL, that one should work. For reference, I use 257.21 ... though any 256 series is supposed to work.

Whatever happened to sons that sit around the computer drinking beer ?

[A bit Later:] Cancel that GT 240 test! it still fails on that one (bummer) works great on 9 series & GTX200 upwards, so will keep looking.

Thanks for the offer to test, I'll take you up on that when I figure out what's going on with those.

---

"Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions.

Managed to contact son, he says last Nvidia driver on the network drive is 257.16 d/loaded 16/06/2010, I think that is driver in use.

(he said don't contact me now, "I'm cooking my dinner".)

LoL, that one should work. For reference, I use 257.21 ... though any 256 series is supposed to work.

Whatever happened to sons that sit around the computer drinking beer ?

The drinking one, who's most common comment was "It isn't my fault Dad, it fell apart in my hands" has found a software tester's job many miles away.

Test back on for later. GT 240 functionality still uncertain.

---

"Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions.

It's all very interesting.

The "Cricket Graph" looks like some crazy-town skyline, or some crazy-town forest, or some (insert analogy here).

Just remember:

As with all things, this too shall pass. :)

Yep, it is passing right now. Last skyscraper disappears and only the desert remains.
I just hope that oasis wont be far away... or at least small settlement. :-)

---

Who the hell is General Failure and why is he reading my harddisk?¿

It's all very interesting.

The "Cricket Graph" looks like some crazy-town skyline, or some crazy-town forest, or some (insert analogy here).

Just remember:

As with all things, this too shall pass. :)

Yep, it is passing right now. Last skyscraper disappears and only the desert remains.
I just hope that oasis wont be far away... or at least small settlement. :-)

Yeah that looks mighty bleak, Hopefully they can get the servers to run on schedule like they've planned.

---

The T1 Trust, PRR T1 Class 4-4-4-4 #5550, 1 of America's First HST's

Now I am getting memory errors I guess...

Work Unit Info:
...............
WU true angle range is : 2.726284
After app init: total GPU memory 939524096 free GPU memory 873984000


Unhandled Exception Detected...

- Unhandled Exception Record -
Reason: Out Of Memory (C++ Exception) (0xe06d7363) at address 0x76EAB727

Engaging BOINC Windows Runtime Debugger...

I guess I should check my e-mail once in awhile. I noticed that there is talk of a memory leak in the .57 boinc. Could this be what is causing these memory errors? I just changed back to the .56 and will see if I get the same errors...

---

Official Abuser of Boinc Buttons...
And no good credit hound!

Now I am getting memory errors I guess...

Work Unit Info:
...............
WU true angle range is : 2.726284
After app init: total GPU memory 939524096 free GPU memory 873984000


Unhandled Exception Detected...

- Unhandled Exception Record -
Reason: Out Of Memory (C++ Exception) (0xe06d7363) at address 0x76EAB727

Engaging BOINC Windows Runtime Debugger...

I guess I should check my e-mail once in awhile. I noticed that there is talk of a memory leak in the .57 boinc. Could this be what is causing these memory errors? I just changed back to the .56 and will see if I get the same errors...

I'm using .56 boinc and still getting that error every so often on GPU work units.

---

- Jarryd

The memory leak which Jord and I are attempting to debug on boinc_alpha is to do with the system RAM usage of boinc.exe when installed as a service on 32-bit Windows computers.

That is rather different from anything a project science application might do with GPU VRAM.

The memory leak which Jord and I are attempting to debug on boinc_alpha is to do with the system RAM usage of boinc.exe when installed as a service on 32-bit Windows computers.

That is rather different from anything a project science application might do with GPU VRAM.

In that case, that memory leak doesn't affect me anyway. I'm using x64 and it's not installed as a service. :)

Richard, do you know of any possibilities of why these Out of memory errors may occur on a GPU?

---

- Jarryd

The memory leak which Jord and I are attempting to debug on boinc_alpha is to do with the system RAM usage of boinc.exe when installed as a service on 32-bit Windows computers.

That is rather different from anything a project science application might do with GPU VRAM.

In that case, that memory leak doesn't affect me anyway. I'm using x64 and it's not installed as a service. :)

Richard, do you know of any possibilities of why these Out of memory errors may occur on a GPU?

The best person to ask would be Jason.

Off the top of my head:

Operating system eye-candy using too much VRAM
other active graphics-internsive tasks (games, video playback?) ditto ditto
A badly-written device driver or SDK component not releasing memory when it's supposed to
A badly-written end user application ditto ditto (surely not!)

The memory leak which Jord and I are attempting to debug on boinc_alpha is to do with the system RAM usage of boinc.exe when installed as a service on 32-bit Windows computers.

That is rather different from anything a project science application might do with GPU VRAM.

In that case, that memory leak doesn't affect me anyway. I'm using x64 and it's not installed as a service. :)

Richard, do you know of any possibilities of why these Out of memory errors may occur on a GPU?

The best person to ask would be Jason.

Off the top of my head:

Operating system eye-candy using too much VRAM
other active graphics-internsive tasks (games, video playback?) ditto ditto
A badly-written device driver or SDK component not releasing memory when it's supposed to
A badly-written end user application ditto ditto (surely not!)

Thanks for that. I might give Jason a message.

---

- Jarryd

The best person to ask would be Jason.

Off the top of my head:

Operating system eye-candy using too much VRAM
other active graphics-internsive tasks (games, video playback?) ditto ditto
A badly-written device driver or SDK component not releasing memory when it's supposed to
A badly-written end user application ditto ditto (surely not!)

Thanks for that. I might give Jason a message.

In addition to Richard's comments, looking at your host and some of your tasks to get an idea, In the 'Out of Memory' circumstance I'd suggest a regular reboot will avoid them. I do see one of Raistmer's exceptions triggering on a '-12', separate issue with a similar looking crash.

Some background: XP's Driver model uses a physical memory management scheme for memory on the card. Over time such a setup is prone to memory fragmentation, so allocations may fail (Causing an 'Out of Memory' exception) even if there is plenty of memory free.

Vista & Win7 has moved to Microsoft's WDDM driver model, which uses a 'paged' memory management model. There are speed penalties for this through CPU use & PCIe bus traffic, but adds reliability as each application video RAM context becomes isolated (ignoring driver bugs), and the ability for the driver to be restarted in the event of failure.

In Addition: Ignoring that there are potential driver bugs we don't know about, since current release codebases don't know about WDDM 'stuff' yet, If the driver crashes (for any reason internal or external to our app) and the machine recovers, then the cuda context is lost and the app can crash with any 'weird' error.

To recover the cuda context must be re-established from scratch, at any point. There is no code to do this yet, and I am looking into it.

I am uncertain whether XP x64 (presumably not using WDDM) will allow these kinds of recovery, but if reliability is a concern, In the long run it may be one good reason to sacrifice speed for reliability by migrating to Windows 7 (and, of course, hoping we can figure out how to properly implement recovery mechanisms)

Jason

---

"Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions.

The best person to ask would be Jason.

Off the top of my head:

Operating system eye-candy using too much VRAM
other active graphics-internsive tasks (games, video playback?) ditto ditto
A badly-written device driver or SDK component not releasing memory when it's supposed to
A badly-written end user application ditto ditto (surely not!)

Thanks for that. I might give Jason a message.

In addition to Richard's comments, looking at your host and some of your tasks to get an idea, In the 'Out of Memory' circumstance I'd suggest a regular reboot will avoid them. I do see one of Raistmer's exceptions triggering on a '-12', separate issue with a similar looking crash.

Some background: XP's Driver model uses a physical memory management scheme for memory on the card. Over time such a setup is prone to memory fragmentation, so allocations may fail (Causing an 'Out of Memory' exception) even if there is plenty of memory free.

Vista & Win7 has moved to Microsoft's WDDM driver model, which uses a 'paged' memory management model. There are speed penalties for this through CPU use & PCIe bus traffic, but adds reliability through each application video RAM context being isolated, and the ability for the driver to be restarted in the event of failure.

In Addition: Ignoring that there are potential driver bugs we don't know about, since current release codebases don;t know about WDDM 'stuff' yet, If the driver crashes (for any reason internal or external to our app) and the machine recovers, then the cuda context is lost and the app can crash with any weird error.

To recover it must be re-established the cuda context from scratch at any point. There is no code to do this yet, and I am looking into it.

I am uncertain whether XP x64 (presumably not using WDDM) will allow these kinds of recovery, but if reliability is a concern, In the long run it may be one good reason to sacrifice speed for reliability by migrating to Windows 7 (and, of course, hoping we can figure out how to properly implement recovery mechanisms)

Jason

Thank you very much for that information.

As far as i know you're right about x64 not using WDDM after having a quick squiz on google. I'll try rebooting every so often see if i get less of those types of errors. :)

---

- Jarryd

Thank you very much for that information.

As far as i know you're right about x64 not using WDDM after having a quick squiz on google. I'll try rebooting every so often see if i get less of those types of errors. :)

Thank you for bringing it up. I've been up to my eyeballs in documentation with exactly these issues. If the restarting helps (or doesn't) Let me know.

Cheers, Jason

---

"Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions.