The only time I've seen them on the 980s or Titans is usually after some update they done and forgotten to block them from being sent.

I would have noticed them before since they take about an hour or more to crunch.

Yes, I am one of those that constantly looks at the Manager to see what is going on (bad habit I know) what happens when you got an extra screen to play with.

My GTX980 is compute capability 5.2

I missed out on the 'at least' in the original message. Edited in now. :)
But even if yours has 5.2, it then has 3.0 capability and more. :D

Eric Korpela wrote:

If nobody has overwritten the scheduler I put in yesterday night, the check is still there. But it does give VLAR to NVIDIA compute capacity 3.0 and above, and AMD CAL target greater than 12, and AMD CAL target == 0 (i.e. a new GPU that doesn't do CAL).

---

Better news.

Jeff Cobb wrote:

Oscar is back up and mysql is recovering.

This does not automatically mean that the site is back, the forums are up or anything.

---

Eric Korpela wrote:

If nobody has overwritten the scheduler I put in yesterday night, the check is still there. But it does give VLAR to NVIDIA compute capacity 3.0 and above, and AMD CAL target greater than 12, and AMD CAL target == 0 (i.e. a new GPU that doesn't do CAL).

Strange, I have never been getting VLAR's for my 980 on main, except when something was wrong with the scheduler. Are you sure his answer was regarding Main, and not regarding Beta?

As far as I know, they haven't added all the changes to Beta yet. But I can ask. As you see, I seem to be on the answer memo list. :)

---

Edit: I still think that an opt-in/opt-out choice for VLAR's on GPU's would be the best solution, where opt-out should be default.

Or where opt-in is default but only if you have problems, can you opt-out.
Of course, just because they run slow on your GTX isn't a good enough reason, as to be allowed to opt-out then is pure cherry-picking. Only if your GTX only returns errors would it be viable.

---

Quick turn around.

Jord in email wrote:

Hi Eric,

Question about the scheduler and VLARs from within the BOINC forums.
People with GTX 980s and Titans never saw VLARs before on their GPUs
on Seti Main. They did see them on Seti Beta, though. So they wonder
if your comment about the scheduler was for Beta, or for Seti Main?

If for Seti Main, then it's a new thing, or it didn't work before. :-)

Eric in email wrote:

You're correct on that. The scheduler that was on main was from Oct 2013, which would be before this change was made.

---

OK, my GTX980s & GTX960 have picked up VLARs, I can't say on the others as they didn't manage to get anything in time :-(

I would rather see "no option unless producing errors/invalids".

My GTX980 is compute capability 5.2

I missed out on the 'at least' in the original message. Edited in now. :)
But even if yours has 5.2, it then has 3.0 capability and more. :D

Eric Korpela wrote:

If nobody has overwritten the scheduler I put in yesterday night, the check is still there. But it does give VLAR to NVIDIA compute capacity 3.0 and above, and AMD CAL target greater than 12, and AMD CAL target == 0 (i.e. a new GPU that doesn't do CAL).

I got this:

30-Jul-2015 01:28:25 [SETI@home] Sending scheduler request: To fetch work.
30-Jul-2015 01:28:25 [SETI@home] Requesting new tasks for NVIDIA GPU
30-Jul-2015 01:28:25 [SETI@home] [sched_op] CPU work request: 0.00 seconds; 0.00 CPUs
30-Jul-2015 01:28:25 [SETI@home] [sched_op] NVIDIA GPU work request: 2910.10 seconds; 0.00 GPUs
30-Jul-2015 01:28:28 [SETI@home] Scheduler request completed: got 3 new tasks
30-Jul-2015 01:28:28 [SETI@home] [sched_op] Server version 707
30-Jul-2015 01:28:28 [SETI@home] Project requested delay of 303 seconds
30-Jul-2015 01:28:28 [SETI@home] [sched_op] estimated total CPU task duration: 0 seconds
30-Jul-2015 01:28:28 [SETI@home] [sched_op] estimated total NVIDIA GPU task duration: 3542 seconds
30-Jul-2015 01:28:28 [SETI@home] [sched_op] Deferring communication for 5 min 3 sec
30-Jul-2015 01:28:28 [SETI@home] [sched_op] Reason: requested by project
...
30-Jul-2015 01:28:35 [SETI@home] Started download of 08my15ad.16840.17658.438086664196.12.49.vlar
30-Jul-2015 01:28:38 [SETI@home] Finished download of 08my15ad.16840.17658.438086664196.12.49.vlar

with a

17-Jul-2015 00:35:14 [---] NVIDIA GPU 0: GeForce GTX 750 Ti (driver version 34475, CUDA version 6050, compute capability 5.0, 2048MB, 410 GFLOPS peak)

I think Eric is mis-remembering the policy. We did try VLAR to Kepler (cc 3.0) when the new cards first came out, three years ago: Keplers were released "big ones first", and we thought they might have enough heft to handle VLARs. This cc 5.0 is a Maxwell - they were released with the smaller cards first. VLARs are not good on any NV cards - for some people they cause screen lag to the point of unuseability, for others they just run slower and hence mess up the runtime estimate averages.

For at least the last two, maybe 2.5, years the policy has been "VLAR to cc 3.0 and above at the Beta project only, no VLARs to NV at the main project. I can find you the exact quote on the Beta message board once it's back up (I had a hand in testing and reporting at Beta, leading to the policy decision).

Edit: I still think that an opt-in/opt-out choice for VLAR's on GPU's would be the best solution, where opt-out should be default.

Or where opt-in is default but only if you have problems, can you opt-out.
Of course, just because they run slow on your GTX isn't a good enough reason, as to be allowed to opt-out then is pure cherry-picking. Only if your GTX only returns errors would it be viable.

There was a call on the Beta forums recently for a 'VLAR to NV' option, with the suggestion it should be defaulted 'on' at Beta, defaulted 'off' at the main project. I'd support that balance.

But with opt-in as default, the newbies, or the set and forget crunchers with GPU's that can't handle VLAR's without making the screen lag, would be immediately peed-off, and uninstall BOINC/SETI.

The newbies have to change something to see screen lag: by default BOINC only uses the GPU when the computer is idle, just for that reason.
Added to that, even non-VLARs will at times make the mouse and screen lag. There is information enough out there that explains why that is, and if the newbie then still wants to uninstall, let them, as they don't want to learn about the why and options around it. (My opinion!)

I also doubt that "set it and forget it users" will have any such problem. They forgot for a reason, didn't they? :)

When they've been getting a bit more used to SETI, and understands that they can change many different settings on the web, then they could try and opt-in for VLAR's, and also know how to opt-out if it doesn't work for their GPU.

Uh-uh, so you think that a big cruncher allows his banks of 980s to slowly run VLARs, when he can opt-out without penalty and run faster work instead? That instead other people with GPUs will opt-in?

Also, everyone reading about the scheduler on Main, read Eric's answer that I posted in https://boinc.berkeley.edu/dev/forum_thread.php?id=8105&postid=63328. :)

---

I remember asking for that on beta, if we got the new opencl_nividia_sah to run them on.

With that new app, VLARs running 1 at a time were 20 minutes. Running 2 at a time was 30 minutes.

However, normal MBs were slower than cuda 50. I don't know if they are dedicate VLARs to only go to the opencl_nvidia_sah. If they could, that would be the best of both worlds.

Normal MBs on Cuda 50 and VLARs on opencl_nvidia_sah.

Currently VLARS on my i7 take 1 hr 35 mins.

The option of allowing them to run on main with the new app would be a great help I think.

The reason I asked for the app and option of running VLARs on main is when we get a VLAR storm.

Then no one gets any GPU work until it clears. If we had the option of running the VLARs on the new app we could help to clear the storm faster.

Yes, there will be those that won't want to run them as it will be slower than normal MBs, but when a big storm hits, there isn't any work for the GPUs anyway.

My 2 cents

Zalster

Good morning everyone. :-) Just got up a little while ago. Got the computer booted, only to find that Main and Beta are down... :-(

I still have some work that I have on Suspend until all this gets worked out.

As to new VLARs on NVIDIA, I don't know if mine was a VLAR, but I noticed that I had a NEW AP unit marked as "V7 7.10" and it seemed to have completed normally. This was on Exeter, my system with the GTX-760. (Also on Lunatics 0.43b.)


TL

---

Have TARDIS, will travel...
Come along K-9!
Join Calm Chaos
Pluto is still a planet

Message boards have just come back up (Main and Beta).

Great, immediately find someone had PMed me there to ask me if the project was down. Well...

---

Good morning everyone. :-) Just got up a little while ago. Got the computer booted, only to find that Main and Beta are down... :-(

I still have some work that I have on Suspend until all this gets worked out.

As to new VLARs on NVIDIA, I don't know if mine was a VLAR, but I noticed that I had a NEW AP unit marked as "V7 7.10" and it seemed to have completed normally. This was on Exeter, my system with the GTX-760. (Also on Lunatics 0.43b.)


TL

VLARs are Multibeams TL. That v7.10 is the newer Astropulse app that was causing problems before the fixes were installed on v43b Lunatics installer

Quick turn around.

Jord in email wrote:

Hi Eric,

Question about the scheduler and VLARs from within the BOINC forums.
People with GTX 980s and Titans never saw VLARs before on their GPUs
on Seti Main. They did see them on Seti Beta, though. So they wonder
if your comment about the scheduler was for Beta, or for Seti Main?

If for Seti Main, then it's a new thing, or it didn't work before. :-)

Eric in email wrote:

You're correct on that. The scheduler that was on main was from Oct 2013, which would be before this change was made.

Still not quite. The test was made in May 2013: SETI Beta message 46039.

As a test I'm sending VLARs to comp-cap 3.0+ NVIDIA cards. Anyone with a Kepler or higher let me know how they work.

I think we have a difference of opinions about this, that's OK, I can live with that :-)

**Jord's opinion!**

Offer people an option to opt-out of something they deem unproductive and they will take it.
Offer people an option to opt-in on something they deem unproductive and no-one will take it, not even those with capable hardware.

We've seen it before, in some sense, as else you wouldn't have had the need for the Rescheduler app on Main. Or before the choice of applications was available, people who would abort Multibeam tasks because they don't pay as much as Astropulse do. Even now there are still some who only have AP checked.

**/Jord's opinion!**

---

Still not quite. The test was made in May 2013: SETI Beta message 46039.

I think you're reading Eric's answer wrong. I think he was answering to my last sentence: "If for Seti Main, then it's a new thing, or it didn't work before. :-)"

You're correct on that. The scheduler that was on main was from Oct 2013, which would be before this change was made.

With this change being the one from last night that added VLAR capability to Main.

---

So does that mean we should expect opencl_nvidia_sah on Main?

Or are we going to have to import it from beta to crunch them?