Astropulse v7 for Linux was deployed last night, my C2D T8100 has finished it's first Wu, and has almost finished it's second:

All AstroPulse v7 tasks for computer 68093

Claggy

One note, everyone may need to set Astropulse v7 to yes in their preferences to get work, all my venues had it set to No,

My two completed tasks had a wingman that errored all their tasks, the replacement tasks are listed as unsent at present, so tasks are either at back of queue, or their there is a lack of hosts asking for AP v7 work.

Claggy

I added my Linux host but before I could set the preferences correctly it loaded up on VLARs. I did manage to get 1 v7 AP. Unfortunately the P4 that hasn't completed a single task since Oct had already been there, Workunit 6466896. Since that host has nothing but Errors, could someone put a hold on any further tasks?

The 1 AP is about 1/3 complete at 3 hours, meaning about 10 hours total run time. Running r546 this host completes a normal AP in about 8 hours. The heavily blanked ones take about 10 hours. It's going to take a while to work through all those VLARs...

That host is using a too old kernel (2.6.16) to support some of the features (e.g. /proc -filesystem) that are minimum requirement ( assuming some kernel between 2.6.23 and 2.6.32 is minimum) to run this kind of app now. Maybe the app based on the stock codebase could be modified to support such "oldtimers", i don't know if that should be done for AstroPulse.
But MultiBeam should also provide such oldies.

---

_\|/_
U r s

How does this AstroPulse v7 work? I looked at my Beta preferences for my Mac and I see AP v6 & v7 is selected. I can't seem to download any GPU APs. I set it to just v7 OR v6 and I still can't download any Beta GPU APs on my Mac. Does AP v7 just work on Linux CPUs?

My Mac will be out of work tomorrow, might as well work a few Beta tasks, but, I can't download any Beta v7 or v6?

So far the only apps that have been deployed here for Astropulse v7 is for Linux, and for CPU only:

Seti Beta applications

Claggy

So far the only apps that have been deployed here for Astropulse v7 is for Linux, and for CPU only:

Seti Beta applications

Claggy

Ok...So, the only way anyone is going to get a validated WU is to work through all those already trashed 42 WUs? Then the third task will be sent out on those trashed WUs? Not Good. There's no way we're going to overtake someone trashing a task a second...

Finished the first one, http://setiweb.ssl.berkeley.edu/beta/workunit.php?wuid=6466896
The second one looks the same, http://setiweb.ssl.berkeley.edu/beta/workunit.php?wuid=6466902
So does the third, http://setiweb.ssl.berkeley.edu/beta/workunit.php?wuid=6466906
:-(

Ok...So, the only way anyone is going to get a validated WU is to work through all those already trashed 42 WUs? Then the third task will be sent out on those trashed WUs?
...

Because the host is reporting computation errors, the third task is added immediately thereafter. There will be a delay in actually assigning those third tasks since they are at the end of the ready to send queue, but it's better than a host turning in "success" results which are wrong and the third task not getting created until the first attempted validation is inconclusive.

The AP v7 work split on the 25th amounted to ~511 WUs, so with very few active APv7 hosts it would take considerable time to get things flowing well. Eric will try to get app versions for other platforms out ASAP, of course.

From earlier:

The 1 AP is about 1/3 complete at 3 hours, meaning about 10 hours total run time. Running r546 this host completes a normal AP in about 8 hours. The heavily blanked ones take about 10 hours.
...

With AP v7, CPU processing of heavily blanked tasks will be faster than unblanked ones. That's the major change, v7 skips the bad data rather than patching with shaped noise.

                                                                  Joe

With AP v7, CPU processing of heavily blanked tasks will be faster than unblanked ones. That's the major change, v7 skips the bad data rather than patching with shaped noise.

                                                                  Joe

Nice. From my experience the heavily blanked GPU tasks can take up to 300-400% longer where the CPU tasks are only around 10-20% longer. Eliminating all that lost GPU time will work for me.

So, there is going to be a method to send just the AP(sse2) tasks to applicable hosts? Seems the other tasks are around 3x as long. I have one that's only 21% complete after 5 hours, the sse2 ones are over 50% complete at that point.

With AP v7, CPU processing of heavily blanked tasks will be faster than unblanked ones. That's the major change, v7 skips the bad data rather than patching with shaped noise.

                                                                  Joe

Nice. From my experience the heavily blanked GPU tasks can take up to 300-400% longer where the CPU tasks are only around 10-20% longer. Eliminating all that lost GPU time will work for me.

So, there is going to be a method to send just the AP(sse2) tasks to applicable hosts? Seems the other tasks are around 3x as long. I have one that's only 21% complete after 5 hours, the sse2 ones are over 50% complete at that point.

Because GPUs do parallel processing, it's impractical for them to skip blanked areas in the same way that CPU processing can. But they'll avoid the huge penalty of creating and inserting the shaped noise blanking so GPU tasks will run at about the same speed regardless of blanking.

BOINC's server code does eventually adapt to send tasks for the fastest app version predominantly. For now, I think tasks for the generic versions need to be sent about as frequently as for the SSE2 versions for Beta testing purposes, and I presume Eric set the app plans that way.

                                                                  Joe

Windows CPU build is deployed too!
Lets start testing :)

Estimation time for:

Athlon x64 Venice: 263h45min
A10-5700 Trinity: 198h51min
Core2Duo 6420 Conroe: 318h1min

Estimation time for:

Core2Quad Q9450 Yorkfield: 179h36min
Athlon XP 2600+ Barton: 264h10min
C-60 Loveland: 1842h46min
Core i5-3470 Ivy Bridge: 152h51min

Estimation time for:
A-10 5700 Trinity: 181h
Athlon 64 3200+ Winchester: 260h32min

Estimation time for:

Core2Quad Q9450 Yorkfield: 179h36min
Athlon XP 2600+ Barton: 264h10min
C-60 Loveland: 1842h46min
Core i5-3470 Ivy Bridge: 152h51min

My Q9400 had similar estimates for the base App. Actual times are around 24 hours for the base App and 10 hours for the SSE2 one. Hopefully the extreme estimates are affecting the credit awards. I'm waiting to see the credits on my tasks, 'cause, I corrected the estimates closer to actual times. I'm waiting...v7 tasks for computer 72013
Oh look, only 10.7 hours left on the currently running v7 7.00...

I'm waiting to see the credits on my tasks, 'cause, I corrected the estimates closer to actual times.

Mine have been 249.83, 232.67 and 405.28, I expect around 235 might be the new norm.

Claggy

I'm waiting to see the credits on my tasks, 'cause, I corrected the estimates closer to actual times.

Mine have been 249.83, 232.67 and 405.28, I expect around 235 might be the new norm.

Claggy

how much was AP paying before abouts?

---

A person who won't read has no advantage over one who can't read. (Mark Twain)

I'm waiting to see the credits on my tasks, 'cause, I corrected the estimates closer to actual times.

Mine have been 249.83, 232.67 and 405.28, I expect around 235 might be the new norm.

Claggy

how much was AP paying before abouts?

Around 700

Claggy

I'm waiting to see the credits on my tasks, 'cause, I corrected the estimates closer to actual times.

Mine have been 249.83, 232.67 and 405.28, I expect around 235 might be the new norm.

Claggy

how much was AP paying before abouts?

When AP V5 was out each unit got 1248 cr.
After credit screw it shrank to ~900
Now 700 but i see a lot of 600 and 500 as well at main now.

But we should concentrate on app testing first.

---

With each crime and every kindness we birth our future.

I'm waiting to see the credits on my tasks, 'cause, I corrected the estimates closer to actual times.

Mine have been 249.83, 232.67 and 405.28, I expect around 235 might be the new norm.

Claggy

how much was AP paying before abouts?

And before the 1248 cr awards, it was ~700 Credits,
Eric then did a normalising script for Seti v6, and didn't do/apply it to AP hence the ~1250 awards,
I think when it was applied, Credit was ~800, before dropping slowly towards 700 cr.

When AP V5 was out each unit got 1248 cr.
After credit screw it shrank to ~900
Now 700 but i see a lot of 600 and 500 as well at main now.

But we should concentrate on app testing first.

Agree, But it's got to be mentioned.

Claggy

When AP was being granted credit under Eric's work-based method with adjustment to Cobblestone rate, credits were between 700 and 800.

When CreditNew was first implemented, the adjustment factor was no longer available but there was some difficulty getting AP data into CreditNew so for an extended period AP credits were still on the work-based method but unadjusted. That was the period when 1284 credits were granted for AP tasks. It was wrong, but enjoyable.

                                                                   Joe