SETI@Home Wow!-Event 2016

[TennesseeTony]

Uhm, are these new v8 GPU tasks heavy on double precision? I've (re)installed the wrong brand graphics cards if so.... Time to fire up the daughter's rig, with a R9-280X and test for myself.

 

[TennesseeTony]

280X is substantially slower than the 980, not enough results to give data, similar tasks take widely varying amounts of time to complete as well. I'll let it run overnight and update.

 

[Pokey]

That is my feeling also. My R9-290s do not seem to be as productive as the GTX-970s. Seti work is single precision.

The other side of this is that the GUPPI work processes faster on cpu and non-Guppi work processes better on GPU, but the work units get assigned to your computer randomly. Hence the need for the GUPPI rescheduler.

At least that is my understanding.................which means I could be wrong.

 

[Pokey]

Unfortunately right now we are having a shortage of work due to the GUPPI splitters producing no usable work

I would expect you and Tony to see an uptick in production now that the guppi work is flowing again. The cpu heavy rigs seem to do much better with them. Mine are gpu heavy and are taking a hit right now.

 

[Kiska]

Unfortunately, with the way Guppi's are, they are all VLAR work which means that the angle range is < 0.12, this type of work, GPU hate them.... Need I say more? There's a reason why people are using that rescheduler to better optimise their production. By sending VLAR work to the CPU and the other work to the GPU.

 

[Assimilator1]

Interesting, never heard of that app before!

Damn, just noticed that LHC has grabbed a bunch of WUs, bad timing, I'll have to pause it.

 

[Pokey]

QuietDad, it looks like our Pisces group is firmly ensconced in 9th place.

I usually like to look at the countries and find the most remote or obscure. This year Saint Kitts and Nevis jumps out (I actually had never heard of them) and Iceland. Sorry Kiska,I don't count Australia, even though it is way the heck and gone down under.

No Tony, TN doesn't count.

So far my grand prize would probably go to Pitcairn Islands last year or Antarctica. They did not show up this year.

 

[Kiska]

Interesting, never heard of that app before!

Damn, just noticed that LHC has grabbed a bunch of WUs, bad timing, I'll have to pause it.

It is located here

Sorry Kiska,I don't count Australia, even though it is way the heck and gone down under.

Thats a shame!

 

[TennesseeTony]

Fun stuff right here, regarding this year's #1 racer:

http://setiathome.berkeley.edu/hosts_user.php?userid=10192508

I think running this for only 24 hours would cause a panic in my energy budget.

 

[TennesseeTony]

1.75 Days left in this race. Doesn't look like I'm going to make it to 50th place this year...oh so close, in 54th place, should climb one or two spots before the end of the race.

On a positive note though, last year I had LOTS of machines running, only 3 running for this year's race, and a 4th for GPU comparison for a short while.

Wait, I still have lots of machines....just not running....I'll be right back.

Muwahahahaahaaa! Take that electric company and my SETI competitors!!!

Release the hounds!!

 

[Pokey]

Way to go Tony!!.............. page 1 & top 50. I was hoping you would give in and put the pedal to the metal................

Hey, where are our thumbs up emojis??? Pretend "thumbs up".

 

[TennesseeTony]

edited: I have taken 50th place as of the time of this posting, 49th will be next hour, and I have a shot now at 48th before the race ends, but it will be really close. /end edit

**********************************

In other news:

The more GPUs I toss in the less impressive the return it seems, as each GPU 'thread' requires up to one whole CPU thread to help it along. And even with Lunatics and the rescheduler, I have nothing but GUPPIs most times on my GPUs. I wonder if it might have been more efficient to have fired up the extra CPU power rather than all the GPU power from the start.

Mjolnir for example: 12 CPU threads available, dual 980Ti's with 4 threads each, and I'm at 97-100% CPU with only 4 CPU threads running. I've lost 2/3rds of my CPU power here.

WUSS: 28 CPU threads available, triple 980's at 4 each (12 GPU tasks), 98% CPU with only 14 CPU tasks running. Lost more than 50% of CPU power here.

BumbleBee: 28 CPU threads, plus 5 (five? messed up the app config I guess, lol) GPU threads, only running 22 CPU tasks and at 98% CPU.

In all instances, each GPU 'thread' is using one whole CPU thread. Is that common for SETI? Anyone else have the same usage? I've seen other projects use a whole CPU/thread for a short period of time, at the beginning and end of the GPU task usually, but this is continuous usage for SETI it seems.

 

[Pokey]

The combinations and permutations are almost endless, but typically in seti giving up a cpu or part of a cpu to run a gpu is no contest because the gpus are so much more productive.
Suffice it to say the learning curve this year was steeper than usual for me anyway.

Anyhow we could solve that by following what BojanR did and use company computers, one of which had 88 cpus and several others which had 72 cpus and then mix in a basket full of AMD Fiji cards, 3 & 4 at a pop.................. ...................... But that is pie in the sky for most of us.

Still I enjoy the annual event. Looking forward to Dec. now.

 

[Kiska]

Well you can bring down cpu usage of the gpu app by using the -use_sleep switch/command line, on the OpenCL apps at least, don't know what the switch is for the cuda app

 

[Assimilator1]

Fun stuff right here, regarding this year's #1 racer:

http://setiathome.berkeley.edu/hosts_user.php?userid=10192508

I think running this for only 24 hours would cause a panic in my energy budget.

GenuineIntel
Intel(R) Xeon(R) CPU E5-2686 v3 @ 2.00GHz [Family 6 Model 63 Stepping 2]
(72 processors)

Wow! That's insane! lol, I assume that's a multi socket multi core CPU & not just 1 CPU

Btw Tony, I've found with both MW & E@H that to get decent output from my GPU I need to dedicate a whole core. Not tried SETI on my GPU as it's too warm to run DC on it atm.
Congrats on entering the Top 50 btw

 

[Kiska]

Btw Tony, I've found with both MW & E@H that to get decent output from my GPU I need to dedicate a whole core. Not tried SETI on my GPU as it's too warm to run DC on it atm.
Congrats on entering the Top 50 btw

You can use -use_sleep with seti opencl app so it doesn't hit the cpu too heavily

 

[Pokey]

Well, the WOW event is over and some of our guys made it up onto the podium. Congratulations everyone. It was fun as always.

 

[TennesseeTony]

Well you can bring down cpu usage of the gpu app by using the -use_sleep switch/command line, on the OpenCL apps at least, don't know what the switch is for the cuda app

Wouldn't that handicap the GPU though?

 

[Kiska]

Wouldn't that handicap the GPU though?

It uses sleep loops in some places which results in lower cpu usage.

 

[Assimilator1]

You can use -use_sleep with seti opencl app so it doesn't hit the cpu too heavily

Wouldn't that drop GPU output a lot though? (your above post isn't an answer btw ).