First FB Sprint of 2018 starts on Mar 21, 6:00am/9:00am PST/EST

Also a list of all the members that signed up for us yesterday. Should have a new list of a couple more today as well.

upload_2018-3-23_16-43-1.png
 
Have 3,255 pending... would think with the Sprint happening that pending WUs would be blazing by in no time flat.
 
Most likely people abusing the system by requesting WU's and then holding them until after the Sprint. I believe I remember Gilthanis talking about teams doing something similar during the Pentathlon last year.

So far my pending hasn't been too bad.
 
Yes, that is a defensive tactic by some teams and is typically used when a project has trouble supplying enough work units to fulfill demand. However, I doubt it really matters much on SETI....lol .
 
I'm doing half as as much as you do. Ignore the 345 errors as I was tinkering with lunatics.....

upload_2018-3-23_14-34-17.png
 
Well I think my other 48 core rig is okay now. She's been running 100% on all 48 cores for a little over an hour now without crashing. So I must have fixed it reseating the ram modules. Weird. Hopefully she at least finishes the WU's that's been running for the past hour.
 
4/6 rigs are down. Heh. I do not think the clocks I use for mining are going to work for these projects.

Anyhow, I'll be back in town in roughly 5 or so hours to get all the other rigs up.
 
We are 50k points from 1st and steadily closing in on them.

Get those systems online!
 
my systems fetched some unexpected yafu WUs ... wtf ... never got those ... lucky the GPUs still full running
 
Do the super optimized seti apps still work? I now have access to a system that has all the fancy cpu things, so I probably can get full speed from the gpu....
 
I'm not sure. Not using any of them, pututu might know.

Also, someone photoshop a giant lawnmower with xtremesystems in it's path because we're about to mow those sons a bitches! I'll post it on the public forum if you don't want.
 
We passed 1/2 way-mark through the sprint. Keep crunching as much as possible and hopefully to see a lot of validation in the last hours of the sprint.
MN Scout said:
Do the super optimized seti apps still work?
Click to expand...
Are you running on Windows or Linux?. I don't see much gain in Windows but if you look at one of the top computers with 4 x 1070 in linux, he/she is averaging close to 200K PPD per rig. If you click on any of the task details, he/she is running cuda 9.0. If I'm not mistaken (haven't tried it since I don't have linux box) you can download this from the forum here. Please post any new findings here if you decide to experiment with it.
 
we still have a better increase value ; next one or two updated we should have them passed

Screen Shot 2018-03-24 at 12.46.40.png

(one data point in the middle seems not transferred; thats why higher current value; divide by two ... )
 
Alright. The rigs are up. Perhaps get a solid day.

Where da lawn mower pic?

:D
 
and funny enough ... we outperform them all in our league though not showing up with any points (I know, its only the sprint; incl. myself); never seems such a green line ...

Screen Shot 2018-03-24 at 13.06.36.png
 
We'd at least make it into the top 10 in League 1... lol

Edit - Only 2 of my machines are running anywhere near optimal.

In fact, i feel that i could easily double production, possibly triple if i have dialed-in setups (both hardware/software).

Edit 2 - Beginning to dial stuff in.
  • Some of the rigs (1070Ti1) just want to play dumb versus other rigs (1070Ti2) with identical software/hardware layouts.
  • 1070 / 1070 Ti's seem to handle 4 tasks well without cuda memory threadsafe errors.
    • 3 tasks per may be a more stable route in the long term.
  • Definitely CPU limited: The G3930s are 100% utilized even @ .04cpu per task.
    • Watching individual GPU load and seeing load anything but pegged at 100% at 3 or 4 tasks per.
  • System ram @ 4GB is barely sufficient. (90% utilization).
  • The 8.22 SETI@home v8 (opencl_nvidia_SoG) app seems to provide best WU times. (all 4 complete in ~35min)
  • The other cuda/opencl apps are all 1h + @4 tasks per.
  • The Astropulse app runs way too long on the GPUs.
  • Clocks are extremely fragile - had to dial stuff down to stock to prevent cuda errors.
  • Threadrippers phoenicis holy batman dude
  • Still need to put Rig 6 together. (+6 1070Ti / +2 1080)
 
Last edited:
atp1916 said:
We'd at least make it into the top 10 in League 1... lol

Edit 2 - Beginning to dial stuff in.

  • 1070 / 1070 Ti's seem to handle 4 tasks well without cuda memory threadsafe errors.
    • 3 tasks per may be a more stable route in the long term.
Click to expand...


I am going to have to play around with this. I am only doing 1 task per gpu currently. 1080ti seems to take 3min 50 sec, 1070's are 4min 20 or something like that.
 
On my Ryzen / 1070 workstation, if i do 50% CPU / 50% CPU time, with 4 task per GPU and .25 CPU per 4 tasks i can get ~20min for 4 simultaneous tasks. That is roughly 5 min per task, which, all things considered isn't too bad i feel.

I feel i could break 100k if i had all machines with enough CPU power per.
 
atp1916 said:
Threadrippers phoenicis holy batman dude
Click to expand...
Phoenicis's GPUs are running optimized cuda 9.0 special. You get about 2x faster run time per WU. See my earlier post on where to get the files. I'm not well versed in linux but I'm trying to see if I can get this up and running for future SETI challenges, kind of invest for the future. ChristianVirtual also uses the optimized apps.
 
Damn.

Windows stuff appears to be from late 2016.

Dualboot time? :D

Edit: Let's see if i can get Windows working.

Edit2: Estimated Time really doesn't mean a whole lot does it...

Trying out:

lunatics_win64_v0.45_beta6-for-sog_setup
+
-sbs 384 -spike_fft_thresh 4096 -tune 1 64 1 4 -oclfft_tune_gr 256 -oclfft_tune_lr 16 -oclfft_tune_wg 256 -oclfft_tune_ls 512 -oclfft_tune_bn 64 -oclfft_tune_cw 64 -use_sleep


Edit3: On the Ryzen 1700 rig, WU times down to ~5 min per 1070 @1task/1core. Let's see how the 1070 Ti's do... with limited CPU. :x

Edit4: Oh. baby. On the G3930/1070 Ti rigs, I hear the fans spinning up :D 1 GPU/.333 CPU seems to be putting the cards to work. Taking big bites now. (~.5% per). Units are clearing in about ~6min. I'll take it!

Edit5: Well, if anyone had thoughts of a comeback: "lol"
 
Last edited:
Switching back over to the mines.

Good first run! .. Definitely need to get some stronger CPUs in these rigs. Also need to figure out if the x1 PCIe lane is hurting things. My numbers were actually very weak for the number of GPUs i was throwing at it all.
 
atp1916 said:
Switching back over to the mines.

Good first run! .. Definitely need to get some stronger CPUs in these rigs. Also need to figure out if the x1 PCIe lane is hurting things. My numbers were actually very weak for the number of GPUs i was throwing at it all.
Click to expand...

Thanks for pitching in! optimized or not you put up some kick ass numbers.
 
One of my 1070's was on PCIe 2.0 x4 and it was as fast as the 2.0 x16,
but i dont' know about 3.0 X1
 
Looking at BOINCstat website (for yesterday's tally), our daily SETI output puts us in the top 10 producers.

upload_2018-3-26_7-56-1.png
 
Tinkering with Ubuntu 16.04 and testing the GPU optimized application yesterday, my 1080 could give me about 50K PPD.

I know the sprint is over, just a good learning experience playing with linux.... Here is the link to my GPU rig.
In case anyone can't see it, below is a snapshot of some of the completed work units.

If atp1916 can get decent CPU with optimized GPU application, we would just be .......

upload_2018-3-26_8-0-50.png
 
You must log in or register to reply here.
Back
Top