Overclock.net banner

Cross Country Project


  • Total voters
    12
41 - 60 of 65 Posts

·
Premium Member
Joined
·
5,969 Posts
You're all going to be like TDP WHY DIDN'T YOU SAVE THAT SUCKA but honestly I couldn't repeat it if I tried. Running Einstein off and on since the 2017 Pentathlon but somehow it wasn't going toward credit for OCN or for me, ever, just stockpiling over time. I managed to fix it in January and it all dumped at once to the tune of 3,147,057 credit. Wish I would have held on to it for another 4 months! Such is the price of not knowing what you're doing with most of this stuff :-D

Anyway, hooray Einstein! My favorite.
You allowed stats export. The Pent pulls team stats which is irrelevant if individual users allow export or not.
 

·
BOINC Cruncher
Joined
·
1,812 Posts
Someone must be hogging up all the nvidia tasks. My 2 1070tis have been sitting mostly idle, only getting a single task at random intervals. I still have plenty of AMD tasks.
 

·
Premium Member
Joined
·
2,822 Posts
I have plenty now, but scrolling through the logs I can see periods when I didn't on my faster cards.


No tasks to send and workload generation seems to be off for FGRPB1G https://einsteinathome.org/server_status.php at the moment.
 

·
⤷ αC
Joined
·
11,240 Posts
Probably has more to do with this gen of Radeon GPUs mostly being poor sellers after mining craze (RX 470/RX 570/ RX 480 / RX 580 / Vega 56) vs the Nvidia counterparts (RTX series, GTX 1070, GTX 1080 + 1080 Ti) , combined with the typically lower efficiency of Radeon on Einstein in general except for the double precision portion. The 7nm Radeon is too new and expensive with no aftermarket cards even if the FP64 is higher and the amount sold is obviously less.

The notation on double precision can be found here: https://einsteinathome.org/content/updated-gpu-requirement-einstein

Granted VRAM is lower on the low end Nvidia GPUs such as the GTX 1060 3GB and GTX 1050 3GB/2GB not to mention the GTX 970 fiasco with "3.5GB" , but the bulk of Nvidia GPUs are 6GB+ these days (everything from Turing GTX 1660 and up). Turing is able to do concurrent int + float operations.

AFAIK you need only about 1.5GB of VRAM unless you're running more than 2 tasks concurrently.

Over at SETI there is a thread on the GTX 1660 (presumably the non-Ti GDDR5 version) and this is the result:
GTX 970 - 6.4 minutes/task
GTX 1060 - 7.3 minutes/task
GTX 1660 - 4.5 minutes/task
https://setiathome.berkeley.edu/forum_thread.php?id=83911&sort_style=&start=20

This year gridcoin team isn't present so that's good. I don't remember the Scottish team though?

Edit: see also https://setiathome.berkeley.edu/forum_thread.php?id=81962&sort_style=&start=220
GTX 1660 TI seems to be impressive perf/watt per post https://setiathome.berkeley.edu/forum_thread.php?id=81962&postid=1988037 and https://www.phoronix.com/scan.php?page=article&item=nvidia-gtx1660ti-opencl&num=3
 

·
Premium Member
Joined
·
2,822 Posts
All my cards, NVIDIA and AMD are seeing no work units available messages now. I have about an hour queued up on my faster systems.


@AlphaC not sure what you are trying to say. My RX-470 is faster in Einstein then my GTX-1070 and GTX-980ti, and I doubt it's using more energy than either.



The Scottish team looks like a temporary amalgamation of teams formed to build a super team and dominate the competition. Look at SUSA in the stats, a number of people from there changed teams for the event, and now instead of on top they are well down the stats.
 

·
⤷ αC
Joined
·
11,240 Posts
I'm just saying general OpenCL performance isn't as high as it could be , Einstein may be an outlier given that the OpenCL benches and [email protected] performance don't reflect that

If you look at the RX 590 position on "OpenCL Test: Global Memory Bandwidth" that's the only time it really matches Pascal besides single precision float. Granted Einstein uses memory bandwidth more.


edit: I looked at Einstein's database of top computers and some RX 570s are getting 1200s , RX 580s around 1000s, Vega around 600s. GTX 1080 ran around 500s-700s, GTX 1080 TI around 450s. No note about the concurrent WUs.

WUProp:
RX 470 is 20.7 min (11.6-266.5) in Windows
http://wuprop.boinc-af.org/results/gpu.py?fabricant=ATI&type=RX+400&modele=Radeon+RX+470&tri=projet&sort=asc


RX 570 is 23.9 min (21.6-27.4) in Linux and 20.1 min (10.7-119.0) in Windows
http://wuprop.boinc-af.org/results/gpu.py?fabricant=ATI&type=RX+500&modele=Radeon+RX+570&tri=projet&sort=asc

RX 480 is 9.8 min (8,9-23,5) in Linux and 13.2 min (3,2-204,3) in Windows
http://wuprop.boinc-af.org/results/gpu.py?fabricant=ATI&type=RX+400&modele=Radeon+RX+480&tri=projet&sort=asc

RX 580 is 17.1 min (8.1-61.3) in Linux and 18.0 min (5.8-168.0) in Windows
http://wuprop.boinc-af.org/results/gpu.py?fabricant=ATI&type=RX+500&modele=Radeon+RX+580&tri=projet&sort=asc


Vega 56 is 8.5 min (6.2-35.6) in Windows
http://wuprop.boinc-af.org/results/gpu.py?fabricant=ATI&type=RX+Vega&modele=Radeon+RX+Vega+56&tri=projet&sort=asc

Vega 64 is 9.2 min (2.1-103.2) in Windows
http://wuprop.boinc-af.org/results/gpu.py?fabricant=ATI&type=RX+Vega&modele=Radeon+RX+Vega+64&tri=projet&sort=asc

Radeon VII is 6.3 min (3.7-11.6) in Windows
http://wuprop.boinc-af.org/results/gpu.py?fabricant=ATI&type=Radeon+VII&modele=Radeon+VII&tri=projet&sort=asc


GTX 1060 is 20 min (1.3-74.6) in Linux and 26.6 min (3.9-94.1) in Windows
http://wuprop.boinc-af.org/results/gpu.py?fabricant=NVIDIA&type=GeForce+1000+Series&modele=GeForce+GTX+1060+6GB&tri=projet&sort=asc

GTX 1070 is 22.3 min (10.1-46.5) in Linux and 21.3 min (4.6-133.0) in Windows
http://wuprop.boinc-af.org/results/gpu.py?fabricant=NVIDIA&type=GeForce+1000+Series&modele=GeForce+GTX+1070&tri=projet&sort=asc

GTX 1070 TI is 19.6 min (11.7-30.1) in Linux and 23.1 min (11.3-81.9) in Windows
http://wuprop.boinc-af.org/results/gpu.py?fabricant=NVIDIA&type=GeForce+1000+Series&modele=GeForce+GTX+1070+Ti&tri=projet&sort=asc

GTX 1080 GDDR5X is 11.4 min (10.8-28.4) in Linux and 23.2 min (5.0-401.0) in Windows
http://wuprop.boinc-af.org/results/gpu.py?fabricant=NVIDIA&type=GeForce+1000+Series&modele=GeForce+GTX+1080&tri=projet&sort=asc

GTX 1080 Ti GDDR5X is 8.6 min (7,3-23,4) in Linux and 18,0 (6,6-63,3)in Windows
http://wuprop.boinc-af.org/results/gpu.py?fabricant=NVIDIA&type=GeForce+1000+Series&modele=GeForce+GTX+1080+Ti&tri=projet&sort=asc


GTX 1660 TI GDDR6 is 15.2 min in Linux and 30.5 min (15.3-69.2) in Windows
http://wuprop.boinc-af.org/results/gpu.py?fabricant=NVIDIA&type=GeForce+1600+Series&modele=GeForce+GTX+1660+Ti&tri=projet&sort=asc

RTX 2070 GDDR6 is 22.3 min (10.8-57.5) in Windows
http://wuprop.boinc-af.org/results/gpu.py?fabricant=NVIDIA&type=GeForce+2000+Series&modele=GeForce+RTX+2070&tri=projet&sort=asc

RTX 2080 GDDR6 is 11,3 (8,1-18,8) in Linux and 11.3 min (8.1-44,5) in Windows
http://wuprop.boinc-af.org/results/gpu.py?fabricant=NVIDIA&type=GeForce+2000+Series&modele=GeForce+RTX+2080&tri=projet&sort=asc
 

·
Premium Member
Joined
·
5,969 Posts
Work generators do not run all the time as there is no need usually. But now there are no FGRPB1G tasks available atm.

A proper SETI setup should be CUDA in Linux. The optimized Windows app is much slower. No comparison to [email protected] at all.

Those 570s and 580s running at 1k-1.2k are 2x tasks. Vega is probably 2x as well at 600s. A 570 is just as fast as a 580 in [email protected] Its all about memory speed. A card with a Mining BIOS gains a nice boost. I saw no change going from 1430 to 1366 core clock. But OCing the memory does wonders.

My own at 1x
GTX 1070: 12:42
GTX 1070Ti: 11:22
GTX 1080Ti: 7:29
RX 580: 10:14
 

·
Watercooled Railgun
Joined
·
4,630 Posts

·
Registered
Joined
·
113 Posts
My 1070 at 2x tasks at a time (1CPU + 0.5Nvidia GPU, two at a time) is taking ~43 minutes per Gamma-ray pulsar binary search #1. Is that as expected?
 

·
Premium Member
Joined
·
2,822 Posts
My 1070 at 2x tasks at a time (1CPU + 0.5Nvidia GPU, two at a time) is taking ~43 minutes per Gamma-ray pulsar binary search #1. Is that as expected?

Seems slow. I am running two tasks on a 1070 and averaging about 1640 seconds per task, so 27:33. Win 10 Pro, NVIDIA 388.13. I have nothing but BOINC running on that system at this time, so those tasks are getting two full threads.
 

·
BOINC Cruncher
Joined
·
1,812 Posts
My 1070 at 2x tasks at a time (1CPU + 0.5Nvidia GPU, two at a time) is taking ~43 minutes per Gamma-ray pulsar binary search #1. Is that as expected?

That's really slow. My 1070 ti's in a headless Linux machine can do 2 tasks in a little over 24 minutes and that is probably slow.
 

·
Premium Member
Joined
·
5,969 Posts
My 1070 at 2x tasks at a time (1CPU + 0.5Nvidia GPU, two at a time) is taking ~43 minutes per Gamma-ray pulsar binary search #1. Is that as expected?
About half normal speed. See my times above in Linux for my NV cards. Are you really running 2x or 4x? Memory OC benefits most.
 

·
Not new to Overclock.net
Joined
·
5,109 Posts
Is there really that much benefit to running two tasks at a time? It seems to be a waste of CPU resources...
My 1080s at stock clocks run a single task each in 11.5 minutes
 

·
Premium Member
Joined
·
5,969 Posts
Is there really that much benefit to running two tasks at a time? It seems to be a waste of CPU resources...
My 1080s at stock clocks run a single task each in 11.5 minutes
My NV cards are slower in Linux at 2x so I'm running 1x. My RX card in Windows gains a bit with 2x.
 

·
Registered
Joined
·
113 Posts
Basically the same story, nothing much else running. 388.43 instead of 388.13 though. Oh well.
 

·
Registered
Joined
·
113 Posts
Any clues here? 100% core load, mild overclocks, but GPU power seems underutilized. This card usually draws about 150 watts when it's hitting its stride and now it's only drawing ~110 watts.
 

Attachments

·
Premium Member
Joined
·
5,969 Posts
Any clues here? 100% core load, mild overclocks, but GPU power seems underutilized. This card usually draws about 150 watts when it's hitting its stride and now it's only drawing ~110 watts.
Are your CPUs 100% used? If so use Process Lasso to permanently push all other exes to like threads 0-6 and the GPU exe to thread 7. My performance on a RX drops a lot without that.
 
41 - 60 of 65 Posts
Top