Right now I am thinking between getting Titan V & water cooled RTX 3070...same money!
What do you guys (& girls, if there is some here)?
Water cooling could fail more easily than Air Cooled. And probably is harder to fix. So my bias has pretty much been Air Cooled.
If you can squeeze out the money I would go for an Rtx 3080 ti instead of an Rtx 3070.
The Titan V is especially good on the GpuGrid project because of its higher memory bandwidth.
The Titan V costs less electricity to run than the Rtx 3080 ti for the same level of production.
Tom M
Then Titan V it is...as I prefer to use "fan based GPUs", because I need to extract air outside of the case (opposed to keeping GPU heat within case on normal cooling of GPU).
I do know that about O3, so it is better & fruitful to do the BRPs...it is 1/3 of points, but on my system it gets done 4x times faster then single O3.
How many tasks are you running on your Gpu's at a time?
I would certainly experiment. Depending on the GPU ram size as well as the speed of the card you may be able to gain total production.
The O3 might gain production at up to 5x tasks for a 12MB card Don't run more than 90% memory.
The Brp7 might gain production at up to 3x tasks. It doesn't seem to gain anything past 3x unless you have a lot faster GPU than I do.
I am assuming you know where the Boinc folder is under the hidden Program-Data folder of Windows.
Tom M
Currently I run my system with Quadro RTX 4000 & Tesla P4, doing 2x O3 tasks. Which averages around ~1h per task, but the BRPs finish within ~15min per task for 1/3 of the points!
As Quadro RTX 4000 does need a help with cooling for doing 2nd task (with Tthrottling down to 75~80% of throttling), so this is also why I want to switch to some card with dual slot layout! Triple slot layout does not work for me, as I need it for Tesla P4.
Klik: I also noticed you are only running windows systems. If you plan on a new gpu, I would really suggest setting up a linux system. If you have never used linux, just let us know- people here will be happy to help. It's not difficult at all and very user friendly.
If you want to stick with windows, then O3 work is not ideal since it is only opencl on windows. You can get more than 3x production with a linux system (same gpu) since it has a cuda version of the application. You can see this in our devices, if interested (they are running rtx a4500 gpus)
If you want to run MeerKAT, then there is a windows cuda application available.
Working on Linux has its advantages, I must say. But also, there is (not that I know of) any kind of BOINC throttling to be installed & worked for Linux. So any card which I was putting in my Linux machines (previously on SETi@home), were either low, powered ones or even the air cooled without fans!
On Windows, there is a FREEware Tthrottling app, which works great on all CPUs & GPUs. I have been using it since I fried the fanned GTX 8500 & air-cooled 240 GT in my systems some time ago. Since then, there were no GPU failures in my systems!
Also, as I do use Quadro RTX A4000 on my work computer, so not a fan of single slot GPUs. It is setup on 75°C on Tthrottle, so it runs on 15~20% only! Which is not great & similar thing can be said for my home Quadro RTX 4000, which runs 75~80% on dual O3 data.
So switching back to dual-slotted GPU is a must, as I used to run Quadro M5000 on it which worked great. Hence the questions for Titan V!
with linux, if you want to throttle the card, the best and easiest thing to do is to just power limit the card until it runs under the power and temp limits you want. power limiting has the secondary effect of reducing temperatures since less power used = less heat generated.
above from a startup script (preferences --> startup applications)
Yes, I do know that...but in Linux you have to write your own script, which is sthg I have never done. & it only limits power to "good card".
While in Win environment with Tthrottle you have FREEware app, which monitors the GPU temp. & even if card goes bad or the TIM paste/pad gets bad (over time), it will throttle down the card. I've had a card with single fan going bad & the card survived with Tthrottle, as it was not limited by power - but with temps!
Tom M wrote: KLiK
)
Then Titan V it is...as I prefer to use "fan based GPUs", because I need to extract air outside of the case (opposed to keeping GPU heat within case on normal cooling of GPU).
Thanks for this suggestion!
non-profit org. Play4Life in Zagreb, Croatia, EU
Tom M wrote: KLiK wrote: I
)
Currently I run my system with Quadro RTX 4000 & Tesla P4, doing 2x O3 tasks. Which averages around ~1h per task, but the BRPs finish within ~15min per task for 1/3 of the points!
As Quadro RTX 4000 does need a help with cooling for doing 2nd task (with Tthrottling down to 75~80% of throttling), so this is also why I want to switch to some card with dual slot layout! Triple slot layout does not work for me, as I need it for Tesla P4.
I do know where the hidden folder is...
non-profit org. Play4Life in Zagreb, Croatia, EU
Boca Raton Community HS
)
Working on Linux has its advantages, I must say. But also, there is (not that I know of) any kind of BOINC throttling to be installed & worked for Linux. So any card which I was putting in my Linux machines (previously on SETi@home), were either low, powered ones or even the air cooled without fans!
On Windows, there is a FREEware Tthrottling app, which works great on all CPUs & GPUs. I have been using it since I fried the fanned GTX 8500 & air-cooled 240 GT in my systems some time ago. Since then, there were no GPU failures in my systems!
Also, as I do use Quadro RTX A4000 on my work computer, so not a fan of single slot GPUs. It is setup on 75°C on Tthrottle, so it runs on 15~20% only! Which is not great & similar thing can be said for my home Quadro RTX 4000, which runs 75~80% on dual O3 data.
So switching back to dual-slotted GPU is a must, as I used to run Quadro M5000 on it which worked great. Hence the questions for Titan V!
non-profit org. Play4Life in Zagreb, Croatia, EU
with linux, if you want to
)
with linux, if you want to throttle the card, the best and easiest thing to do is to just power limit the card until it runs under the power and temp limits you want. power limiting has the secondary effect of reducing temperatures since less power used = less heat generated.
_________________________________________________________________________
#----------------------------
)
above from a startup script (preferences --> startup applications)
Skip Da Shu
)
Yes, I do know that...but in Linux you have to write your own script, which is sthg I have never done. & it only limits power to "good card".
While in Win environment with Tthrottle you have FREEware app, which monitors the GPU temp. & even if card goes bad or the TIM paste/pad gets bad (over time), it will throttle down the card. I've had a card with single fan going bad & the card survived with Tthrottle, as it was not limited by power - but with temps!
non-profit org. Play4Life in Zagreb, Croatia, EU
Well, all I can say (as a
)
Well, all I can say (as a Linux noob) thats Windows doing its job ...
cheers
sfv
... and to limit temperature
)
... and to limit temperature to what you want ...
nvidia-smi -i 0 -gtt 76
--
petri33
So with 4 GPUs and if I want
)
So with 4 GPUs and if I want a 70c limit, it should be:?
nvidia-smi -i 0 -gtt 70
nvidia-smi -i 1 -gtt 70
nvidia-smi -i 2 -gtt 70
nvidia-smi -i 3 -gtt 70
You could probably omit the
)
You could probably omit the -i flag and [num] parameter and it will apply the same value to all cards. Works that way for power limiting.
_________________________________________________________________________