NVIDIA GTX 1660 Supra advice

Bill F
Bill F
Joined: 24 Dec 05
Posts: 50
Credit: 66,173,262
RAC: 131,561
Topic 230152

I am looking for advice on GPU optimization settings for a new system with a NVIDIA card for the Einstien app's

GenuineInte Intel(R) Xeon(R) CPU E5620 @ 2.40GHz [Family 6 Model 44 Stepping 2] (4 processors)  32 GB NVIDIA NVIDIA GeForce GTX 1060 6GB (6143MB) driver: 536.67

Would someone with experience on this GPU like to share some pointers with me ?

 

Thanks

Bill F

 

 

In October of 1969 I took an oath to support and defend the Constitution of the United States against all enemies, foreign and domestic;
There was no expiration date.

Keith Myers
Keith Myers
Joined: 11 Feb 11
Posts: 4,720
Credit: 17,645,773,094
RAC: 6,356,056

With only 6GB of memory you

With only 6GB of memory you are limited to only 1X task per app because the O3AS GW tasks often exceed 4GB of VRAM usage.

But you can run run 2X multiples of BRP7 tasks since they have a very low VRAM usage.  But in reality the BRP7 tasks run best at 1X for the most part.

I'd stick to 1X tasks for both O3AS GW and BRP7 tasks.

 

Bill F
Bill F
Joined: 24 Dec 05
Posts: 50
Credit: 66,173,262
RAC: 131,561

Keith Thank you for the

Keith

Thank you for the response you mentioned BRP7 twice and GW, But not FGRP what is the memory requirements like on those ? Should I stay to X1 on those as well ?

Thanks

Bill F

 

 

In October of 1969 I took an oath to support and defend the Constitution of the United States against all enemies, foreign and domestic;
There was no expiration date.

Gary Roberts
Gary Roberts
Moderator
Joined: 9 Feb 05
Posts: 5,844
Credit: 109,749,951,642
RAC: 28,294,476

Bill F wrote:... But not FGRP

Bill F wrote:
... But not FGRP what is the memory requirements like on those ?

FGRP GPU tasks are finished.  Please read the Tech News announcement about this.  It was posted a month ago.

Cheers,
Gary.

Bill F
Bill F
Joined: 24 Dec 05
Posts: 50
Credit: 66,173,262
RAC: 131,561

Keith With the reduced GW

Keith

With the reduced GW requirements will a 6GB card now support two tasks with a .5 setting ?

 

Thanks

Bill F

In October of 1969 I took an oath to support and defend the Constitution of the United States against all enemies, foreign and domestic;
There was no expiration date.

Keith Myers
Keith Myers
Joined: 11 Feb 11
Posts: 4,720
Credit: 17,645,773,094
RAC: 6,356,056

The latest news posted on the

The latest news posted on the front page of the project says the GW O3AS tasks use much less VRAM now.

So you should be able to run 2X on your 6GB I think. Try it and see if that is so.

There are people running 3X on 8GB cards so two should fit on 6GB.

 

Gary Roberts
Gary Roberts
Moderator
Joined: 9 Feb 05
Posts: 5,844
Credit: 109,749,951,642
RAC: 28,294,476

I'm running 2 tasks

I'm running 2 tasks concurrently on 4GB cards with no issues.  For best performance, you need to stagger the start time of the 2nd task so that the GPU is as much in constant use as possible.  There is significant CPU only use when a task reaches 49.5% and again at 99.5%.  If both tasks reach these milestones simultaneously, the GPU will be completely idle during those periods.

An easy way to get an appropriate staggering is to temporarily suspend all other tasks and allow the first task to reach ~40% on its own.  Then resume the remainder and a 2nd task will launch and go through ~1 min of startup and then be GPU only.  Meanwhile, the first task continues on to 49.5% (the CPU only stage) as the 2nd task takes off on the GPU.  Unless you  have a very slow CPU, the first task will finish the CPU only stage before the 2nd task gets there.  This will tend to rinse and repeat automatically at 99.5%.

Depending on task to task variations, the staggering will drift over time.  About once a day or so, check and readjust.  An easy way to do that is to prevent the start of the replacement task for whichever task is 'ahead of the curve'.  Once you know the 'best completion time possible',  all tasks should complete reasonably close to that value.  Divide that completion time by 4.  The number you get will be the optimum task1 to task2 separation.  For when a task1 has completed and a replacement has restarted, the task2 will be up around the 90% mark and the gap between the replacement task1 and the old task2 will be ~3 times the one-quarter gap described previously.  Don't be fooled by that.  Think about it and you should understand why that is.

One thing that really helps is to make sure the CPU isn't operating in powersave mode.  I see further improvement by setting the CPU in performance mode.  I use Linux and there is a utility called cpupower which can set the required mode very easily.  You might think that this would only help for the CPU only portions of the run.  However, I've noticed that the GPU portions run slower if the CPU is in powersave mode.  One of my CPUs has a 800/4000 frequency range.  The GPU portion of a task slows noticeably if the frequency is allowed to drop to 800.

EDIT:  Please note that the above is specifically for running 2 O3AS tasks concurrently on a 4GB or 6GB GPU.  For anyone running more than two, the staggering scheme described above would NOT be appropriate.

Cheers,
Gary.

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.