Well that's consistent with what I found.
I misinterpreted what you said. I've seen absolutely no improvement in processing times for the new 1.57 app versus the old 1.52 app.
I thought you were saying you were getting the same results.
I still think you should try running the BRP6 1.57 apps solo and see what you get.
Not interested. SETI is my primary project. MW and Einstein are just my backup projects when SETI goes kablooey. I'd just can't reconcile the referenced speedup in both the thread I mentioned and here in this thread. I looked for people doing similar concurrent tasks and projects on a card to try and compare apples to apples. I'm sure I would see a shorter computation time if I ran a task solo. Same would apply for a SETI or MW task. I just choose to run 3 projects. The majority of the time, any card is doing only SETI and only briefly is any one MW or Einstein task running on any system. That is only one Einstein task running at a time on any of my cards.
That is only one Einstein task running at a time on any of my cards.
Well that answers the question you do run Einstein solo when it runs. It still does't answer in my mind why your "more powerful" card produces less per hr then my GTX660.
No, I guess I didn't state the case clearly. I have two 970's. Both set to run 3X or 3 tasks per card at all times. When a MW or Einstein task runs, it is running concurrent with two SETI tasks. Project resource share is low enough for both MW and Einstein that there never is more than one MW or Einstein task running at any time on either system. With regard to your higher production with the 660, my resource share is likely much lower than yours. I give 80% of time to SETI and only 10% to MW and Einstein. Project rotation is every 70 minutes because with the old 1.52 app, they would likely complete in 2 cycles of rotation. These 1.57 tasks are taking at least 3 cycles of rotation, thus my lower throughput.
When a MW or Einstein task runs, it is running concurrent with two SETI tasks.
Ah, you are not running solo and that is consistent with my finding that BRP tasks don't play nicely with others. I suspect many apps from other projects would have this same "problem".
As I understand it resource share does not affect run time but rather how much of the time it runs. I certainly could be wrong on that one though.
As I understand it resource share does not affect run time but rather how much of the time it runs. I certainly could be wrong on that one though.
Correct. Resource share only affects amount of time a system devotes to processing of any project. In my experience, MW is worse. Einstein is fairly benign on my systems as far as slowing down SETI completion times. MW on the other hand drastically lengthens the SETI processing time. It uses more GPU and PCIe bus resources than any of my projects.
I know why. The FP32 and FP64 performance is better on the old 660 vs the 900 series. 1/24 FP32 vs 1/32 FP32. Look at this table. There are a lot of good things to say for the older designs with regard to math performance.
How many concurrent 1.57 tasks per 960 do you run Tom? Any other projects on the cards when Einstein is processing? I'd sure like to see that kind of improvement on my systems from 1.52 > 1.57. As long as we're comparing apples to apples.
Well that's consistent with
)
Well that's consistent with what I found.
I misinterpreted what you said.
I've seen absolutely no improvement in processing times for the new 1.57 app versus the old 1.52 app.I thought you were saying you were getting the same results.
I still think you should try running the BRP6 1.57 apps solo and see what you get.
Not interested. SETI is my
)
Not interested. SETI is my primary project. MW and Einstein are just my backup projects when SETI goes kablooey. I'd just can't reconcile the referenced speedup in both the thread I mentioned and here in this thread. I looked for people doing similar concurrent tasks and projects on a card to try and compare apples to apples. I'm sure I would see a shorter computation time if I ran a task solo. Same would apply for a SETI or MW task. I just choose to run 3 projects. The majority of the time, any card is doing only SETI and only briefly is any one MW or Einstein task running on any system. That is only one Einstein task running at a time on any of my cards.
RE: That is only one
)
Well that answers the question you do run Einstein solo when it runs. It still does't answer in my mind why your "more powerful" card produces less per hr then my GTX660.
No, I guess I didn't state
)
No, I guess I didn't state the case clearly. I have two 970's. Both set to run 3X or 3 tasks per card at all times. When a MW or Einstein task runs, it is running concurrent with two SETI tasks. Project resource share is low enough for both MW and Einstein that there never is more than one MW or Einstein task running at any time on either system. With regard to your higher production with the 660, my resource share is likely much lower than yours. I give 80% of time to SETI and only 10% to MW and Einstein. Project rotation is every 70 minutes because with the old 1.52 app, they would likely complete in 2 cycles of rotation. These 1.57 tasks are taking at least 3 cycles of rotation, thus my lower throughput.
RE: When a MW or Einstein
)
Ah, you are not running solo and that is consistent with my finding that BRP tasks don't play nicely with others. I suspect many apps from other projects would have this same "problem".
As I understand it resource share does not affect run time but rather how much of the time it runs. I certainly could be wrong on that one though.
RE: As I understand it
)
Correct. Resource share only affects amount of time a system devotes to processing of any project. In my experience, MW is worse. Einstein is fairly benign on my systems as far as slowing down SETI completion times. MW on the other hand drastically lengthens the SETI processing time. It uses more GPU and PCIe bus resources than any of my projects.
GTX660 1.52 times around
)
GTX660 1.52 times around 12,800 x2 i5 4.00 Ghz
GTX960 1.52 times around 16,800 x2 i5 4.00 GHz
GTX660 1.57 times around 10,500 x2 same system
GTX960 1.57 times around 12,700 x2 same system
Both NVIDIA in same system
Both runs with 100% CPU running x4 ATLAS
So Thank you Bernd and gang.
Not sure why GTX960 is slower than GTX660 it does use much less power
I know why. The FP32 and
)
I know why. The FP32 and FP64 performance is better on the old 660 vs the 900 series. 1/24 FP32 vs 1/32 FP32. Look at this table. There are a lot of good things to say for the older designs with regard to math performance.
How many concurrent 1.57 tasks per 960 do you run Tom? Any other projects on the cards when Einstein is processing? I'd sure like to see that kind of improvement on my systems from 1.52 > 1.57. As long as we're comparing apples to apples.
I run two GPU tasks at a time
)
I run two GPU tasks at a time on both GPU's in the system as well as
four ATLAS tasks (which are CPU only) so total of 8 tasks at once.
Bottom line - I get the same thruput on the
GTX960 with 1.57 as I got with the GTX660 on 1.52.
Tom you and Keith are
)
Tom you and Keith are confirming my case that BRPs do not play nicely with others.