Crunching on GTX 750 Ti
16 Feb 2015 1:09:15 UTC
Topic 197972
(moderation:
Just upgraded to an EVGA GTX 750 Ti FTW from an Intel HD 4000. I'm excited about the increase in output and just want to confirm the numbers I'm seeing.
For one task at a time:
GPU Usage: ~90%
MCU Usage: ~70%
GPU Boost: 1320 MHz
GPU Memory: 2700 MHz
The card boosted to 1320 MHz GPU clock and I'm thinking about pushing it further but don't know how much benefit it provides. The difference between stock clocks and max boost with my 780 Ti cards is a minute or two at best on Arecibo WUs. Maybe boosting the Memory Clock would help as well? Also, with the GPU and MCU usage already so high, will there be much benefit in running multiple WUs at time?
Appreciate any advice anyone is willing to send my way.
Crunching on GTX 750 Ti
)
I run one GTX 750, and one GTX 750 Ti. I run only Perseus Arm Survey on both of them, 2 tasks at a time. I don't currently recall the 1X to 2X productivity improvement I saw, but think it was somewhat less than I have seen on some other GPUs, but still plenty to make me want to do it. Starting from your current configuration--that change may get you more improvement than overclocking will.
I've not explored any form of overclocking on either of my 750s, though my 970 did respond pretty favorably to raising the GPU memory clock, and somewhat less appreciably to raising the primary GPU clock.
Both of my hosts also handle a dissimilar older GPU--a GTX 660. I unload the host CPU substantially by curtailing BOINC CPU work by more than the "1 core" popular here, and I use Process Lasso to tamper with default affinities and priorities with a view to reducing the latency the GPU tasks experience when they are waiting for CPU support task service.
I like both of 750s a lot--quite a bit of output at rather low power, an easy physical fit in my boxes (the short length reduces conflicts with cables, drives, and such) and quite moderate price. The base 750 model, in particular, can be had a quite attractively low price.
Thanks. I agree with you
)
Thanks. I agree with you about the price/productivity ratio. I was able to find a good deal on the model I picked up.
I should have mentioned in my first post that I'm currently testing on Arecibo WUs only and will branch out in the near future. The Perseus Arm tasks would have substantially lower GPU Usage, I believe.
I am also currently running on 75% CPU Usage. I've seen a few posts about the potential of Process Lasso and running GPU tasks in "real time" recently but haven't checked it out for myself yet.
You may want to opt in for
)
You may want to opt in for the beta. For BRP6 the results seem to be very nice with app 1.50. At 70% memory controller load an overclock will definitely help. Although the cards typically can't go much higher than 3000 MHz, so proceed with care. A core OC should also help, definitely on the beta tasks (because they released some other "performance break"). And at 90% GPU load I'd go for 2 concurrent WUs. Even if you "just" reach about 95%, that's still a 5% boost for little additional power. With the BRP6 beta the additional CPU load is far lower as well :)
And you could run the HD4000 along your GPU. This requires quite some main memory bandwidth for the iGPU and hence slows other tasks down a bit. The slow-down is too much for high-performance cards, but could be alright for a GTX750Ti. And once the changes from the beta find its way into the Intel version, the negative impact on other tasks should be significantly reduced. I don't know when this will happen, though. Bernd has done some excellent work recently, but is quite busy.
MrS
Scanning for our furry friends since Jan 2002