Does anyone with a 660ti or 7970 have any data on power consumption (both idle consumption and when running BRP4, not the system requirements or TDP values)?
I recently upgraded to the 7000 series cards from my previous GTX 550Ti cards. I have two 7970's and the power draw is between 160 - 176 watts (measured at the wall with a kill-a-watt) running two jobs at a time with 92% GPU utilization.
Here is a nice plot that helps visualize how good the 7970 is. For the total BRP4 production I used the best numbers reported in the sticky OpenCL / Cuda thread. I measured power for the 430, 550Ti, 7870 and the 7970. Power for the 660Ti is Magic's number and the 650 power came from a post by Gary Roberts detailing his budget build with this card.
I have not achieved the same production with my 7970 as others report, looks like I still need to get my X79 board to run PCIe x16 at version 3 which will probably increase power just a bit.
As a side note for the many 550Ti's out there, the 7870 makes an excellent upgrade power wise. For an extra 7 watts the 7870 doubles production over the 550Ti.
That sounds about right - and 35-40 minutes to complete the two tasks?
Alec
Hi Alec,
My average time is between 42 and 44 minutes running the 2 tasks.
That card probably runs faster with a newer processor and the one I use this 660Ti SC is my older Athlon II X4 635 that is also running the 2-core T4T tasks at the same time and once in a while I will add a Gamma Ray or a GWP task too.
Hallo Robert!
That´s an important information for me. Up to now I did think, that CUDA for NVIDIA is a much more effective programm coding than OpenCL for the AMD cards.
From (Server Status Page)/(GPU Productivity within last week) one can calculate the mean number of taks per GPU. This number is always much higher for NVIDIA than for AMD, at now 7,4/4,7 and averaged over the last 1/4 of a year I get 9,1/5,6. But the reason for this difference might have other reasons, for example a different age of the fleets of cards or different usage etc..
It ssems to me, that the long time knowledge: NVIDIA is much more capable than AMD for E@H, became a no longer true prejudice.
So the programmers (Bikeman and coleagues) did great progress in programming OpenCL!
Let´s see, what the new generation of cards from NVIDIA comming this springtime, will show.
Thanks for this info.
Kind regards and happy crunching.
Martin
So the programmers (Bikeman and coleagues) did great progress in programming OpenCL!
Thanks :-) But to be fair, the code is almost identical for CUDA and OpenCL (the FFT libs are different but that should not make a very great difference). So the credit for the good performance of the latest generation of AMD cards should go to AMD alone ;-)
The poorer statistics for the AMD/ATI card population at E@H is probably also biased by many AMD APUs where the GPU part isn't that fast.
Hallo Bikeman!
Thanks for clearifying this. All this are importand informations for the next decision to buy hardware. - And better understanding, what´s going on in the project. -
Hi, Robert, how did you meassured the power draw?
I mean, I guess its not the value of the whole system, so is that numbers the difference in the Kill-a-Watt values with the card working vs. idle or working vs. the system without the GPUs?
System: i7-3770, Nvidia GTX 660, 4 hard disks. Measured at the wall.
CPU & GPU idle = 70W
8x CPU tasks, GPU idle = 138W
8x CPU tasks, 1x GPU task = 219W
8x CPU tasks, 2x GPU tasks = 226W
no data for idle CPU & active GPU
Power Draw with 660ti, 7970?
)
I really don't check mine often since it is easier for me if I don't think about that until the light bill shows up every 2 months
But I have a GeForce 660Ti superclocked and at idle it is around 112 watts and running cuda X2 on my quad is about 260 watts @ 65C
Hi Magic, That sounds
)
Hi Magic,
That sounds about right - and 35-40 minutes to complete the two tasks?
Alec
I recently upgraded to the
)
I recently upgraded to the 7000 series cards from my previous GTX 550Ti cards. I have two 7970's and the power draw is between 160 - 176 watts (measured at the wall with a kill-a-watt) running two jobs at a time with 92% GPU utilization.
Here is a nice plot that helps visualize how good the 7970 is. For the total BRP4 production I used the best numbers reported in the sticky OpenCL / Cuda thread. I measured power for the 430, 550Ti, 7870 and the 7970. Power for the 660Ti is Magic's number and the 650 power came from a post by Gary Roberts detailing his budget build with this card.
I have not achieved the same production with my 7970 as others report, looks like I still need to get my X79 board to run PCIe x16 at version 3 which will probably increase power just a bit.
As a side note for the many 550Ti's out there, the 7870 makes an excellent upgrade power wise. For an extra 7 watts the 7870 doubles production over the 550Ti.
RE: Hi Magic, That sounds
)
Hi Alec,
My average time is between 42 and 44 minutes running the 2 tasks.
That card probably runs faster with a newer processor and the one I use this 660Ti SC is my older Athlon II X4 635 that is also running the 2-core T4T tasks at the same time and once in a while I will add a Gamma Ray or a GWP task too.
-Magic
Hallo Robert! That´s an
)
Hallo Robert!
That´s an important information for me. Up to now I did think, that CUDA for NVIDIA is a much more effective programm coding than OpenCL for the AMD cards.
From (Server Status Page)/(GPU Productivity within last week) one can calculate the mean number of taks per GPU. This number is always much higher for NVIDIA than for AMD, at now 7,4/4,7 and averaged over the last 1/4 of a year I get 9,1/5,6. But the reason for this difference might have other reasons, for example a different age of the fleets of cards or different usage etc..
It ssems to me, that the long time knowledge: NVIDIA is much more capable than AMD for E@H, became a no longer true prejudice.
So the programmers (Bikeman and coleagues) did great progress in programming OpenCL!
Let´s see, what the new generation of cards from NVIDIA comming this springtime, will show.
Thanks for this info.
Kind regards and happy crunching.
Martin
RE: So the programmers
)
Thanks :-) But to be fair, the code is almost identical for CUDA and OpenCL (the FFT libs are different but that should not make a very great difference). So the credit for the good performance of the latest generation of AMD cards should go to AMD alone ;-)
The poorer statistics for the AMD/ATI card population at E@H is probably also biased by many AMD APUs where the GPU part isn't that fast.
Cheers
HB
Hallo Bikeman! Thanks for
)
Hallo Bikeman!
Thanks for clearifying this. All this are importand informations for the next decision to buy hardware. - And better understanding, what´s going on in the project. -
Kind regards and happy crunching.
Martin
It will be interesting to
)
It will be interesting to see, where will stand the new AMD HD 7790 on the chart. Looking at:
http://en.wikipedia.org/wiki/Comparison_of_AMD_graphics_processing_units#Southern_Islands_.28HD_7xxx.29_series,
it have the best GFLOPS/W (Single-precision) theoretical ratio among all current desktop GPUs from AMD and Nvidia.
Hi, Robert, how did you
)
Hi, Robert, how did you meassured the power draw?
I mean, I guess its not the value of the whole system, so is that numbers the difference in the Kill-a-Watt values with the card working vs. idle or working vs. the system without the GPUs?
System: i7-3770, Nvidia GTX
)
System: i7-3770, Nvidia GTX 660, 4 hard disks. Measured at the wall.
CPU & GPU idle = 70W
8x CPU tasks, GPU idle = 138W
8x CPU tasks, 1x GPU task = 219W
8x CPU tasks, 2x GPU tasks = 226W
no data for idle CPU & active GPU