It seems that a lot of GPU problems revolve around specific versions of drivers. Though NVidia has their own support structure, you can often learn from information reported by others who fold.
The obvious is that the automatic detection has failed and switched the cards. You can manually correct it by assigning opencl-index values as extra slot options to the two cards. (-1) means automatic, while 0 is the first device and 1 is the second. Of course you do not know which card is the first and which is the second so if you do not succeed the first time switch the values between the cards.
I recently had the same thing happen with a GK106 [GTX 650 Ti] and a GM107 [GTX 750 Ti]. The "slow" card was faster than the "fast" card. I didn't see any point in convincing FAH to interchange the names to make it look right, since they were both getting the same assignments.
Hardware configuration: Intel i7-4770K @ 4.5 GHz, 16 GB DDR3-2133 Corsair Vengence (black/red), EVGA GTX 760 @ 1200 MHz, on an Asus Maximus VI Hero MB (black/red), in a blacked out Antec P280 Tower, with a Xigmatek Night Hawk (black) HSF, Seasonic 760w Platinum (black case, sleeves, wires), 4 SilenX 120mm Case fans with silicon fan gaskets and silicon mounts (all black), a 512GB Samsung SSD (black), and a 2TB Black Western Digital HD (silver/black).
Without a log we have little to go on other than guessing. Please copy & paste the folding log here, including the system and config portions (Click refresh first). The log will allow us to see what is actually going on.
There are lots of potential causes:
Make sure you have a full CPU core dedicated to the 780Ti.
Make sure you are giving it enough time to make an accurate estimate. Generally 3%-5% is the minimum before it gets reasonably accurate.
Check the Windows system event logs to make sure you haven't gotten a video card reset that forced your clock rates very low.
Check the folding log and compare it to the client to make sure that it matches the number of frames done.
Run the client on full 24x7 (no running on idle) for you won't get the maximum PPD that your card is capable of if it is not running all the time
P5-133XL wrote:Without a log we have little to go on other than guessing. Please copy & paste the folding log here, including the system and config portions (Click refresh first). The log will allow us to see what is actually going on.
There are lots of potential causes:
Make sure you have a full CPU core dedicated to the 780Ti.
Make sure you are giving it enough time to make an accurate estimate. Generally 3%-5% is the minimum before it gets reasonably accurate.
Check the Windows system event logs to make sure you haven't gotten a video card reset that forced your clock rates very low.
Check the folding log and compare it to the client to make sure that it matches the number of frames done.
Run the client on full 24x7 (no running on idle) for you won't get the maximum PPD that your card is capable of if it is not running all the time
Thanks for the quick answer!
Make sure you have a full CPU core dedicated to the 780Ti. - by adding a CPU slot my PPD increased - thanks
Make sure you are giving it enough time to make an accurate estimate. Generally 3%-5% is the minimum before it gets reasonably accurate. - waited till 15%
Check the Windows system event logs to make sure you haven't gotten a video card reset that forced your clock rates very low. - GPU is under perfect load (90% in GPU-Z)
Check the folding log and compare it to the client to make sure that it matches the number of frames done. - matches
Run the client on full 24x7 (no running on idle) for you won't get the maximum PPD that your card is capable of if it is not running all the time
[b] - is there a bonus for 24/7?[/b]
As mentioned above my PPD increased (to ~60k, Project 9201) by adding the CPU slot. It still seems a little low, shoudn't it be around 200K?
Hardware configuration: Intel i7-4770K @ 4.5 GHz, 16 GB DDR3-2133 Corsair Vengence (black/red), EVGA GTX 760 @ 1200 MHz, on an Asus Maximus VI Hero MB (black/red), in a blacked out Antec P280 Tower, with a Xigmatek Night Hawk (black) HSF, Seasonic 760w Platinum (black case, sleeves, wires), 4 SilenX 120mm Case fans with silicon fan gaskets and silicon mounts (all black), a 512GB Samsung SSD (black), and a 2TB Black Western Digital HD (silver/black).
Yes, there is a bonus. The faster the WU is returned, the higher the bonus. And the points curve is exponential, not linear. So the points will change significantly if running part time vs. full time.
7im wrote:Yes, there is a bonus. The faster the WU is returned, the higher the bonus. And the points curve is exponential, not linear. So the points will change significantly if running part time vs. full time.
Adding a CPU slot, may increase your PPD (from the CPU also folding) but not to the GPU. What I was saying was that if the CPU folds or something else is using 100% of the CPU, you may need to manually decrease the number of CPU cores folding so that the GPU has at least full CPU core all to itself. If an Nvidia GPU does not get a full CPU core for its own exclusive use you will starve it and it will slow down and not use 100% of the GPU. It is generally good to have a little bit of CPU idle left.
The 90% GPU load is actually a little low. My experience is that it should be 95%-99% except for the very start of the WU, for several minutes, and periodic dropouts, every couple of frames, of a few seconds to 0% while it writes checkpoints. Are you sure that the GPU core isn't getting at least 100% of one CPU core?
Absolutely, there is a bonus for folding 24x7. It is called the quick return bonus (QRB) and it will generally produce far more points than the base point values, for your GPU, if you return the WU's as fast as they can i.e. run continuously and not part-time. Points are calculated based upon the download time, so if you shutdown in the middle the clock continues to run while there is no folding occurring thus your PPD is dropping when you are not folding and because of the QRB it is not a linear drop but an exponential one. If you need to shutdown, ideally what you do is click the finish button and wait to allow the WU to finish (Which can be a very long time) and send without getting a new one whereupon you can shut down without harming PPD.
What I did in the first place was to delete the CPU slot, because I didn't want the CPU to fold. This is where I got around 1700 estimated PPD with the GPU. Since I added the CPU slot again the estimated PPD is around 70k. Im sure there was enough CPU time for F@H (no other extensive processes runing).
Well, now it seems to work (on a Core 15 project 99% GPU load - with decreased CPU threads) - thanks to your information and help.
And I like the idea of the QRB so it will be 24x7 for me.... at least for a while
For QRB you need to get a passkey and then return at least 10 WUs with 70% (from memory) success rate.
It is still helpful to post your log (including the system and config sections), so people here can see if there are any other reasons for low ppd.
I agree with billford about what a 780Ti can get
Edit by Mod: Correction. Your total successful returns of bonus-eligible WUs must be >80%
