i learned that overclocking cpu doesnt help alot. How about overclocking gpu ?
From an AMD standpoint overclocking the GPU will most likely lead to "hung WU" which means that most likely your video cards drivers crashed. I'm actually underclocking my R9 290X TRIX cards just to keep them stable and that is an ongoing experiment.
Overclocking will most likely result in a reduced PPD due to instability if you are using AMD GPU's. I can not say if it is the same for Nvidia.
i am running the FahControl more than 6 hours with 1x 4770k (turbo enabled)and 3x 290 (1040-1300).And ppd saying around 460k-476k.Do you think lowering them to stock speed will give more ppd?
I also use the FAH Control client. I have not been able to keep any of the R9 290X TRIX cards running at 1040-1300. I'm currently at 947-1000 on most rigs. One computer is at 937-1000. Another is at 917-1000 and still shaky. Every time I see a hung WU I reduce the core clock speed by 10 and reboot. If I get to 900 on core clock speed and I'm still getting hung WU's then I reduce the memory speed by 50 and start again. It is an iterative process but I did drop from 1250 memory speed straight down to 1000 initially after I ran into issues even after reducing the core clock speeds and then I started the process over again. If you are wondering where I got the 947 from it is an arbitrary number that another user posted so I restarted there.
Watch your logs for each GPU. If the logs shows a WU is 25% complete but the client shows a higher number such as 30% then that WU is hung and will not finish without intervention. If you let it go long enough it will eventually show as 99.99% complete in the client and just sit there. That is the indication that something happened and then you need to start bringing down your clock speeds to find what is stable. You do not need to wait for the client to show 99.99% before you act. In fact, if you do you will lose hours because work isn't really being done.
If you have not run into this issue and have been running for some time (days) your cards may be able to handle it. That has not been the case with any of my 15 R9 290X TRIX cards though so I would say the odds are against you.
Some tips: If I do in fact see a mismatch I reboot the machine without pausing folding in order to reset the WU. It may take a little while in order for the client to show movement again but the log should show it fairly quickly. Some have reported that you can pause folding and then start again but I always reboot because I have found the driver crashes often times make the system unstable. I prefer to start over.
To reiterate: Watch the logs. The client is not a reliable source of information. Frequently check the percent done in the log and compare it to what the client says. Don't trust the client display. That goes for PPD estimates as well. They are also very unreliable in the client.
If you do run into hung WU's it would probably save you a lot of time and effort to start where I'm at and if you run stable start adding 10 to your clock speeds if you want. I can say the overall PPD have not changed much even after all the downclocking. The biggest killer of PPD is hung WU. You want to avoid those at all costs.
thanks for the information , can you show me an example of hung wu?
Not at the moment but it really isn't hard to identify. In the client you have three tabs. Status, System Info, and Log. You can ignore the System Info tab for this.
The status tab is the GUI that shows you graphically how far along your WU's have progressed for each hardware slot. It is also very unreliable. The WU graph is shown in the Work Queue part of the status tab. If you aren't sure which WU belongs to which hardware slot just click on the hardware slot in question and the associated work queue will be highlighted as well. The percentage shown in the work queue is what you need to compare to what is shown in the log.
The percent on the status tab will always increment but it is not necessary in sync with the log and the log shows the real answers.
When you click on the log tab you have filters at the top. I always filter the log on "Slot". So for instance let's just say I filter on slot 1 and the log shows my percent complete is 30%, but when I go back to the status tab slot 1 shows 50% complete. They are out of sync which means the work unit is hung (ie, video driver crash most likely). The status tab may say that work unit has 4 hours left to run and it will continue to progress graphically, but when you check the log it will stay at 30%. When you ultimately take action via a reboot it will start over at 30% again so the time before you notice it is simply lost.
The Log only shows full percentage increments. So if the log shows the work completed is 2% lower than what the status tab shows for the same WU then the WU is hung and you need to take action to get it moving again.
If you can't take action at the time what will end up happening is that the WU on the status tab will slowly progress up to 99.9% complete but it will never finish because it actually never moved beyond 30% based on this example. Again the percentage that counts is what is in the log for that WU.
So you would need to downclock your cards a bit and then reboot. Once you do, the status will either immediately show the reset back to 30% or it could take a bit. However, if you check the log you will see the client finds its previous checkpoint and then starts advancing again from 30%. It may take some time for the status page to reflect what the log is showing though which is why I always say check the logs. They are what counts.
Hope this helps.
