If you are running Genefer or Genefer-WR and want to try overclocking, or are already OCing and getting errors, here are some tips. As a general rule the GPU memory needs to be DOWNclocked 10-15% and the temperature needs to be at reasonable number. Overheating will dramatically increase the chances of errors. Setting GPU fans manually to MAX while running GFN should help some (or a lot). Having a GPU with 2 or more fans run cooler than reference single blower ones. Heat is the enemy in electronics.

Now if you really want to run GFN and overclock with a much lower chance of errors you need to lower you GPU memory even further as you increase the GPUs main core clocks. Mid powered GPUs do run cooler than high end GPUs in general but when OCing the high end GPUs will increase temperature faster than mid powered GPUs when OCing the same percentage amount.

Examples for nVidia Fermis (400 & 500 series) that worked for me:

GPU--------------core clock/shaders---memory---voltage

Stock GTX460-----675/1350-----------1800-------0.975
OCed GTX460-----750/1500-----------1700-------1.000
OCed GTX460-----850/1700-----------1500-------1.012

Stock GTX470-----607/1215-----------1674-------1.012
OCed GTX470-----662/1325-----------1550-------1.025
OCed GTX470-----738/1475-----------1400-------1.037

Stock GTX560ti---822/1645------------2004-------1.012
OCed GTX560ti---900/1800------------1700-------1.025
OCed GTX560ti---937/1875------------1600-------1.037

Stock GTX570-----732/1464------------1900-------0.988
OCed GTX570-----800/1600------------1700-------1.000
OCed GTX570-----850/1700------------1600-------1.012

Stock GTX580-----772/1544------------2004-------1.013
OCed GTX580-----825/1650------------1700-------1.025
OCed GTX580-----850/1700------------1600-------1.037

I don't have any Kepler or Maxwell GPUs. But due to much lower Double Precision than Fermis (except the Titans) they should run cooler to start with but the same rules should apply.

Examples for AMD Radeons that worked for me:

GPU--------------core clock---memory---voltage

Stock HD5830---800---------1000
OCed HD5830---900---------750

Stock HD7950---925---------1250-------1250mV
OCed HD7950---1000-------1200-------1250mV
OCed HD7950---1035-------1150-------1250mV
Don't know why stock voltage is higher than the 7970. Bought this used so previous owner may have messed with the GPU BIOS. This is the boost edition yet the base clock is at the boost clock. further evidence that the BIOS has been changed.

Stock HD7970---955---------1375-------1225mV (non GHZ edition)
OCed HD7970---1050-------1300-------1250mV
OCed HD7970---1125-------1250-------1265mV

It seems that the memory on the Tahiti GPUs doesn't need to be lowered as much as the other GPUs I've tested. So the the AMD Radeon HD 7870/7950/7970/R9 280X series has the best DP along with the Titans currently with AMDs R9 290X (Hawaii) not far behind. The soon to be released 390X may (should) be a powerful GPU with good DP (Fingers crossed).

Due to inconsistent manufacturing quality of GPU chips not everyone may achieve success in overclocking, but you won't know until you try.

Would users running GFN post their settings that run successfully & consistently for more than 3-4 times to help others who want to try to OC their GPUs. You can also post about failures.

PS: Mike or other moderator can you sticky this post so it stays visible near the top over time. Remove this PS afterwards
____________
Largest Primes to Date:
As Double Checker: SR5 109208*5^1816285+1 Dgts-1,269,534
As Initial Finder: SR5 243944*5^1258576-1 Dgts-879,713

Lowering memory below stock is the key. It only slows computation very slightly. Overclocking the core increases computation speed significantly more than the slight decrease with memory downclocked. You should lower memory 2-3 times the percentage of the increase in core clocks. So don't try and be brave and run memory at stock speed. Even if you run at stock core speed you should lower memory at least 5-10% below stock. This is the key to avoiding errors.
____________
Largest Primes to Date:
As Double Checker: SR5 109208*5^1816285+1 Dgts-1,269,534
As Initial Finder: SR5 243944*5^1258576-1 Dgts-879,713

Exit 0 status no finish file: a mild overclock yes but 1 card is repeatedly fine while the other gives that error after a few hours. Mem clock standard, fans 80%+, 80C - any ideas why they would fail? Card does everything else fine.

80C is probably too hot on the Fermi cores. I ran into trouble with the GFN short and some PRPnet GFN work when a couple of my Fermi based cards ran that hot. Get it down to the mid-70s and you will likely be okay.

Here are some of my results with a Gtx 970. It was the Gigabyte Windforce OC model (not the G1).

1: I bought that card for a gaming rig. So naturally, I started there.
- +190/+0/+0 for the Core/Mem/Voltages was stable (didn't try OCing Mem). Days of benchmarking, gaming, stress testing, never a failure.
- Allowing for overvoltage, I managed to get +240/+0/+87.5. Once again, I didn't try mem. OCs.

2:HOWEVER, due to the P2 crap that plagues the 9xx series on distributed computing tasks, I didn't need much time to have crashes on the thing. I tried Folding Home, GPUgrid, POEM Home, and Primerid. While some were more resilient to OC than others, the final setting that was good for every single one of them was +164/"+500".
-I say "+500" because, in that power state, the mem. runs at 3001 (6002 effective), instead of the 3505 (7010 effective) mhz that it is supposed to. Thus, I "overclocked" the RAM to match the correct speeds.
-And yes, +164. +165 caused failures once every 2~3 weeks; a rare cenario, but a crash regardless. With +164, the thing's been going for months without problems. And while the overvoltage does allow the boost clock to go higher (so the final clocks go up too), it won't actually allow me to OC any higher, due to - again - the P2 driver bug.

As a final note, I didn't try to see if I can OC any higher with the particular instances of GFN-Short and GFN-WR. As said, given that I found a "be all, end all" stable OC, and that I switch between projects a lot, I decided to just stick with that. I also won't be OCing the RAM; it just won't let me (screw you P2!).

But I may try lowering mem. speeds to try OC the core higher. Do you guys think it will be worth it?

Experimenting with dowclocking the memory is a good idea, trying to oc the gpu core more that way.
But oc-ing with ov and running 24\7 is not a good idea imo.
I never do that with my hardware, i only oc or occasionally downclock.

When temp was an issue with some of my gpus that had poor cooling in the past, i did downclock. However i avoid cards with poor cooling, so i have not had the need to downclock for many years.