Topic: to OC or not to OC? (Read 1874 times)
My computer makes it very, very clear when I've overextended the card...the driver either crashes (and usually restarts) and my MHash/sec drops to almost nothing, or it just locks up outright and I have to hard reboot it. I've also found that I can run my card pretty fast provided I'm not doing anything else with the system. So, I guess it will depend on whether you are actually using the computer while it's mining...or, at all (since sometimes, just bugging the computer when the clocks are unstable will cause one of the previous two outcomes).
Overclocking does increase power usage. Anyone who is posting that the total power usage is the same for an overclock is not actually measuring the power use directly.
I have been able to undervolt and mild overclock and keep power usage the same or undervolt with stock clocks and lower power usage.
I do not recommend personally going over 15% for overclocking, though there are many who are doing so successfully. I do not want the system to lock up and me not know about it for hours. One crash can kill all of the benefit of overclocking.
Meh. I overclocked a processor I had past 100%. 1.86ghz to 3.74ghz. Nothing to be afraid of if you know what you're doing.I have been able to undervolt and mild overclock and keep power usage the same or undervolt with stock clocks and lower power usage.
I do not recommend personally going over 15% for overclocking, though there are many who are doing so successfully. I do not want the system to lock up and me not know about it for hours. One crash can kill all of the benefit of overclocking.
Even for mining, I still think going as far as you can is best. Let's look at a worst-case scenario...
I don't believe I've seen a card go less than a 50mhz overclock without being stable. So lets say, for the sake of argument, that 50mhz is the LEAST you could gain with an overclock. Also, for the sake of argument, let's say the worst case stock top speed of a card is 900mhz (?).
So, you try overclocking to 975mhz, and it fails overnight. No mining for 8 hrs. Which means, 950mhz * 8 hrs = 7600 mhz-hrs of lost production. You settle on 950mhz, and it is stable. 7600mhz-hrs / 50mhz = 152 hrs. 152 hrs = 6.3 days.
So worst case, if you're looking at overclocking vs not overclocking, you can make up for lost production within a week.
On the other hand, if you DON'T go for the best overclock you can get, and miss out on say, 25mhz, you could lose the equivalent of 8 hrs of production per week. That'll add up quick, especially on some of the higher-end cards.
Overclocking does increase power usage. Anyone who is posting that the total power usage is the same for an overclock is not actually measuring the power use directly.
I have been able to undervolt and mild overclock and keep power usage the same or undervolt with stock clocks and lower power usage.
I do not recommend personally going over 15% for overclocking, though there are many who are doing so successfully. I do not want the system to lock up and me not know about it for hours. One crash can kill all of the benefit of overclocking.
I have been able to undervolt and mild overclock and keep power usage the same or undervolt with stock clocks and lower power usage.
I do not recommend personally going over 15% for overclocking, though there are many who are doing so successfully. I do not want the system to lock up and me not know about it for hours. One crash can kill all of the benefit of overclocking.
Don't OC, if the manufacturers could squeeze out a greater clock rate without risking a high rate of failure they would have.
Not entirely true.
The manufacturers didn't make these cards for hashing bitcoins, they made them for playing video games. A bit of tweaking to adjust them for bitcoins is almost common sense!
Don't OC, if the manufacturers could squeeze out a greater clock rate without risking a high rate of failure they would have.
My 3 HD 6950 are able to run at stock frequency at 0.95V insted of 1.1V. I did stability tests and everything went fine, the power consumption is way lower. Not all the chips are the same and, insted of overclocking i would try to lover the voltage. At the end you are going to make more profit anyway due to the less power energy drained and the cards will not be stressed.
Don't OC, if the manufacturers could squeeze out a greater clock rate without risking a high rate of failure they would have.
You could always use my "quick n dirty" method.
Start guiminer (because it's easy to make it stop/start)
Overclock +10%, immediately.
Overclock in 25mhz intervals afterward
Once computer locks up, restart
Reduce overclock to 25mhz below the point of locking up.
Start guiminer, and keep an eye on it for stability. If it crashes, reduce overclock another 25mhz.
Took me all of about 5 minutes to overclock my 5850 and 5770 in this manner. They've been stable and mining away for the past 10 hours, no problems so far...
Much easier, faster, and less obnoxious than the "+5mhz, wait 15 minutes, +5mhz, wait 15 minutes" methods. I really don't get why people do that... If you need to eek out that last 10mhz of performance, then at least get a generic idea of the mhz range you need to fine-tune your overclock in first.
Start guiminer (because it's easy to make it stop/start)
Overclock +10%, immediately.
Overclock in 25mhz intervals afterward
Once computer locks up, restart
Reduce overclock to 25mhz below the point of locking up.
Start guiminer, and keep an eye on it for stability. If it crashes, reduce overclock another 25mhz.
Took me all of about 5 minutes to overclock my 5850 and 5770 in this manner. They've been stable and mining away for the past 10 hours, no problems so far...
Much easier, faster, and less obnoxious than the "+5mhz, wait 15 minutes, +5mhz, wait 15 minutes" methods. I really don't get why people do that... If you need to eek out that last 10mhz of performance, then at least get a generic idea of the mhz range you need to fine-tune your overclock in first.
I recommend taking a weekend and playing with it. My general method is to bump up the core 5 mhz at a time, and then let it sit for 15 minutes, while watching for the 'verification failed, check hardware' messages.
Once I see one of those, back it down 5 mhz, and leave it running, checking in every 30 minutes. Normally if the machine doesn't produce a verification failed message within 2 hours, then it will stay stable 24/7.
I have 2 GPUs out of the 10 running that will pop up the verification failed message every 3-4 hours. The percentage of time/shares lost to these errors is less than the loss in mHash/sec if I were to drop it 5 more mhz, so I'm letting them run with it.
Once I see one of those, back it down 5 mhz, and leave it running, checking in every 30 minutes. Normally if the machine doesn't produce a verification failed message within 2 hours, then it will stay stable 24/7.
I have 2 GPUs out of the 10 running that will pop up the verification failed message every 3-4 hours. The percentage of time/shares lost to these errors is less than the loss in mHash/sec if I were to drop it 5 more mhz, so I'm letting them run with it.
i agree. play with it.
i also note that, while overclocking the GPU and overvolting the card can add more stress (i.e., heat) and reduce the life-expectancy of the card; underclocking the memory (typically down to 300 Mhz or so - in little steps...) does two things, one of which is entirely counter-intuitive:
1.) it increases Mhash/sec! whooda thunk it? and...
2.) it lowers the temperature at which the card runs - most often lowering it quite markedly.
and let's remember that it is heat which is the great enemy - nothing else. just using a card hard won't hurt it, as long as it stays relatively cool.
so i would heartily recommend adding overclocking to your repertoire of skills...
I recommend taking a weekend and playing with it. My general method is to bump up the core 5 mhz at a time, and then let it sit for 15 minutes, while watching for the 'verification failed, check hardware' messages.
Once I see one of those, back it down 5 mhz, and leave it running, checking in every 30 minutes. Normally if the machine doesn't produce a verification failed message within 2 hours, then it will stay stable 24/7.
I have 2 GPUs out of the 10 running that will pop up the verification failed message every 3-4 hours. The percentage of time/shares lost to these errors is less than the loss in mHash/sec if I were to drop it 5 more mhz, so I'm letting them run with it.
Once I see one of those, back it down 5 mhz, and leave it running, checking in every 30 minutes. Normally if the machine doesn't produce a verification failed message within 2 hours, then it will stay stable 24/7.
I have 2 GPUs out of the 10 running that will pop up the verification failed message every 3-4 hours. The percentage of time/shares lost to these errors is less than the loss in mHash/sec if I were to drop it 5 more mhz, so I'm letting them run with it.
With overclocking you run the risk that your miner will crash or your computer will lock up. Until you find settings that are stable for 24/7 100% operation, you could run into many of these crashes.
To make it worth it, the extra Mhash/sec you can get from overclocking must offset the downtime from these crashes. If you're able to babysit your computer throughout the day to reboot, tweak OC settings, restart miners, etc. then it's probably worth it to spend some time on OCing. If you're going to be away at work/school/whatever for long stretches of time during which your miner may crash and not make any BTC for you, then I'd stay stock.
To make it worth it, the extra Mhash/sec you can get from overclocking must offset the downtime from these crashes. If you're able to babysit your computer throughout the day to reboot, tweak OC settings, restart miners, etc. then it's probably worth it to spend some time on OCing. If you're going to be away at work/school/whatever for long stretches of time during which your miner may crash and not make any BTC for you, then I'd stay stock.
1.7 gh/s is pretty good, if you've never overclocked before I wouldn't risk it until those cards are no longer profitable and then try to squeeze some more life out of them when you know what you're doing.
ahh i see, well yeah im getting 1.2 gh/s right now, you sure itll reach 1.7? thats very tempting though
edit: I thought you where getting 850Mh/s not that the core was 850Mhz, lol
It's never too early to start learning. Perhaps start out with a mild overclock, just to get the hang of things. MSI Afterburner is fairly easy to use, especially if you are already using MSI hardware (you won't have to tweak the config file).
best to err on the side of caution 
1.7 gh/s is pretty good, if you've never overclocked before I wouldn't risk it until those cards are no longer profitable and then try to squeeze some more life out of them when you know what you're doing.
ahh i see, well yeah im getting 1.2 gh/s right now, you sure itll reach 1.7? thats very tempting though
1.7 gh/s is pretty good, if you've never overclocked before I wouldn't risk it until those cards are no longer profitable and then try to squeeze some more life out of them when you know what you're doing.
so yeah i got two 6990s running @850...
should i oc both of them?
what are the long term risks?
will it ruin their life expectancy and stuff?
i have never oced anything in my life so im not sure about the risks for mining.
should i oc both of them?
what are the long term risks?
will it ruin their life expectancy and stuff?
i have never oced anything in my life so im not sure about the risks for mining.
Jump to: