Safe HD7950 bitcoin mining.

[DerekZ06]

I just got into the very resource intensive bitcoin mining. I want to know what are the safe voltages, currents, and temps for 24/7 usage of the HD7950. The card model is a HIS iceQ Turbo 3gb 7950 Boost clock. I've underclocked and undervolted the ram attempting to save some power. I also forced the fan speeds of either 63% or 74% because I found out those speeds, the motor whine fades and I only hear the rushing air.

Currently, I am at 950mhz core, 950mhz ram, forced vcore of 1.125 with an actual vcore of about 1.055V. It is pulling an average of 12.9 amps for a total of 156watts to process about 440Mhash/s. As seen here:

Is that safe operating conditions? Also, my VRM temps are always 25C no matter what.

I've also tried more extreme scenarios such as 1100mhz core clock and 1175 core clock, but without forcing a voltage and having the card manage it's own voltage. Which I think is a no no.

Here is a screenshot of 1100mhz:

Running at that clock increased my current draw by 5.2 amps to 18.1 (217watts), while only slightly increasing my Mhash/s a second to 512.


Here is where I tried 1175mhz:

That increased my average current to 20.2amps (242watts) and again only gave me a slight boost in Mhash/s to 537.5. I found GPU-z decreases that amount slightly tho, closing it would up it to about 560 Mhash/s. Also, it appears to be actually pulling 22.3A (267watts) with the average being lower due to the card's occasional drop to idle. Pretty sure the card's not supposed to pull over 225watts.



Would this card be able to safely operate at those frequencies higher than 950mhz 24/7? Or is that even to high in itself?

 

[thujone]

depends on how much you pay per watt for electricity... but honestly with the price of bitcoins these days... i don't think you really need to worry about how much your electricity costs.

 

[2timer]

Why are you looking at current draw from the wall? That just affects your electricity bill, by itself it's not a measurement of safety.

If you want to get your voltage, first you have to have a target core frequency. What do you want to run at? What is your performance goal? Then you step down your voltage incrementally until your GPU crashes in a burn test, and that gives you the lowermost limit.

In terms of thermals, 60 C is pretty good, I'd start to worry above 70.

 

[DerekZ06]

depends on how much you pay per watt for electricity... but honestly with the price of bitcoins these days... i don't think you really need to worry about how much your electricity costs.

Electricity is 4.58 cents per kw/hr. Then divided 5 ways.

Why are you looking at current draw from the wall? That just affects your electricity bill, by itself it's not a measurement of safety.

If you want to get your voltage, first you have to have a target core frequency. What do you want to run at? What is your performance goal? Then you step down your voltage incrementally until your GPU crashes in a burn test, and that gives you the lowermost limit.

In terms of thermals, 60 C is pretty good, I'd start to worry above 70.

I'm not really worried about energy from the wall. But the current the gpu is using because I don't want the gpu to die due to electromigration. That higher current running through the cores tiny wires can't be good for it's health. I don't want the card to deteriorate to quickly.

I'm not entirely sure what I want the card to run at yet. I just don't ever want this card to die on me. I want to replace it when I'm ready, not when I need to use my computer to do something. Hopefully, two gpu generations later ~36+ months. Or after the gpu has paid for itself and the next gen.



I should also mention Im using a Rosewill Capstone 450watt psu with a single 37amp 12V rail. The cpu is overclocked as well and, during testing, the power draw would hit the high 90's of watts.

 

[2timer]

If you keep your GPU within the specified voltages and temp ranges, then there's no reason to be paranoid about it failing prematurely. Just keep an eye on the temps and if they look off, take out the card, reapply thermal paste, and reseat. Over time and long use, the thermal paste can dry up and not perform as well. Make sure your case has plenty of airflow and of course don't run your machine if ambient temps are too high.

 

[Subyman]

Fwiw, you can lower the memory speed a lot. The core is all that matters.

 

[DerekZ06]

Fwiw, you can lower the memory speed a lot. The core is all that matters.

I tried to do this, but I found lowering it below 900mhz causes corruption. The ram on this card is very finicky. It will only even overclock to 1400mhz, and it seems like everyone else's card on forums and reviews is able to go up and over 1800mhz.

If you keep your GPU within the specified voltages and temp ranges, then there's no reason to be paranoid about it failing prematurely. Just keep an eye on the temps and if they look off, take out the card, reapply thermal paste, and reseat. Over time and long use, the thermal paste can dry up and not perform as well. Make sure your case has plenty of airflow and of course don't run your machine if ambient temps are too high.

Will do, I actually had that happen to me once already on an HD7770. Since you brought it up, I'm now wondering if it's a good idea to use Coollaboratory Liquid Pro thermal paste. I bought some for de-lidding my cpu and paid a ton for it and still have a ton of it left. Bet it would significantly lower the temps, and bind the heatsink to the gpu... There's also likely resistors close to the gpu die that it could cause shorts on. I wonder if it's been done before.

 

[Accord99]

At only 950 MHz, you can probably drop the GPU voltage down from anywhere from 1.000V to 1.050V (perhaps even less), this also reduces current use. On a 7950 I have, going from 1.094v to 1.000v (@900 Mhz) reduces current use from ~115 to ~96A.

As bitcoin is barely affected by memory speed, you could underclock and undervolt the memory as well. Normally, you can underclock the memory no lower than (core clock - 150 MHz) but it's possible by changing MSI Afterburner's configuration to run at below 500 MHz in order to maximize bitcoin/watt.

 

[BallaTheFeared]

It's not really "safe" any prolonged load on your gpu is going to accelerate wear.

Low voltage/clocks = higher profit margins, but slower return.

Bitcoining a single card is, imo, a total waste of time. Take an hour out of your day to pick up pop cans on the side of the road if you really need an extra $2.50 a day, then go home and enjoy your graphics card to play games.

If you really want bitcoin, get a farm going or buy one of the new ASIC's when they're released, makes the 7950 look stupid tbh.

http://www.butterflylabs.com/

http://bitcoinmagazine.com/asicminer-starts-hashing/

http://launch.avalon-asics.com/

 

[DerekZ06]

At only 950 MHz, you can probably drop the GPU voltage down from anywhere from 1.000V to 1.050V (perhaps even less), this also reduces current use. On a 7950 I have, going from 1.094v to 1.000v (@900 Mhz) reduces current use from ~115 to ~96A.

As bitcoin is barely affected by memory speed, you could underclock and undervolt the memory as well. Normally, you can underclock the memory no lower than (core clock - 150 MHz) but it's possible by changing MSI Afterburner's configuration to run at below 500 MHz in order to maximize bitcoin/watt.

Are you able to drop the core voltage down that low in afterburner, or is that the actual VDDC voltage reported by gpu-z? My card isn't stable if I use just 15mV less. But, I'm pretty close to the 1.000V - 1.050V range if I go by GPU-Z.

If I use the stock core clock of 850mhz, based on pass experience, I know I can use a voltage of 1.025V selected in afterburner. I can't remember what the VDDC reported by GPU-z comes up to be exactly anymore, but it was less than a full volt.


I tried to underclock the memory lower, but found I got corruption below 900mhz. However, I tested that when I was using a core clock that was higher than 900mhz + 150mhz, which was 1100mhz. I found that the 950mhz ram clock worked fine when using the 1100mhz core clock and just haven't touched it since. Possibly the reason I couldn't get core clock of 1200mhz stable when I've played games at that frequency in the past.

 

[Accord99]

Are you able to drop the core voltage down that low in afterburner, or is that the actual VDDC voltage reported by gpu-z? My card isn't stable if I use just 15mV less. But, I'm pretty close to the 1.000V - 1.050V range if I go by GPU-Z.

It's the target voltage set in Afterburner, I'm stable at 900 Mhz using a 1.000V setting in Afterburner which becomes roughly 0.94v in GPU-Z. That's strange that your 7950 can't go much lower, I've used several 7970/7950s and they're all capable of running in the 1V range with a 900 MHz core clock.

I tried to underclock the memory lower, but found I got corruption below 900mhz. However, I tested that when I was using a core clock that was higher than 900mhz + 150mhz, 1100mhz. I found that the 950mhz ram clock worked fine when using the 1100mhz core clock and just haven't touched it since. Possibly the reason I couldn't get core clock of 1200mhz stable when I've played games at that frequency in the past.

It's possible, I've found the 7900s to do strange things when setting their memory clocks too low though they've always been fine running at 150 MHz below the core clock. However, looking closer at the GPU-Z image you put up, the video card's memory system isn't using much power anyways.

And seeing as how the core temperature is pretty low, and how board makers set their default voltages sometimes to 1.175v or even higher, I don't think you have much to worry about.

 

[Subyman]

It's not really "safe" any prolonged load on your gpu is going to accelerate wear.

Low voltage/clocks = higher profit margins, but slower return.

Bitcoining a single card is, imo, a total waste of time. Take an hour out of your day to pick up pop cans on the side of the road if you really need an extra $2.50 a day, then go home and enjoy your graphics card to play games.

If you really want bitcoin, get a farm going or buy one of the new ASIC's when they're released, makes the 7950 look stupid tbh.

http://www.butterflylabs.com/

http://bitcoinmagazine.com/asicminer-starts-hashing/

http://launch.avalon-asics.com/

I wouldn't mess with bit coins either. The diff is going up too high for Gpus. Litecoin is doing okay. I make around $7.50 a day after electricity on a 7970 + 6870. Then if they end up taking off, you already got a bunch because you started earlier.

 

[Will Robinson]

It's not really "safe" any prolonged load on your gpu is going to accelerate wear.

Low voltage/clocks = higher profit margins, but slower return.

Bitcoining a single card is, imo, a total waste of time. Take an hour out of your day to pick up pop cans on the side of the road if you really need an extra $2.50 a day, then go home and enjoy your graphics card to play games.

If you really want bitcoin, get a farm going or buy one of the new ASIC's when they're released, makes the 7950 look stupid tbh.

http://www.butterflylabs.com/

http://bitcoinmagazine.com/asicminer-starts-hashing/

http://launch.avalon-asics.com/

But not as stupid as buying an NVDA card to do BitCoin right?.....^_^
Remember all the Green Team dissing BitCoin mining?
Well since all that FUD,thousands of people have bought and paid off their AMD cards through mining.
How's the payback on folding and Cudie poo going?

 

[philipma1957]

I wouldn't mess with bit coins either. The diff is going up too high for Gpus. Litecoin is doing okay. I make around $7.50 a day after electricity on a 7970 + 6870. Then if they end up taking off, you already got a bunch because you started earlier.

could you send me a pm on how to setup litecoin?

 

[Sleepingforest]

But not as stupid as buying an NVDA card to do BitCoin right?.....^_^
Remember all the Green Team dissing BitCoin mining?
Well since all that FUD,thousands of people have bought and paid off their AMD cards through mining.
How's the payback on folding and Cudie poo going?

Yes, while you make back your money, Nvidia users cure diseases.

 

[Soulkeeper]

I just fried my 7850 after two days of running bitcoin with 1050MHz core with 1.14v
I think the 4 phase vrm on my cheapo diamond card just couldn't handle it because the core temp never went above 50c. I had copper heatsinks on all the mem and a Twin Turbo II on the gpu. The vrm mosfet/driver chips were the only thing not sinked.

 

[Binky]

A 7950 should do about 500mh at about 1000mhz, and maybe 570 at 1100. Try using cgminer with the diablo core.

 

[DerekZ06]

A 7950 should do about 500mh at about 1000mhz, and maybe 570 at 1100. Try using cgminer with the diablo core.

I noticed this. So I did some testing and found having a triple monitor setup affects the output as well as several background apps. Disabling all but one monitor, closing the web browser, and disabling sidebar gadgets puts it up to and slightly over the expected output.

I just fried my 7850 after two days of running bitcoin with 1050MHz core with 1.14v
I think the 4 phase vrm on my cheapo diamond card just couldn't handle it because the core temp never went above 50c. I had copper heatsinks on all the mem and a Twin Turbo II on the gpu. The vrm mosfet/driver chips were the only thing not sinked.

That scares me. But before I started, I pulled my card out to make sure it had VRM heatsinks on it since I don't have VRM temp sensors.

I now have my card running at 800mhz clock rate and 1.019V, partly because I fear I'll harm it, but mostly because I was starting to get annoyed with the 63% fan speed and I like my PC's dead silent.

Unfortunately it is only doing about 375 Mhash/s a second now, but the fan is at it's idle speed and it's only pulling 8 amps and at 64C. I'll still use the higher clocked profile when I am away from the PC.

It's also interesting to note, that right now, even at 800mhz I still can't put the voltage below a volt.

 

[thilanliyan]

You should be able to go even lower on the voltage. I have my His 7950 IceQ X2 cards at 1.0v 1000MHz core. But to get to 1.0v I had to use the newest Afterburner with the "Force Constant Voltage" option ticked...otherwise the lowest I could go was 1.125v. The His iTurbo software would only let me go down to 1.125v.

 

[DerekZ06]

You should be able to go even lower on the voltage. I have my His 7950 IceQ X2 cards at 1.0v 1000MHz core. But to get to 1.0v I had to use the newest Afterburner with the "Force Constant Voltage" option ticked...otherwise the lowest I could go was 1.125v. The His iTurbo software would only let me go down to 1.125v.

I have the latest version of afterburner with "Force Constant Voltage" checked. I never bothered to install the included HIS software.

It will run at slightly lower voltages and appear to be operating well. However, what has forced me to keep them high is GPU-Z. Every time I changed a setting in AfterBurner, in order for me to get an accurate reading on the average current draw and other average values, was to open a new instance of gpu-z --because it does not have an option to reset sensor values. I found that if the voltage is any lower than it is now and I launch GPU-z, the system will lock up and force me to hard reset the tower.

From what I've gathered so far, it seems to have to do with GPU-z crashing at OpenCL detection while OpenCL is being used to mine bitcoins, when the GPU voltage is lower than my current settings. After the system is back online, the next time I open GPU-z I get this error:

So far, my work around is not re-enableing OpenCL detection. The reason why I feel it is not GPU-Z at fault, but my card, is that this only occurs at lower voltages. This likely is just GPU-Z exposing an instability caused by the lower voltage. Im sure I have the latest Open_CL version, version 2.8. And my GPU drivers are also the latest 13.3, albeit it is still in beta. I was extremely disappointed to see AMD not release a stable version of the 13.2.



I don't think I got a very good chip. I've had terrible luck with this whole build. Even my i7-3770k requires more voltage than comparable systems to be stable. I had to settle for just 4.2GHZ on it because I am unwilling to supply it with over 1.25 vcore 24/7, while other people at the same clock are at as low as 1.1V. After deliding the cpu, the highest I could get to was 4.4ghz with a forced constant voltage of 1.3V. I could never get anything over 4.3ghz stable using dynamic offset voltage. Anything over 1.3V actually would cause the stability to go backards. ie. If I set 1.35v, it wasn't even stable at 4.0 Ghz. It's a shame too, because of everything wasted having to delid the chip, the cpu won't even break 62C on the cores or 54C at the CPUTIN sensor. It's cool enough to keep pushing, but won't go higher. I'm at the same clocks before the delid, just 30C cooler. I'm certain it has a defective cache. If I lower the voltage I get BSOD 0x124, and if higher the voltage I get the same BSOD 0x124, in both cases windows says something like GENUINE.INTEL.PROCESSOR <CACHE something something something >. The cpu has a narrow voltage window.

I can't look back on the exact errors because I reinstalled windows after I found the overclock the computer would be rock solid at. I made it pass 12hrs Prime 95 in-place FFT, 18hrs prime95 blend w/ 85% ram allocated, and 8hrs of IntelBurnTest at maximum. 4.2ghz is the highest it can do it at. I'm not sure if others who are using less voltage hold their cpu to the same standard. If they do then I have just terrible terrible luck. Probably just terrible luck, if I follow a guide and set an overclock considered "normal" and easy, I won't even get to windows. Both the CPU and the GPU have amazing temps but I can't do anything with it. Maybe there's a deeper issue, but I can't figure it out.

 

[BallaTheFeared]

It's the dark underside of overclocking nobody wants to talk about, the avg to below avg chips.

 

[DerekZ06]

It's the dark underside of overclocking nobody wants to talk about, the avg to below avg chips.

Yea that's what I kinda figured. Only see success stories out because everyone likes to brag a little and show off their success a bit.

I would have if I could have.

What really sucks, is that pretty much every overclocking guide assumes you got a good chip. For my processor, I remember the first guide I followed. I set everything exactly as it said, and it failed so bad I had to clear the CMOS. Actually, now that I think about it, there's not a single guide I found that worked on my cpu. Mostly because they all recommended using offset voltages. I had to fix the voltage just to reach 4ghz and through the course of a few days got it to 4.4ghz. Then spent over a week slowly transitioning it, setting by setting, to do 4.3 on dynamic offset voltage to pass stress testing. Only to find out that when it wasn't under full load, allowing the voltage and clock to fluctuate, there was a lot of random 'anomalies' and other very preculiar glitches

 

[wbynum]

I wouldn't mess with bit coins either. The diff is going up too high for Gpus. Litecoin is doing okay. I make around $7.50 a day after electricity on a 7970 + 6870. Then if they end up taking off, you already got a bunch because you started earlier.

What is your combine kH/s with those two cards? With the latest difficulty, you should be grossing a max of ~$7 per day before electricity. I used the following two sites to come up with that number (figure optimistically 1000 kH/s combined with the two cards).

http://www.litecoinpool.org/calc?hashrate=1000&difficulty=78.3

https://github.com/litecoin-project/litecoin/wiki/Mining-hardware-comparison

 

[DerekZ06]

Just thought I'd update on some things because I finally applied the coollaboratory liquid pro. Also, I previously said my ram won't overclock much more than 1400mhz. Well, I just noticed that I had it undervolted in my overclocking profiles. I must have done that a while ago and just forgot about it.

The liquid pro on the heatsink and die:

Temps with stock TIM:

@ room temp of 75 F.

Temps with liquid pro:

@ room temp of 77 F.

I expected a bigger drop than 3C, I may have been able to improve a lot more if I lapped the heatsink. That thing in current condition is far from smooth, it's like the meant for it to have a rough finish.

 

[wbynum]

Few things:

- To get a decent bump in MHash you need to bump the intensity (-I value) up to 9.

- Both temps 60c and 65c and very cool for mining. I have a few cards running at ~80c during the heat of the day without issue. With the nice HIS blower, heat should not be a problem for you.

- Not sure if you are bumping up the memory or not, but if you are, stop. Bitcoin mining does not use memory. Running a lower memory setting will make the card run cooler and will actually give a better MHash setting sometimes.

BTW, at an intensity of 9 and a 1000mhz core clock on my 7950's I get ~530 MH/s per card.

Hope this helps.