Author

Topic: Overclocking a water cooled 5970 (Read 3324 times)

zvs
legendary
Activity: 1680
Merit: 1000
https://web.archive.org/web/*/nogleg.com
June 10, 2012, 01:11:54 AM
#13
You're pretty much maxing out the capabilities of the 5970. You could probably push 950-1000 MHz with an increase in voltage to 1.25v or so but your power consumption is going to spike. I personally run mine right now at 800 MHz core, 500MHz memory with stock volts to get the highest MH/w efficiency. I find it more helpful to just run them at conservative clocks, cool temperatures(I'm around 45-50C core temps, 55-60C VRM temps) and let them run forever. I had around 40 days up-time on my watercooled rig before I moved it to another location. Now in a more permanent location I fully expect 4-6 months up-time before I sell the cards off.

Note: I found a 120mm fan cooling the back of the card by the PCIe connector helps to cool the PCB and VRM for that core. I have 3 cards per rig and just plop the fan on top of the cards to blow down on all three. Otherwise just let the cards grind away  Smiley
I run my 5970's at 815/150 @ 1.010v during the day (sdk 2.1), about 750mhash. 

Occasionally at night time I'll boost them to the highest stable I've found at 1.050v (ranges from 820-850 for various gpus)
legendary
Activity: 1344
Merit: 1004
June 05, 2012, 06:02:24 AM
#12
How bad does overclocking affect the power efficiency? I thought it would improve, but I guess that I am mistaken?

Increasing core clock by itself should increase MH/j efficiency a little, but you should gain a lot of raw processing power in too.
Increasing core voltage GREATLY decreases efficiency and no increase in core clock will make up for the MH/j loss, but if you pay very little or nothing for electricity, running with minimal undervolt or stock volts is an option. I pay $0.12069/KWh in Wisconsin and found that while I lose MH/j going from 0.95v core to 1.0625v core, the increase in core clocks should increase my net profits about 5% or so according to the spreadsheet I made.
newbie
Activity: 52
Merit: 0
June 05, 2012, 04:46:31 AM
#11
How bad does overclocking affect the power efficiency? I thought it would improve, but I guess that I am mistaken?
sr. member
Activity: 274
Merit: 250
June 03, 2012, 10:55:55 PM
#10
My golden spot 2 years ago for watercooled 5970 was 970Mhz@ 1.2 Vcore with 90*C VDDC temp.
Blaizing speed Cheesy and blaizing power bill Cheesy

Now i got 3 watercooled 5970 and i`m using them with those parametrs: 760@1 with no idea what temp on VDDC (linux)
If You dont pay for electricity, or You dont care for such things, Push it HARD! Tongue
newbie
Activity: 52
Merit: 0
June 02, 2012, 10:19:30 AM
#9
I bought the card with the cooling block installed, so I don't know how well everything is physically connected, but it seems as if every component is getting proper cooling.
hero member
Activity: 697
Merit: 500
June 01, 2012, 03:31:25 PM
#8
You're pretty much maxing out the capabilities of the 5970. You could probably push 950-1000 MHz with an increase in voltage to 1.25v or so but your power consumption is going to spike. I personally run mine right now at 800 MHz core, 500MHz memory with stock volts to get the highest MH/w efficiency. I find it more helpful to just run them at conservative clocks, cool temperatures(I'm around 45-50C core temps, 55-60C VRM temps) and let them run forever. I had around 40 days up-time on my watercooled rig before I moved it to another location. Now in a more permanent location I fully expect 4-6 months up-time before I sell the cards off.

Note: I found a 120mm fan cooling the back of the card by the PCIe connector helps to cool the PCB and VRM for that core. I have 3 cards per rig and just plop the fan on top of the cards to blow down on all three. Otherwise just let the cards grind away  Smiley
hero member
Activity: 518
Merit: 500
June 01, 2012, 10:30:58 AM
#7
Are the vrm's labelled VDDC(1,2,3) ?

If so, those are running about 66C right now, I will have to see what they hit when it gets hour again next week, but I can't imagine that they would hit 100

well, was just sayin because one of my friends had a 5970 on water, and he didnt know that he had to cool the vrm down appropriately(and using linux it was hard to find out the vrm temp on a headless system). we spent almost 4 hours until I came up with putting some cooling on the vrm
newbie
Activity: 52
Merit: 0
June 01, 2012, 09:08:35 AM
#6
The current is running ~69A pretty solid, did see one very small gap
newbie
Activity: 52
Merit: 0
June 01, 2012, 09:05:03 AM
#5
Are the vrm's labelled VDDC(1,2,3) ?

If so, those are running about 66C right now, I will have to see what they hit when it gets hour again next week, but I can't imagine that they would hit 100
hero member
Activity: 658
Merit: 500
June 01, 2012, 08:56:34 AM
#4
Yep the number 1 thing is get the three little shiny VRM up between the gpus to stay cool. Use the best TIM you can find. I have tweaked both of my 5970's using the stock fans to run at 900/900 on air and keep those VRM under 100C at 1.125V. That's the bottleneck. If they get anywhere near 125C they cut the power and the card throttles. Keeping them below 100C will greatly increase the life of the cards.

Run gpu-z go to sensors and watch the VRM temps. Above all else if you can keep the vrms cool you can overclock to your hearts content.

If you look at the vddc current it'll be 60 something amps on a good overclock. If you see little lines/gaps and not a solid red bar your card is throttling.

newbie
Activity: 52
Merit: 0
June 01, 2012, 08:52:38 AM
#3
Those are the other 2 temps in gpu-z right?  If so then they have never been more than 10C hotter than the gpu temp. I.e. ~60C or less.
hero member
Activity: 518
Merit: 500
June 01, 2012, 08:48:18 AM
#2
make sure you cool the VRM appropriately(this one usually gets cooled by the stock cooler), otherwise the card might shut down and you spend lots of time looking for whats wrong(I did). The higher you push the card, the hotter the VRM gets.
newbie
Activity: 52
Merit: 0
June 01, 2012, 08:44:35 AM
#1
I've been overclocking an Xfx 5970 black edition with an Ek water block.  I have never seen temps above 50C so far and today it is running at about 40C due to a cold weather front that popped up this week.  (also have it in the garage where it is cooler)

My question is, since I have no chance of over heating the card, what is the chance of causing damage to the card with aggressive overclocking and over voltage.  I am currently running 1.139V @ 950 Mhz.   I'm getting about 415Mhash per core for ~830 for the whole card using cgminer, catalyst 12.3.  I think my settings are -v2 -w128 -g2 -i7.  I could probably set the intensity higher since it is a 2500k system and I don't really get on it too much for more than web browsing or burning a dvd.

Any suggestions for improving my hash rate?
Jump to: