7970 clocking down to 500Mhz while mining.

[Minn0W]

Hi all,

I have a HD 7970 that appears to keep clocking down to 2d mode (500Mhz) instead of staying in 3d mode (1000Mhz)

I have had to modify the cooling because it did not fit corectly and the GPU was overheating. Would easily reach 97 deg C with stock cooler. I believe the die was manufactured incorrectly because the metal shield around the die was taller then the die its self, So i milled off the excess corners of the copper surface off an Artic Accelero Xtreem Plus II.
Now she runs much much cooler, usually around 70 deg C.
I say it was a Saphire 7970 because in an attempt to fix this speed cycling issue, i have flashed it with the 7970 Ghz Edition firmware. And it all works perfectly, but i am still having this issue.

This is happening with DiabloMiner and poclbm.

I am now using cgminer with these settings...

Code:

"intensity" : "9",
"vectors" : "1",
"worksize" : "256",
"kernel" : "poclbm",
"lookup-gap" : "0",
"thread-concurrency" : "0",
"shaders" : "0",
"gpu-engine" : "1000",
"gpu-fan" : "0-100",
"gpu-memclock" : "0",
"gpu-memdiff" : "0",
"gpu-powertune" : "0",
"gpu-vddc" : "0.000",
"temp-cutoff" : "97",
"temp-overheat" : "95",
"temp-target" : "85",
"api-listen" : true,
"api-port" : "4028",
"expiry" : "120",
"failover-only" : true,
"gpu-dyninterval" : "7",
"gpu-platform" : "1",
"gpu-threads" : "5",
"hotplug" : "16",
"log" : "5",
"no-pool-disable" : true,
"queue" : "2",
"scan-time" : "45",
"temp-hysteresis" : "3",
"shares" : "0",
"kernel-path" : "/usr/local/bin"

And here is a screen shot over GPU-Z while running cgminer for a couple of minutes...
http://vitl.co.nz/7970-Mining.jpg

You can see the GPU cycling down quite often.

I have just tested the GPU while playing crysis 3, and i did cycle down, so i though i would try lowering the voltage on my 12V rail, it was set to 12.8v, and is now set to 11.5v.
Crysis 3 ran at 1000Mhz for the full 10 minutes i was playing it. But bitcoin mining is still giving me issues.

Does anyone know what on earth is going on here !?!?

Update 1:
The behavour appears to be a function of the temperature, it will cycle down when the card goes over 55 deg C.
I am starting to wonder if i need a heat sink on what looked like the oscillator on the board.

Update 2:
I  have just put heat sinks on what i believe are transistors for the power circuit....

http://www.vitl.co.nz/front_full.jpg

And it has improved a little but still not good. It still starts to cycle down to 500mhz @ 55 deg C but not as often.
Putting a fan heater on fan mode against the card appears to almost fix the issue, but this is far from ideal.

Update 3:
Something interesting happening...
As soon as the core clock goes over 1000Mhz, the core voltage switches from 1.072v to 1.119v.
And only when that happens, the card will cycle down to 500Mhz periodontally.

Update 4:
Getting VERY good results using Saphire Trixx by lowering the voltage.
Download:
http://www35.zippyshare.com/v/4022846/file.html

Update 5:
Purchasing water cooling system.
Lowering the voltage did help a lot, bit it still all depend on the temperature.

[1713992518]

1713992518

[1713992518]

1713992518

[1713992518]

1713992518

[akipfer]

desita minerar bitcoin amigo, GPU não compensa

bora LTC

[gateway]

Im pulling in around average of 700~ with my sapphire 7970 OC...

gpu 1182
mem  = 1043
vddc - 1231
boost 20

cgminer

[vm1990]

hi the die is suppose to be lower than the surrounding metal this is because in previous card people put after market heatsink on and over tighten things which over time with heat cause the die chip to sink into the PCB board and crush the solders into each other

also the things you put heat sinks on are VCC's they control the voltage and if they get to hot they throttle the entire card

of your still having problem and still have the original heatsink then look at it and where there where white thermal pads is what need cooling (there there for a reason)

[Minn0W]

hi the die is suppose to be lower than the surrounding metal this is because in previous card people put after market heatsink on and over tighten things which over time with heat cause the die chip to sink into the PCB board and crush the solders into each other

also the things you put heat sinks on are VCC's they control the voltage and if they get to hot they throttle the entire card

of your still having problem and still have the original heatsink then look at it and where there where white thermal pads is what need cooling (there there for a reason)

I do think the VCCs may be my issue. After putting heat sinks on them, it is running better, but not great.
I still have to run my VDDC < 1120 to prevent the card from throttling.

The factory heat sink was absolutely terrible though.

But my water block should be in the mail tomorrow, and it will be interesting to see what i can get out of the card when it is water cooled.

[vm1990]

non of the standard card have terrible cooling with default heat sinks, i have bigger issues with after market one that dont have all the correct parts to cool important parts of the card same as water cooling i hate the ones that just cover the gpu chip bit the ones that cover nearly all the card there the best ones cool everything important and keep all of the card nice and cool

[Minn0W]

Well I bought a water block, pumped my voltage too high, Killed a VRM, Chopped the blown one off the board with a knife, and im back in action.
Quite happy i was able to recover the 7970 after blowing a VRM and the fuses on the board.

[roy7]

Sounds like powertune, I was having the same issue. Apparently to turn it "off" you need to set it to +20%. Then it'll stop lowering your clock now and then to save power.

[Minn0W]

Interesting... I will set powertune to +20 and see if it differs to my usual behaviour.

[Minn0W]

Setting powertune to +20 did help, but only very slightly.
With my 1 dead VRM chip, and VDDC set to 1125, and GPU Core set to 1180, i still get my card dropping down to 500mhz every time it goes full load.

I will set the GPU to 1200 and decrease the GPU voltage until it crashes.
TBC....

[Minn0W]

Instant crash :-( My card is not exactly running at its full potential.

Anyone know where i could purchase a new VRM cihip...
I will illustrate the one i has to chop off my board to get it working again...
http://vitl.co.nz/VRMs.png

I would like to populate my board correctly.
I need either a faulty 7970 or the replacement parts.
Anyone selling a dead 7970 ?