Mobility 5870 Undervolting Results

System:
Clevo w870cu
i7 820QM
Mobility 5870
4GB DDR3-1333
Hitachi 7K500 + 5K500.B

I'm on medical leave and have a decent amount of free time, so I thought I would do this to clear up the speculation on whether or not undervolting the 5870 is possible or worth while.

*Disclaimer* - I'm not responsible for a bad flash or any other damage that comes from this guide. If you're not confident in your ability, don't try it.

For my flashing method I used Jstarnino's post here, with one exception: I used GPUz 0.4.6 to dump the BIOS. Be sure to keep a backup of the original vBIOS.

These tests were all done at 800/1100, because that's the setting I use for gaming, and the most common overclock I see with this card. I used AMD Clock Tool to measure temps, and OCCT for stability testing.

For stability testing, the core setting had to pass 20 minutes of OCCT with error checking, which was setting to the highest detection setting. I tested 800/1100 for 1 hour as the last test.

The OCCT temps are after 10 minutes of the GPU test being run with the shader complexity set to 8.
Ambient temperature was a controlled 22*C.

Results
( Voltage) ( Highest Stable GPU Clock) (OCCT Temps [ Core/MemIO/Shader]) ( Idle MemIO)
( 1.15) ( 910) ( 84.5 / 110.5 / 91) ( 61)
( 1.10) ( 870) ( 81.5 / 105 / 87.5) ( 60)
( 1.05) ( 830) ( 79 / 101 / 84.5) ( 58)

MemIO is part of the GPU, it isn't the memory temps, it's the memory controller.

I'm really surprised that it was able to overclock all the way down to 1.05. I was thinking that 800 core would be tough to pull off at 1.1, let alone 1.05. I'm not going any lower though, the curve shows it's probably not going to do 800 at 1v. Still, the results are pretty positive, I just wish I had a watt meter to test the power draw.

 

100oC+ on mem? That doesnt sound safe to me

 

910 is a pretty hefty OC - you sure your mem/vram thermal pads are seated correctly? 100c+ alone is worrying but the fact the the core is more than 20c apart from the mem is a hint something is amiss. There shouldn't really be more than a 10c difference...

 

ya , ur Mem IO is too hot.. try doing something about that..

 

Ya, I know the TIM needs to be redone, I just haven't had the opportunity to do it yet. Maybe I'll get it done today. It is common for the MemIO to be 10-15 degrees above the shader temps though, even on desktop cards. Especially when running OCCT, which is as heavy as Furmark.

Kind missing the point, a 9% undervolt with a 11% overclock is good news, no? Especially with my temps being kinda high to begin with. This means less draw, less heat, and more battery with no downside.

 

thats great to badi'm skiping that gen

 

lol... but JohnnyFlash , 10-15C isn't regular.. what worries me is that 110C is too near the shutdown spec.. u don't want ur memory burning out u know..

 

MemIO is not the memory, it is located on the GPU die and it is a known problem that the GPU heatink is not flushed with the GPU die, causing a higher MemIO temp. he either needs thicker paste, or a copper shim.

 

What MemIO temps do you get running OCCT or Furmark? These aren't gaming temps. Running Crysis for an hour doesn't come close to 100.

Ya, the problem is a combination of the video card cage and thermal pads being too thick. I've already bent the cage as far as I can, now I'm going to order some 1mm and .5mm thermal pads and swap them out. That should fix it. For now I just loaded up the AS5, it was really my only choice. Running Crysis doesn't break 80, but OCCT still hits 100 eventually. Desktop users have reported temps as high as 115 from gaming, so I'm not as worried.

 

Remove thermal pads from memory chips and substitute them with thermal compound. This will solve Your MemIO temps problem.

 

The pads that are propping up the heatsink are mainly the ones on the regs. It's still a good idea, and would help a little. Maybe once my MX-3 gets here I'll try it.

 

There's a post around here about someone trying that method and burning out his memory, he may have done a bad job. Remember MemIO is not the memory.

 

Likely similar to G73's HD5870M. The chip/die is higher than the memory, hence why the thermal pads are needed.

 

i bought this to replace the thermal pads on ram/chips

for the lower chips i reused the thicker pads... none of my GPU sensors go above 78C using furmark

did not see this initially, thanks for clearing that up!
however why would it be so different than the gpu core? slightly confusing.

nice results, that's one heck of an overclock at 1.15v. Treat that card with care and don't let it overheat so much, it's a killer!

 

Even on desktop cards going back to the HD48xx series, the MemIO has run as the hottest part of the GPU. It's also located on the side the heatsink isn't making the best contact with. Once I get my pads I think everything will be fixed.

 

Why would you want to undervolt so much on such high overclocks? Wouldn't that make the GPU unstable? I can understand undervolt on 2D/Battery/IDLE.

 

thats why he tested to see if it was stable.
he's attempting to be as efficient as possible. extra energy used once 100% stable is wasted as heat.

also johnny, the pad material i linked is quite squishy (silicone based?) and will compress quite a bit if ncessary.

 

Exactly what xeroxide said, if it passes ATiTool and OCCT, it's stable. Undervolting for idle is a good idea for battery life, but under load: less voltage = much less heat and power draw. Once I put an extreme CPU in, I'm going to need to conserve on heat and power where I can. Even if you aren't overclocking there's zero downside to this, only benefits.

They are, but they're really cheap. They rip apart if you touch them and they're double stacked in places, so they need to be replaced.

 

I run my card at 1,02V @ 840/1050 24/7 (the laptop is loaded 8 threads + GPU even when I'm not at home, I run BOINC when idle, I don't turn it off), heatpads replaced by thermal compound, copper mod & bottom cover hole drilling mod ( http://forum.notebookreview.com/msi/457412-msi-gx740-5870-1gb-ddr5-116.html post #1160). Furmark tops at 82*C (that's the MemIO). I have never experienced a system freeze or BSOD in 6 months of use.

 

What voltage does it say in the drop down menu in the AMD Clock Tool? As far as I know, voltage steps in 0.05v are only allowed.

 

I have been using RBE to edit the vBIOS voltage table [as well as any other setting of my card, EVER - and it DOES work, because with the same clocks and the same ambient temperatures I noticed different (read: lower) temps while lowering voltage by only 0,01V]. I don't play with software tools .

 

I know you flashed the voltages. AMD Clock Tool only reads the current voltage being used. When you click the voltage table drop-down menu in the tool, it will only show what voltage the chip is actually running at. It doesn't give you the option to change anything.

So, I'm curious whether it lists 1.02 or 1.05 for you.

 

It reads VDDC: 1.05V.

It also shows 'Supported VDDC Voltages: 0.8000V; 1.0200V; 0.9000V', which are exactly the same values that I've set in RBE and I've flashed to my card. I hope this helps.