Clevo 7970M - How to raise power limit (power play table regedit)

So on my quest to make a spare 7970M in a spare testing laptop draw as much power and heat as it could for testing mods I found that beyond a certain point, no matter the extra clocks or voltage it would not draw more than 160W from the wall, and maximum temperatures or frame rates in the Firestrike stress test I was using were not changing.

CORE (set/reported)	MEM	VID (core)	Max Temp	Wall Draw
925	1350	1.000V	64 C	~133 W
925	1350	1.100V	71 C	~151 W
1000	1350	1.100V	73 C	~154 W
1050	1350	1.100V	72 C	~155 W
1100	1350	1.125V	73 C	~156W
1100	1450	1.125V	74 C	~157W

I loaded a vbios into an old editor tool and saw a 84W TDP power limit. There is no overdrive possible with the 7970M. Trying to alter this the various vbios editors either didn't work or bricked, since they wouldn't work with UEFI (as this laptop is set up with).

I looked into powerplay table editing and found this method worked.


1: open regedit, find the exact key. If convention is followed, 0000 would be master card and 0001 the slave if present (this laptop has had two AMD cards in it previously for testing/flashing)
2. double click the key containing the soft power play table
3. find the hex value corresponding to the TDP wattage (84 decimal = 54 in hex)
4: change it and reboot.

I changed 54 (84W) to 74 (116W) and saw wall power draw increase to up to 180W.
1100/1400/1.125V would also crash 3dmark with no other change (but system would remain stable) indicating now, the voltage was insufficient to maintain the *actual* clocks... I suspect the clocks reported by afterburner are inaccurate as they are more like a "requested" clock and actual power limit throttling is invisible to these utilities. (Furmark also supposedly runs at a constant 850mhz... suspicious)

Those 3dmark crashes resolved by raising VID to 1.175V. After doing that and keeping the high overclock, the laptop would blackscreen crash (at about 78C and rising) requiring reboot after approx 30 second worth of stress test (I assume VRM overheat).

Just to check, I edited back to 54, and wall draw was back to topping at 160W and not crashing at 1.125V any more.

For the record; it's a P370EM with a Clevo 7970M, stock vbios 015.024, stock specs 850/1200/1.000V, 84W power limit, no enduro.

DISCLAIMER: these seem to be pretty weak cards with little headroom in the VRMs so be conservative! I have just tested this with an carefully cut and contact-tested fresh thermal pads (and liquid metal on the core) and they seem to be cutting out with only 30W of extra power so pushing lots of extra heat through old dry VRM thermal pads is a recipe for the Magic Smoke.

 

Can you post some benchmarks? I'm really interested in the results.

 

So the only way to check if graphics is holding the "clocks" is to compare results from benchmarks. For my clevo p170em I've used VBE7 to change Power limit(100W) and my 7970m go stable @ max with 1090/1600/1.1v(enduro).
3dmark P score
FS score

 

Good lord, that is quite an impressive benchmark you did there. That's beating an stock 880M and is close to a 970M.

 

I'm aware there's plenty of modded vbios out there that have raised PL already but this isn't requiring a vbios flash, and it works with UEFI.

Benches incoming.

The effectiveness of replacing some of the thermal pad mass over the VRM area with chunks of copper is one of the mods I was considering.

I think hwinfo may report actual clocks: had a quick look just then and it was reporting ~700mhz under furmark compared to the usual "850mhz" in AB, GPUz etc

 

Results speak for themselves

https://www.3dmark.com/compare/fs/14363370/fs/14363117#


Left: 1050MHz core, 1500MHz memory, 1.125V, 116W power limit
Right: 1100MHz core, 1500MHz memory, 1.150V, 84W power limit

lower clocks + higher power limit = higher score

The reason for lower clock/volt on the core was 1100MHz @ 1.150V would not complete with the higher power limit

 

It is kinda scary to see a HD 7970M being on par with a stock 970M

 

ok so apparently all my top 7970m benches got removed from hwbot. I remember I had 9.8k 3dm11 GPU score running the core at 1.225V and over 1.2 GHz.

That was running a vBIOS which allowed software overvolting. Nospheratu modded it to enable the voltage slider in Sapphire Trixx. I then removed the power limit (set to 160W) with VBE7. I also manually trippled the 3D battery clocks, which more than doubled FPS on battery.

I never had any issues with power delivery on the card, but I did modify it to have the stronger Dell 7970m version low side power FETs.

I could not upload the vBIOS to the website so I linked it here:

http://www.mediafire.com/file/gemlddpn1mbgka6/C7970m_unlocked.rom

This vBIOS is meant for Clevo enduro cards. I think I could mod a non-enduro if someone needs it. It may or may not work right on Dell cards (I think early vBIOS were compatible between Dell and Clevo, but later vBIOS had problems). Dell cards overclock for crap anyway though.

 

Seein that HD7970M cards die a lot lately, isn't this a bit risky?

 

I thought it was strange that apparently today I got "world records" on hwbot for FS, FSE, and TS for a single 7970M (with non-extreme CPU) with a bone stock heatsink with liquid metal on the core, and I have put no effort into tweaking or finding maximum anything.

 

Unfortunately, it's true that the cards were dying. It seems to me that even on our forum someone wrote that the problem concerns the "hardware" power section, I think...
After 3 years my 7970m also joined the "club" so i go with 8970m(clevo 4GB). This card was running in my clevo @ 975/1425/1.05v instead of 1000/1500/1.05v from 7970m for daily use.
And currently i have a cursed by all 1060 vortex, which works on stock vbios on p170em with perma bios.
Btw I even found old recordings(fps was little higher without recording):

 

Would be interested how a 8970M modded would perform compared to the 7970M

 

7970m vs 8970m
2gb(hynix) - 4gb(samsung)
850/1200/1.05 stock - 900/1200/1.05 stock
Daily use with max perf @ 1.05v
1000/1500 - 975/1425
Until today i don't know what will by the max core/mem/voltage timing for 8970 because i bought 1060 vortex and AMD cards... well, they just lie in the "warehouse".
I would love to go forward with 8970m oc but changing the cooling from amd to nvidia and vice versa(paste, thermal pads, copper plate) demotivates me very much...
Besides, 1060m has quite a decent gaming performance for me(1080p), as well as working temperature, so i gave up.I will wait for next gen AMD mxm.

 

Some more playing around.

Raising the power limit beyond 116W (PP table "74") to 132W ("84") lets it draw over 200W from the wall in firestrike GT1, where the usual average is ~180W. Thing is, 1050mhz/1125mV on the core is right on the edge. A hot day and ambient a few C higher and it'll black screen.

Backing off a little bit, 1025core/1450mem/1110mV is stable and passes Firestrike stress test all day long. (average wall draw ~173W, max temp 86C). With the higher power limit I actually hit 7000 GPU score in FS: https://www.3dmark.com/fs/14454779

 

FYI I couldn't push this card beyond 1.15V even when watercooled with a core maximum of 54C at 1140mhz... and my copper tube did contact the vrm/memory heatsink.