MSI 6950 TFIII PE/OC- testing w/ CL3P20

Its been a while since I have picked up an ATI GPU for play..these cards caught my eye since I first saw them in Vegas for MOA 2010 semi-finals. I fancy the idea of having a solid air cooled GPU for benching..as I dont always feel like juggling more than 1x pot when using DICE [lazy..ok ].

I have just begun digging into this GPU lately to get the feel of how it responds to voltage increases/temps. This particular model, did not unlock to a 1500 SP's via flashing or editing the BIOS. I plan on trying for a full unlock with a 6970 Lightning BIOS [as it too supports triple voltage control]. *More on that later





For starters:

Test Rig: [pic above]
Asus Maximus III Gene
Intel i5 660 ES
2x2 G.Skill Pi's
TX650 - GPU testing on air
TPQ1200 - GPU testing on water/DICE..LN?


Now this is the hwbot section..so I figured I would just summarize a lot of the stock testing, so I didnt hear all the ..

Quick tests at stock voltage:

I tested stock clocks under load with Kombustor- 64c max load temp after 17min .. needless to say- the stock cooler on this thing is . With a little adjustment to the temp curve in AFE, I was able to hold the GPU to 54c max load temp after 17min test again

Kombustor run on stock clocks/volts/fan profile

*Testing with Kombustor and Furmark was short lived, as OCP [of the current limiter on the software..not sure] would prevent clock speeds over 875mhz from running. It was on to 3D mark testing from here...


Stock OC limits:

GPUv stock @ 1.1v
GPU core limit- 925mhz
vMEM limit- 1330mhz

load temps with stock fan profile- 64-66c
load temps with CL3P's fan profile- 52-55c

Stock GPU- 3D'05 : 29k
Stock GPU- 3D'06 : 21k



A look at the PCB Layout:

input phases..for vmem and vtt [input at bottom of pic/output phases at top of pic] *can add 1x 16v cap for input


input phases for GPUv

output phases for GPUv

--------

**Now Before further testing could occur past ~980mhz core, I had to perform OCP mod for GPUv. At this time I also did a few other things.. and why not right..? 'You've' already got the "hood" open..might as well wrap up a few other loose ends I will just say that this is the fastest and easiest GPU mod/s I have ever done.. it literally took longer to get the stock cooler off, than it did to perform the following:




Mod Summary...Short and sweet!:

GPUv OCP for total and per phase



vMEM OCP



* As well, I applied MX4 TIM and a simple 'washer-mod' for some more bolt tension on the core.


-------



Overclock testing on stock cooler:

WOW..I actually ran into PSU limitations before running out of room on the voltage sliders, or hitting temp limitations! With core speed set to 1050mhz the 12v inputs were drooping to 11.72v..and the GPU was for the first time, starting to make audible squeal. I am very impressed/surprised that the cooler on this GPU was able to hold better than 63c, with GPUv at 1.32v and the above core speed! Mem and VTT were also increased a bit.. have a look below....

OC limits due to PSU:

GPUv stock @ 1.32v
GPU core limit- 1050mhz
vMEM- +.02v
vMEM limit- 1410mhz
AUXv or VTTv- +.02v

load temps with CL3P's fan profile- 61-64c

3D Mark '05 - 30k


3D Mark '06 - 23k


Best Vantage run on air so far: P18k


3D Mark '11- 5418 @ 1050mhz core, 1410mhz memory 1.35v core/+2mv mem in AFE


OC'd Just Cause 2 benchmark testing @ 1075mhz core: 8x AA, 8x AF @ 1680x1050

compare to GTX470 on water @ 900mhz core:



3D Mark '11-5356



Not too much difference going on between those pics..? But, what you cant see is the difference in temps and cooling required for that performance

*Here what cooled the GTX470 for it to be able to run 900mhz core..





That is:
Zalman Reserator- pump removed
PA120.3 in push/pull
6x Comair Rotron 24v, 130mm fans @ 13.4v
MCP355 with Petra's top
GTX470 temps- idle @ 34c / load @ 42c = 8c temp delta




That in mind.. its even harder to believe this is doable on a stock cooler..all I can say is-> to MSI for giving us all a cooler that can finally hold up to some god-awful punishment and blatant voltage abuse!! Ive never been able to confidently take a card to its voltage slider limit on a stock cooler..and have it do anything other than sizzle!!

With the power phases still strong at 1.32v load and +1000mhz core..I am anxious to see what kind of performance gains I will be able to squeeze out of a 20-30c drop on load temps..


---------

next up...







Preparation..

Removing the stock cooler is a cinch... 4x screws/standard drive..and done! All of the VRM and RAM cooling plate stays in-tact.. : ...its a beautiful thing! I 'helped' MSI's design a bit..just because I cant help but to add my ..

*Under the VRM/RAM cooling plate, there is thermal tape for contact in obvious places where the VRM's and RAM IC's contact the cooler.. The issue I have with it.. is its "tape". Simply put -> It covers everything... not just its intended target. Call me crazy/OCD..yeah ok, I wouldnt be able to sleep at night knowing this "bubble-gum" is laying all over the power phases and memory like a heating blanket....

So... I used a razor and cut away all of the thermal tape in between where it would contact the RAM IC's and VRM's...so that only the size/shape/amount needed for good contact was left on the cooler. [pics of this to come dont worry]

Heres some shots of the breakdown and 'bolt-up'

'improved' cooling potential..for the lulz


naked core


Installed..and ready for ..to fire

-----
post continued on pg2. here
-----
Edited by CL3P20 - 6/10/11 at 8:53am

Gonna be fitting it up with an MCW60 and an army of Zalman heatsinks soon. First task is going to be seeing if the input voltage is still drooping once its fed by my TPQ1200 ..if so, I will be adding some additional capacitors to the input phases.

*Inductance seems good, voltage from idle to load is very stable on the output side. The only reason why I stopped raising core speed was due to the PSU starting to clip ..I dont have a way to monitor current draw from the GPU alone..but reviews online estimate around 450W between 1001-1100mhz core speed [assumed voltage 1.3-1.4v]. No wonder the TX650 was starting to give me issues

Lots going on at the moment- Updates to come soon

**off topic- the Ballistix I got for the mem comp are a total bust.. high density..single sided..and giving me issues in every way!

*updated OP with stock and OC'd 3D Mark '05 and '06 runs, as well as SS for temp verification in 17min Kombustor run.