MSI 6950 TFIII PE/OC- testing w/ CL3P20 - Page 2

continued post:


For starters, I want to draw some attention to the GPU VTT phases and supply area on the PCB. Please keep in mind, I am adjusting all voltages on the GPU..and monitoring them under idle and load.






VTT and vMEM have droop.. noticeable at high clock speeds. For the 12v input VTT phase, +1ghz core speeds will cause droop -.25v to -.29v. For memory, it took 1410mhz to see the same level of droop occur on the 12v input phase.

Now these figures are all using a TX650 PSU still..which I have already stated, is dipping to 11.81v under load with max clocks on air. I intend to revisit these findings once the GPU is fed with the TPQ1.2kW, to see if the droop issue goes with the PSU or not.


Now.. for a quick temp check/comparison from stock air->my water loop




More to come shortly...

measure points are here..



Stock vmem is 1.6v idle
Stock vtt is 1.15v idle

yessir.. You can set mem voltage in AfterBurner.. I havent found a need to go beyond the +4mv limit its giving me yet..may be needed for 1500> ..although cold may do it too.

I am still hitting voltage limitations on the TFIII, so I havent gotten around to seeing if this one will tolerate -0 or not yet. Looks like a Frequency Switching mod is in order..I think Im at some-sort of 'supply limit' with clock speeds +1070mhz. There are virtually no gains in 3d testing over this speed, even though the gpu finishes the benchmark, with no arties. ...still working on this mod though, just located it and started testing recently.



Voltages on the GPU using TPQ1200 [using 1050mhz core/ 1370mhz mem]:

+12v input for GPUv- 12.02v idle, 11.99-12.00v load
+12v input for VTTv- 12.0v idle, 11.97v load
+12v input for vMEM- 12.02v idle, 11.97v load

GPUv= 1.32v idle, 1.33v load [fluctuation measured at the input to the inductors/mosfet output = -.2v under load]
VTTv= 1.17v idle, 1.17v load
vMEM= 1.62v idle, 1.61v load


Voltages on the GPU using TX650w [using 1050mhz core/ 1370mhz mem]:

+12v input for GPUv- 11.97v idle, 11.91-11.94v load
+12v input for VTTv- 11.94v idle, 11.89v load
+12v input for vMEM- 11.95v idle, 11.90v load

GPUv= 1.32v idle, 1.33v load [fluctuation measured at the input to the inductors/mosfet output = -.6v under load]
VTTv= 1.17v idle, 1.17v load
vMEM= 1.62v idle, 1.61v load


The low +12v inputs from the overloaded TX650 caused a LOT of squeal on the GPU's power circuitry.. I was convinced I was going to blow something if I tried to push clock speeds further. On the TPQ1200 - there was NO noise from the power circuitry on the GPU at ALL! IT WAS DEAD QUIET AT FULL LOAD!

I would really like to see if I can get this GPU over 1100mhz now..im just around the corner now, and temps are still under 46c load on water..even at 1.38v and 1080mhz core ..so I am hopeful.


More to come

Yeah..from what I see..its enough to use the 650W, when OC'ing this card to ~900mhz core.. but amperage demands on the +12v rail are brutal above this point, and you really need a strong PSU to keep the GPU fed and quiet.

* could game on the TX650 with this card at stock all year long.. but with heavy OC..its a different story altogether.

Performing GPUv mod and experimenting with 'switching frequency' mod as well [penciling only for now ]. Updates to come as soon as I make some headway. I would like to bench +1100mhz core, which will not be possible without these mods from what I am seeing.

Unfortunately I cannot find any records/posts of a Switching Freq. mod for this uP controller..but I have located the pin/resistor for the mod with a datasheet. The datasheet does not detail if raising or lowering resistance is required to raise the switching frequency.. so looks like lots of testing is in order.

->Still trying to locate a 6970 Lightning BIOS to try for a full unlock as well.. *hint*hint*


More to come soon-