Plastic stock cooling system housing removes by a slight squeeze on both side of it:
Radiator of the cooling system is fastened with four screws. Unscrew them to take the rad off:
Stock system’s heatsink will later be used as an additional radiator. Just take it off for now:
Distances between the corner holes around the GPU — 58.4 mm (2.3″).
1 mm thermal pads between memory chips, power transistors and stock cooling system heatsink are used as a thermal interface. The pads are of good quality, therefore, there is no need to change them.
A closer look of the board:
GPU 4-phase power subsystem is located on the right part of the card:
Voltage controller NCP5388 is located on the back side of the graphic card:
The controller does not support I2C control, however, thanks to NVIDIA engineers we now can change VID voltage via SMBus.
On top of the controller there is memory power phase:
Memory chips made by Samsung have the following marking K4G10325FE-HC05:
Technical specifications of memory chips allow them to operate at 4000 MHz frequency:
|Small Classification||GDDR5 SDRAM|
|Density, Refresh||1G, 8K/32ms|
|Interface, VDD, VDDQ||POD_15, 1.5V, 1.5V|
|Package||170FBGA (Halogen Free & Lead Free)|
|Temp, Power||Commercial Normal|
|Speed ( Wafer/Chip Biz/BGD : 00 )||0.5ns (4.0Gbps)|
Increasing memory voltage can end up real bad — one phase apparently is just enough to power ten memory chips of DDR5 with a total of 1280 Mb. Extreme guys can easily find some volt mods on the Internet, here is just a small recommendation for them — you’ll need to use extreme cooling (LN2) to gain success in it.
Voltage measurement points can be found on the back side of the board:
According to multimetre there is 1.54 V on memory chips.
Now, after all details are revealed, it is time to go with alternative cooling way.