The photo above was taken for "eye candy" reasons, although as I began testing the NV-68 with Corsair XMS Xpert-3200 I did achieve 7.4GB/s bandwidth on this system running XPert at 11x250HTT. With the AMD on-die memory controller I ran SPD timing’s of 2-3-2-5, as opposed to my initial Intel P4/Xpert OC 15x250FSB (875 Asus) at 2-4-4-8 resulting in just 5.7GB/s bandwidth, but that's another review.

Test System
CPU	A64 3500+ (2210MHz (Winchester core 90nm)) Socket-939
Mainboard	DFI LANPARTY NF4 UT (BIOS 310)
Memory	G-Skill DDR600 2x256MB DC
Graphics	BFG 6800GT OC (370/1000MHz) NVIDIA 71.89 WINXP Drivers
Power Supply	Thermaltake Silent Pure Power 680
CASE	Thermaltake Kandalf aluminum Tower
Cooling	PolarFLO TT CPU waterblock, Danger Den NV-68, Hydor L-40II Pump, Danger Den Double Heater Core, 2x120mm/92CFM fans, 1/2"
Operating System	WindowsXP Home SP2

Test Methodology

Two thermistors running from the TTGI SF-610 were used to measure external (Ambient) temp, and a second was placed as close to the NV45 GPU as possible without adversely affecting mounting (contact between heatsink and GPU). Primarily GPU temperatures will primarily be determined by Riva Tuner 2.0 15.5 which monitors the NV45 on-die thermal diode. The nVidia on-die thermal diode seems to be fairly accurate, regardless thermistor placement is not an exact science. To record results including temps, frequencies etc. during IDLE, a combination of monitoring utilities will be captured in screenshots. For CPU temps and frequency the A64 on-die thermal diode will be monitored using ITE Smart Guardian. To indicate FSB speeds, and processor data CPU-Z will also be used. Finally for additional real-time clock speed WCPUCLK will be used. The screenshot below exemplifies how temps, and frequencies were recorded for this review.



For LOAD the combination of utilities will only vary slightly. CPU-Z will be used throughout, insofar as HTT and CPU speed these will remain constant running GSkill DDR600 (2x256MB) with the A64 3500+ at 9x300HTT = 2.7GHz on our DFI LANPARTY NF4 ULTRA-D.

GPU-Core-temp as indicated in Riva Tuner will be our primary source throughout.

GPU Ambient found at the very bottom of the RivaTuner GUI will be recorded also as this represents the current temperature of the PCB board.

The Room temp found in all charts is the recorded ambient temp using a thermistor from the TTGI SGF-610. This thermistor was placed just inside the case, which remained open to ensure there was no differential between Case and Ambient. This is crucial due to the size of the radiator.

To stress the BFG 6800GT GP and memory the program used will be Rthdribl v.1.2 (Real-Time High Dynamic Range Image-Based Lighting) just as the name implies the program demonstrates DirectX(R) 9.0, and Pixel-Shader 2.0. Most useful for our purposes is the drag-and-drop window so screenshots can be taken simultaneously with temperature monitoring/frequency GUI's.

Many H20 enthusiast's are now using double, even triple radiators most of which are mounted externally on their case. Our goal is to determine how well the NV-68 can remove a specific amount of heat from the card it's mated with. Whether the radiator is in the case or outside the case has no bearing on how well the waterblock itself "performs" this task. Obviously if the air surrounding the radiator is 35c then we know the waterblock cannot "cool" the GPU any lower, this process is dictated by thermal equilibrium. For those skeptical about the use of internal thermal diodes, a quote from the following article may help; "...the diode/motherboard combination can display temperatures that are off by up to 6C or so from the actual temperature. This error is constant, so it does not affect the usefulness of the internal diode as a test device. A change to the cooling system that changes the temperature reading by 3C IS a 3C change regardless of whether the diode shows it as 43 to 40, or 38 to 35, or 40 to 37. The diode may lie, but it always tells the same lie."


Testing in linear Single-feed setup



Testing in linear Dual-feed setup



The Thumbnails below each correlate to data found in the Temp Chart (Excel Spread Sheet) on the bottom of this page.

BFG running Stock-air cooler @ IDLE 370MHz Core/1000MHz Memory (CPU 9x300HTT = 2700MHz) LEFT/RIGHT - LOAD

BFG @ IDLE 370MHz Core/1000MHz Memory (CPU 9x300HTT = 2700MHz) Single Feed (PolarFLO TT) - NV-68 LEFT/RIGHT - Dual Feed

BFG @ LOAD 370MHz Core/1000MHz Memory (CPU 9x30HTT = 2700MHz) Single Feed (PolarFLO TT) - NV-68 LEFT/RIGHT - Dual Feed



Danger Den NV-68 Proprietary Cooling


For our proprietary cooling tests I replaced the PolarFLO TT for CPU cooling with Alphacool's Xtreme Pro Set and their NexXxos XP BOLD AMD 64. With the NV-68 on a proprietary system graphic card temps should improve significantly. As done previously we'll show screenshots taken running the BFG 6800GT at IDLE and then under LOAD using Rthdribl. The screenshots below exemplify the NV-68 on a proprietary system. All components used in the previous system have been left in place, with the primary feed now running from the radiator directly into the NV-68.

NV68 proprietary BFG @ IDLE 370MHz/1000MHz (CPU 9x300HTT = 2700MHz) LEFT/RIGHT = BFG @ LOAD