AnandTech: DFI UT P35-T2R: Tweakers Rejoice!

Quad-Core^4

DFI UT P35-T2R Overclocking Benchmark Testbed
Processor	Intel Core 2 Quad QX6800 ES - G0 Stepping Quad Core, 2.93GHz, 8MB Unified Cache, 11x Multiplier, 1066FSB Intel Core 2 Duo X6800 Dual Core, 2.93GHz, 4MB Unified Cache, 11x Multiplier, 1066FSB
CPU Voltage	1.300V Base
Cooling	Various - Tuniq Tower, Swiftech Apogess GTX, Dual Rotary Cascade
Power Supply	OCZ ProXStream 1000W, PCP&C 1200W
Memory	OCZ Flex PC2-9200 (2x1GB, 4x1GB)
Memory Settings	4-4-4-12 (DDR2-1066)
Video Cards	NVIDIA 8800GTX Ultra, NVIDIA 7600GS, NVIDIA 8800 GTS 640MB
Video Drivers	NVIDIA 163.44
Hard Drive	Western Digital 7200RPM 250GB - WD2500KS
Optical Drives	Plextor PX-755A
Case	Cooler Master Stacker 830 Evo, Open Test Bed
BIOS	DFI 9/13
Operating System	Windows XP Professional SP2
.

Our BIOS settings for 450FSB with a quad-core CPU using 2GB of RAM

Parts used for this stability test

Qx6800 ES G0
OCZ ProXStream 100w
2GB OCZ Flex 9200 RAM
Tuniq Tower with additional Panaflo 120mm fan
WD2500 KS 250GB Hard Drive
Plextor 755a DVD/RW
NVIDIA 7600 GS

450FSB Quad-Core BIOS Settings
CPU Feature
Thermal Management Control	Disabled
PPM (EIST) Mode	Disabled
Limit CPUID MaxVal	Disabled
CIE Function	Disabled
Execute Disable Bit	Disabled
Virtualization Technology	Disabled
Core Multi-Processing	Enabled
Exist Setup Shutdown	Mode 2
CLOCK VC0 divider	AUTO
CPU Clock Ratio	8x
Target CPU Clock	3600MHz
CPU Clock	450FSB
Boot Up Clock	AUTO
DRAM Speed	266MHz/667MHz
Target DRAM Speed	DDR2-1125MHz
PCIE Clock	100MHz
Voltage Settings
CPU VID Control	1.39
CPU VID Special Add	AUTO
DRAM Voltage Control	2.1
SB 1.05V Voltage	1.07v
SB Core/CPU PLL Voltage	1.55v
NB Core Voltage	1.61
CPU VTT Voltage	1.60v
VCore Droop Control	Disabled
Clockgen Voltage Control	3.45v
GTL+ Buffers Strength	Strong
Host Slew Rate	Weak
GTL REF Voltage Control	Enabled
x CPU GTL1/3 REF Volt	115
x CPU GTL 0/2 REF Volt	114
x North Bridge GTL REF Volt	117
DRAM Timing
Enhance Data transmitting	FAST
Enhance Addressing	FAST
T2 Dispatch	Disabled
Channel 1 CLK Crossing Setting	More Aggressive
Channel 2 CLK Crossing Setting	More Aggressive
CH1CH2 Common CLK Crossing Setting	More Aggressive
CAS Latency Time (tCL)	5
RAS# to CAS# Delay (tRCD)	5
RAS# Precharge (tRP)	4
Precharge Delay (tRAS)	9
All Precharge to Act	4
REF to ACT Delay (tRFC)	30
Performance Level	6
Read delay phase adjust	AUTO
MCH ODT Latency	1
Write to PRE Delay (tWR)	12
Rank Write to Read (tWTR)	10
ACT to ACT Delay (tRRD)	3
Read to Write Delay (tRDWR)	8
Ranks Write to Write (tWRWR)	AUTO
Ranks Read to Read (tRDRD)	AUTO
Ranks Write to Read (tWRRD)	AUTO
Read CAS# Precharge (tRTP)	3
ALL PRE to Refresh	4
PCIE Slot Config	1X 1X
CPU Spread Spectrum	Disabled
PCIE Spread Spectrum	Disabled
SATA Spread Spectrum	Disabled

450FSB Stability Test

Water-Cooling

After attaining a decent 3.6GHz performance overclock on air-cooling, the decision was taken to move over to water-cooling for the CPU. Our QX6800 G0 ES is not one of the coolest running G0 CPUs around. Up to 3.6GHz the current G0 stepping quad-core CPUs all seem to be able to cope with less than 1.4V. However, cruising over 3.6GHz results in a non-linear VCore scaling process per CPU MHz increase. We found ourselves tapped out for further overhead with 100% CPU load tests. We were able to run most benchmarks at around 3.9GHz with the same air-cooling, with around 1.5Vcore to hold the CPU steady. Results placed on various forums have shown some users Prime95 stable for all software up to 3.8GHz with good CPUs using top-end air-cooling.

The following Water Cooling components were used for this test:

Swiftech Apogee GTX CPU waterblock
2x Petra Top Laing DDC Ultra water pumps in series
Swiftech Micro-Res
Thermochill PA120.3 Radiator, with 3x Panaflo 120mm fans
Danger Den 1/2" ID Tubing

This swap to water-cooling allowed us sufficient temperature overhead to run Prime95 type tests for 11 hours with speeds of 3.72GHz @ 465FSB on the 333 strap with 4GB of RAM, while holding 470FSB on the 266 strap and 2GB of memory stable at around 467FSB. A 5-7 FSB margin above Prime95 type testing is often usable for those who do not require absolute 100% load stability to find sleep at night.

Prime95 and Everest

We found that CPU scores in 3DMark06 using the DFI P35 T2R were higher than we were accustomed to on the acclaimed 975X chipset. We managed to push 4GB of RAM to 490FSB with water-cooling for a quick run of 3DMark06. The board at this point however was right on the ragged edge and would not run SuperPi 32M without a failure.