Tests will be performed across AMD and Intel systems, utilizing an Athlon64/nForce3-250, and Intel Prescott/Canterwood platforms. Both systems’ CPU will be liquid cooled. Our AMD test rig will feature Cool-Cases CF-1 Rev.2 CPU-Kuhler kindly supplied by Watercooling.de
and the Intel system will feature Danger Den
's venerable RBX CPU-block, and their Z-chipset will cool the 875-NB. Both water cooling systems will use the Hydor L-40 pump, and DD double heater core.Test System AMD.CPU -
TIM - Arctic Silver AS5
Cooling - Cool-Cases CF1 Rev.2 waterblock. Hydor L-40 (2800l/h 230cm max height) Danger-Den double heater core, 2x120mm/90-CFM Sunon's, Danger Den Bay Reservoir, 1/2" > 1/8" (CF-2) > 1/2".
Motherboard - Chaintech Zentith ZNF3-250 (the BIOSless version)
Graphics - Sapphire X800Pro (256MB, Catalsyt 4.6)
HDD - Maxtor DiamondMax Plus 9 (SATA150 120GB)
RAM - Corsair Twin-X1024 3200XLPRO (2.8V DIMM's 1,2)
PSU - OCZ Technology Power Stream OCZ420ADJ 420W
Case - TTGI-USA TT-201T3 and TT-561
Software - WndowsXP SP1, CPUZ, H'oda WCPUID, Podien CPUCool, DigiDoc, Sandra SiSoftware 2004 SP1/SP2, Aida32, Sciencemark 2.0, Hexus PiFast, 3DMark2001SE 330, 3DMark03, Gun Metal Benchmark, Aquamark 3, [H]ardOCP UTK3 Benchmark.
This is a fairly new Athlon64 3400+, unfortunately I chose the Chaintech Zenith ZNF3-250 as a Socket-754 nForce3-250 platform, based on the success of their nForce3-150 motherboard. The ZNF3-250 released almost 90-days ago, still lacks a BIOS upgrade providing multiplier access. The board's design leaves much to be desired, and not appropriate for mention here. Suffice it to say I've been unable to explore the 3200XLPRO's full potential, nor exploit the maximum performance from my A64 3400 CAAOC with the ZNF3-250. On Chaintech's behalf they did call me personally to tell me their still working on a BIOS. To date the best I could squeeze out of the 3200XLPRO on the Chaintech was an unstable 220FSB at 200MHz memory speed, for 2484MHz. The memory holds CL2 regardless of frequency.
From a prima facie perspective Low Latency would be a promising attribute, however; overclocking the A64 3400 on the ZNF3-250, I attained higher frequencies (235FSB) using Corsair's Twin-X4000, from its 3-4-4-8 timings? These higher latencies, in concert with the A64 has given me my highest 3DMark2001SE score to date, seen below.
This surprised me, as I was unable to manually adjust the 3200XLPRO to maintain CL3-4-4-8, nor did SPD settings allow me to surpass 220FSB on the ZNF3-250. In fact at 220FSB the 3200XLPRO wasn't stable in many tests, which is why I've omitted the A64 results at 220FSB. Onto the tests.
Since I've begun water-cooling I place H20 related hardware (pump/radiator) outside of the case, and leave the side panels off. Even when air-cooling I've kept the side panels off my PC's for the last three years. This has resulted in a 10C to 20C drop in overall system temps, and translates to a drop of 5C to 15C on-die CPU temps. This method is not for everyone, and not the most attractive. As a hobbyist it's a small sacrifice to effectively double the life of my processor. Some have critiqued my methods, however; I find very few PC-cases are sealed so well as to be air-tight. Nor have I seen many cases ventilated effectively enough to creating a vacuum. The simplicity of Thermal dynamics actually confounds many of us Enthusiast's. The hypothesis is sound. Expose the components to the open air, and allow Thermal Equilibrium to do its thing. I've included the photo below to exemplify my test system orientation.
At default speeds (200FSB) Corsair's Plug and Frag automatically adjusts timings for CL2-2-2-5 every time. No need for manual adjustments. I inserted the memory in DIMMs 1 and 2, and as seen below CPU-Z verified the timings just as Corsair specifies.
I tested all possible DIMM combinations, finding no change in performance. For the first AMD benchmark I used the Synthetic, SiSoftware Sandra memory benchmark in Buffered mode. I ran the Athlon64 3400 at default speed (11x 200FSB = 2209MHz) setting the bus multiplier option known as Hyper Transport at 4X. This is important; there are several settings for HT, and 4X results in an 800MHz Front Side Bus speed. The screenshot below exemplifies 3200XLPRO at 200FSB, 4xHT, CL2-2-2-5 2CL 2CMD, taking advantage of Plug and Frag simply means running from SPD (Serial Presence Detect).
Our next Sandra memory Benchmark run in Buffered mode, will exemplify the use of AMD's Hyper Transport. I raised HT-4X to HT 5X which results in a 1005MHz bus speed. The remainder of these tests will be run at HT-5x unless otherwise specified.
This is very respectable bandwidth for PC3200, and is consistent with most other article's utilizing an nForce3-250 chipset and Athlon64 3400, such as this review at PCStats.
Adia32 Benchmark READ 200FSB (2209MHz) SPD CL2-2-2-5 2CMD.
Aida32 WRITE 200FSB (2209MHz) SPD CL2-2-2-5 2CMD.
ScienceMark 2.0 Membench. 200FSB (2209MHz) SPD CL2-2-2-5 2CMD.
HeuxPiFast Challange 200FSB (2209MHz) SPD CL2-2-2-5 2CMD.
3DMark2001SE 340 200FSB (2209MHz) SPD CL2-2-2-5 2CMD.
3DMark2001 results (1024x768) 200FSB (2209MHz) SPD CL2-2-2-5 2CMD.
3DMark03 200FSB (2209MHz) SPD CL2-2-2-5 2CMD.
Aquamark 3 200FSB (2209MHz) SPD CL2-2-2-5 2CMD.
[H]ardOCP UTK3 Bench - Low (1024x768) 200FSB (2209MHz) SPD CL2-2-2-5 2CMD.
[H]ardOCP UTK3 Bench - CPU (1024x768) 200FSB (2209MHz) SPD CL2-2-2-5 2CMD.
Unfortunately I was unable to do much overclocking with Corsair's 3200XLPRO, for the reasons mentioned earlier. This was not due to memory limitations, but attributable to the test platform.