Test Setup & Overclocking
Berry has compared the 4 kits he had in his possession on the system outlined below. Here are his findings when it comes to overclockability:
Berry’s Test Setup
|CPU ||A64 3700+ (San Diego Core, E4)|
|Mainboard ||DFI RDX200 CF-DR (Bios C12)|
|Video card ||ATI X800XT PE|
|Other ||Silverstone TJ06|
Maxtor 160GB SATA
All tests were run with the VDIMM set to 2.75v max, as there have been user stories of Micron based modules dying because of 2.8v. I was able to test for compatibility with the modules on MSI K8N Neo2 and DFI NF4 Ultra-D; my Shuttle SK83 did have issues with Infineon/Micron based modules this is not Team Group's fault but do check up on compatibility with your board. PC4000 TXDR2048M500HC3DC 3-4-4-8 Samsung UCCC
At 200Mhz they run smoothly at CL2.5 3-3-7, and at rated speeds of CL3 4-4-8 275Mhz proved stable (although I had to change the advanced timings in the DFI RD board’s bios to 7-15-1-2-2-4). I also had the chance to test these modules in another DFI board, the NF4 Ultra-D, 285-290Mhz was possible at CL3 4-4-8. On the RDX200 motherboard they were stable up to 289Mhz.
PC4000 TXDR2048M500HC3DC 3-3-3-8 Micron based
At 200Mhz these run at CL2.5 4-4-8 using SPD but CL2.5 3-3-7 poses no problem. I had seen tests of Micron based modules that ran at CL2 2-2-8 but I had no such luck with this set. At 275Mhz timings were CL3-3-3-8 (advanced setting: 7-15-1-2-1-2).
Playing around with the Drive Strength option in the BIOS seems to have quite a bit of effect on the overclocking of this kit, current maximum overclocked lands on 287Mhz at CL3 3-3-3-8
PC3200 TXDR2048M400HC2DC 2-3-3-5 Infineon BE5
The modules are binned to run at tight timings, meaning CL2 3-2-5 and they do exactly that at 200Mhz. Overclocking with this setting went ok until 209Mhz. 217Mhz was obtained by loosing settings to CL2.5 3-2-7. The maximum I obtained was 233Mhz@CL3 3-2-7, not too shocking, but these modules are meant to run at LL (low latency) at PC3200 speeds.
PC3200 TXDR2048M400HC25DC 2,5-4-4-8 Team labeled
SPD detects CL2.5 4-4-8 at 200Mhz they don’t promise much, but do run smoothly at CL2.5 3-3-8 up to around 210Mhz. The maximum overclocked was 215Mhz@CL3 4-3-8.
While Berry had the chance to try out more modules than me, I had the opportunity to test the Infineon CE6 based modules.
PC4000 TXDR2048M500HC25DC 2.5-3-3-6 Infineon CE6
JNav89GT's AMD Test Setup
|CPU ||AMD Athlon 4000+ (San Diege, E4)|
|Mainboard ||DFI SLI-DR NF4 Ultra|
|Video ||ATI Radeon X850XT PE|
|Cooling ||Modified Prometeia Mach1 R402a by Chilly1|
|Power Supply ||Zippy 460W|
|Operating System ||Windows XP SP2|
On 2 of my testing platforms I was able to attain a rock solid 270mhz at 3-3-2-8. Stock timings of 2.5-3-3-8 were able to run 255, but in all honesty Cas latency really is not as strong of a predictor of performance for the A64 as the tightened trp can provide. Ideally we could run 2-2-2-6 timings, but no IC has offered us this performance at higher frequencies in a 64x8 configuration that will allow 1GB modules or 2GB dual channel kits. At 3-3-2-8, the timings at 270 MHz are relatively tight and can provide some excellent performance. Past 270 MHz I could run and bench 2d applications with total stability up to 278 MHz, but past 270mhz I could not retain any appreciable 3d stability. This is not Team Groups fault, it is an inherent quality of the Infineon module they are using, and similar 3d issues can be found in threads regarding Mushkin, G.Skill, and OCZ’s use of these IC’s as well.
PC4000 TXDR2048M500HC3DC 3-3-3-8 Micron based
After playing with the CE6’s for a time the Micron based PC4000 arrived (the same set Berry tested). With default timings of 3-3-3-8 at 250Mhz, the initial impression one might get would be these are somehow inferior modules. While each kit can certainly be different, in my testing the Microns edged out the Infineon CE6’s slightly in usable frequency. I was able to run and bench at 280MHz at 3-3-3-8 timings at 2.8v (any higher I was concerned with premature IC death as noticed with Crucial Ballistixs). However, extended 3d sessions did begin to error/lockup even though 2d/memtest programs were totally clean. I was able to run 275mhz error free for 24/7 operation. I feel that with some extended time and bios tuning I could get 280MHz to work well at 3d for extended periods. I will say that this Micron did take some time to tune though. Honestly, I’m not sure it would be any more difficult to use, for any other user though, but my bios tune changed quite a bit from the CE6’s. My ease in running the CE6 was primarily a result of my prior experience with some Mushkin Redline XP4000 2GB modules that I had the opportunity to test as well, so I already had a good base tune from which to work.
For our battery of benchmarks Berry has run a series of tests of different speeds, the CPU multiplier was left at 11x:
200Mhz @ tightest timings for each set
PC3200 kits compared at different maximum overclocks
250Mhz & 275Mhz @ tightest timings for the PC4000 sets
Everest Memory Benchmark
Sisoft Sandra Memory Benchmark
Cinemark 2000 CPU benchmark
SuperPi 1Mb / 32Mb