Performance

Stability test:

• MemTest+ 1.60 30x test #5
• Prime95 24.13 Blend 2h and Large 30min
• 3DMark01 Standard
• Quake4 Custom Timedemo

The participants of this test:

• Mushkin XP4000 with Winbond UTT-CH5 (2x512 MB)
• TwinMOS Twister Pro PC3200 with Samsung TCCD (2x512 MB)
• G.Skill 4000 2GBHZ with Samsung UCCC (2x1024 MB)
• Corsair PC3500 rev1.1 with Winbond BH-5 (2x256 MB)

2-2-2-5-7 - 3.xV
Without any cooling, the Mushkin XP4000 does not run stable with a higher voltage than 3.1V.
The difference between cooling with a fan and cooling with the RAM Freezers is four MHz and 5.6°C. First that looks like a small difference, but at such high frequencies and low latencies it is quite big.
If you want the same frequency with air cooling you will have to give more juice (0.2V - 0.3V), which means a shorter lifetime and Mushkin's warranty does not cover such high voltage.
So we see that UTT-CH5 reacts well to cooling.


2.5-3-3-5-7 - 2.9V
The TwinMOS Twister is an example of Samsung TCCD based memory. We saw an incredible advantage of 25 MHz to the standard heatspreaders and 11 MHz to air cooling.
It reached almost the same temperatures as the Mushkin XP4000 but we squeezed some more MHz out of the Samsung TCCD by cooling with the MIPS RAM Freezers.


3-4-4-8-7 - 2.6V
The Samsung UCCC chips are a good example for a setup where you don't need the RAM Freezers. The memory stays cool without cooling and overclocking does not increase with cooling either.


2-2-2-5-7 - 3.xV
With the old Corsair PC3500 Winbond BH-5, you can see almost the same results as with the Mushkin XP4000. Without cooling they stay a little bit cooler and overclock higher. But above 3.2V they will also start giving errors. We fired them with dangerous 3.85V. That's not recommended for 24/7 use, but there are some crazy folks out there, who bench at such high voltages. With the RAM Freezer we could run our stability test at unbelievable 276 MHz at tightest Timings. So if you want to have the last few MHz to bench, get the freezers.