Warranty, RAM Cooling And Final Note
Vendors, Damage And Warranty
We’ve discussed damage to ICs and electromigration in the context of overclocking before, and we’ve mentioned vendor-advertised default values for RAM performance.
There are many RAM vendors, but not all manufacture the ICs they use to build modules, and there are only a handful of memory manufacturers, including Samsung and SK hynix. The kit vendors purchase memory ICs and solder them onto PCBs. Different vendors may list different values for voltages and timing latencies using the same chips.
Vendors test each chip in order to program the EEPROM on-board the module with voltage and timing latencies; it is the integrated memory controller and parameters set via the motherboard firmware that control the memory’s final voltage and timing in the system, so once power has been turned off, there is no way to check what parameters or voltage were applied to the memory, short of actual physical damage.
Often, the pre-programmed memory frequency and latency values on the DRAM’s EEPROM are lower than the actual advertised value; the memory will boot with conservative settings for compatibility with a wide variety of systems. From there, voltages, latencies and clock rates must be set via the motherboard firmware. This is by design from the vendor’s point of view, and will not violate the warranty.
Straying above the datasheet voltage value (1.5 or 1.65 or even 1.8 volts) or loosening the RAM timings leaves no trace on the chips themselves. In fact, the bottleneck for memory voltage is the IMC (e.g. 1.55 volts for Intel’s Sandy Bridge CPUs).
Keeping the above factors in mind, there are two major sets of values you should check before overclocking the memory chip:
- The maximum memory speed, clock rate, and voltage allowed by the CPU and/or the IMC. These values can be exceeded, but that really depends upon the silicon lottery, and there is a higher chance of causing physical damage.
- The maximum voltage and speed for the specific memory module (not the vendor’s advertised values). This requires a glance at the original memory module manufacturer’s datasheet.
There is some disagreement in the overclocking community about how much RAM benefits from active cooling. RAM doesn’t really perform too much better at cooler temperatures, but serious overclocking still heats up the memory chips. On the whole, unless a system is intended for extreme benchmarking or testing, passively cooling the memory chips is sufficient, though having an additional fan blowing air over the area certainly won’t hurt.
If you are already using a CPU liquid-cooled loop, putting in a water block for the RAM is relatively easy if that's the route you want to go. Alternatively, vendors have water cooling-ready memory packages, like the Kingston HyperX H2O series.
This brings us to heat spreaders and vendor-supplied heat sinks. It used to be that spreaders were almost purely aesthetic. There are also reports that certain vendors have, in the past, glued heat spreaders on to disguise the quality of the memory modules or ICs. But current-generation heat spreaders from reputable vendors do make a difference in operating temperatures, and when the same kit is provided in two different packages—one with a heat spreader and one without—going for the heat spreader makes technical sense.
A Final Note: “Overclocking” Memory vs. Regular Memory
Apart from the aesthetics and marketing, there is no specification difference between “overclocking” or “pro” or “gaming” memory and regular memory if the same IC is used. However, given the chip-to-chip variability of memory modules, vendors will often test and pre-select higher-performing modules/chips, label them as “overclocking” memory, perhaps add a heat spreader and, most important, set the pre-configured timings in the EEPROM to values optimal for overclocking, given a certain training regimen. This only applies to reputable vendors, though—any memory can be overclocked, and marketed as such.
MORE: Best Memory
MORE: All Memory Content