64-Bit: Benchmarks With 4 GB To 16 GB
Part 1: Gaming
Older, single-threaded, 32-bit applications benefit very little from large expansions of system memory when only one application is running. We know from previous benchmarks and articles that 32-bit games running in 64-bit environments don't benefit much (sometimes at all) from system memory sizes above 4 GB, at least not in terms of frames per second.
Another rule is that an increase in performance can only be seen when a) less data is transferred between the RAM and the swap file or the data is transferred faster and b) when the writing and reading of temporary files is extremely fast. This brings us back to our earlier 32-bit discussion. Using a RAM disk can shorten load times significantly, but it doesn't really impact the actual frames per second. Also, giving the graphics card more allocable memory is likely to impact your subjective perceptions. Unfortunately, this acceleration cannot easily be quantified.
We tested the games one more time. Using the 4 GB of RAM test results as a reference point for the 100% mark, we calculated the individual relative percentage values using higher memory capacities to get a better overall comparison.
So far, so annoying. Other than the 64-bit client of Half-Life 2 with the Bombast mod installed, hardly any game sees an increase in frames per second using more RAM. The increased amount of system memory allocable by the graphics card noticeably decreases the number of visual artifacts in GTA IV. Load times could be shortened if the temporary directories were relocated to a RAM disk.
Here, however, you have to weigh what is more meaningful: running the entire system equally fast from an SSD or creating a RAM disk that requires at least 12 GB of RAM in order to be useful.
Part 2: Applications
We only expect to see performance increases when programs are able to allocate and, more importantly, need more memory space. Only then can you avoid the swap effect and the performance impact we described earlier. This can also be somewhat alleviated with an SSD. We therefore only expect a jump in performance when the amount of RAM needed is nearing the amount that is installed. In the following charts, we see two examples with measurable, though small, improvements.
The synthetic benchmarks don't show any significant performance advantage, so we decided not to create a lot of pointless charts with identical bars. The fact is that a good-sized RAM disk provides a better performance increase than having more memory once you cross a minimum threshold.
At this point, we want to refer to our 32-bit benchmark results, as they're also valid for the 64-bit equivalents. However, in terms of usability on multitasking systems, we'd say that 8 GB of RAM is the minimum, even when not factoring in system memory used by the graphics card. If necessary, 6 GB will do as an interim minimum for enthusiasts in triple-channel systems.