Originally posted by: Blain
You're kidding right? Look down the left side at the amounts used. :roll:
Everything about the setup stayed the same except the amounts of memory and the FSB when I OC'd 3%. It's all an "apples to apples" comparison.
He's saying the benchmarks aren't doing anything that will specifically test RAM differences. I've explained already that a RAM deficiency will not show up first in framerates, that means that using framerates as a benchmarking method is a poor choice.
Unfortunately subjective 'feel' IS where RAM will show up first, and there aren't very many ways to translate that into objective numbers.
Level load times is about the most objective thing there, and when I tested that for BFV (about a year ago) the numbers looked like the attached table. Even though the average FPS looks GREAT for the 256MB case, the actual game was virtually UNPLAYABLE because of all the pausing and such. The benchmarks show it some on minimum FPS, but reality was WAY worse than just a slightly lower minimum FPS with a good average. WAY WORSE.
Also, while 512MB was playable after you had been in the game a few minutes, the first minute or so was pretty choppy. So choppy it would be difficult to like load the level, get in a helicopter and expect to be able to fly it without crashing. These are things that affect gameplay significantly, but are not shown in the numbers.
Believe what you want, but the reality is that the reason people don't provide hard numbers for memory differences is because hard numbers for memory differences mean very little. They are not an accurate representation of real world performance.
It's like if you were presented with benchmarks comparing different CPUs with a 5900XT in DOOM 3 at 1600x1200. WOuld you trust the numbers from that benchmark? Of course not, it won't be sensitive to differences in CPU becuase the CPU has no problem feeding the 5900XT to it's maximum capability. Similarly, normal benchmarking simply doesn't show RAM size limitations.
By the way these were with a AXP at 2200 MHz / 200 FSB / 5900xt. Also, I had truncated the maximum frame rate at 80FPS, because I was looking for changes in performance I would actually see, not changes that dropped momentarily very high framerates that were well over my refresh rate. I don't care about framerate changes from 150 FPS to 120 FPS, so when I do benchmarks I filter out garbage like that to increase the SNR in the area I actually care about.