Most Super 7 motherboards already have 1Meg of L2 cache, which is automatically L3 cache when a K6-3 is in the motherboard. At most, this extra cache costs $10.
Really? I didn't know Super 7 motherboards already have one meg of L2 cache. But your estimation of $10 isn't accurate; I just checked Price Watch, and an extra 512K of cache cost anywhere from $32 to $63. I'm not entirely familiar with L2 cache latencies on a Super 7 motherboard, but wouldn't a larger cache, by rule of thumb, necessarily be slightly slower? (Yes, another nitpick.)
L2 cache speed is very important in the K6 architecture. Just moving the L2 cache speed up to 112 MHz from 100 makes a jump from 29.1 to 31.0 in Winstone 98 performance (see cpu.simplenet.com cache speed is very important in the K6 architecture. Just moving the L2 cache speed up to 112 MHz from 100 makes a jump from 29.1 to 31.0 in Winstone 98 performance (see cpu.simplenet.com.
You are forgetting that the SDRAM might have also been overclocked. (Yet another nitpick.)
All these nitpicks add up to the conclusion that estimating the benefits of adding an on-die L2 cache is problematic, at best.
As for the results on cpu.simplenet.com, I find it curious that a K6-2 at 450/100 has a Winstone 98 score of 29.1, while the Celeron 450/100 has a score of 29.3. Given that the Celeron 450/100 is slightly faster than a Pentium II 450/100 (from Tom's Hardware Guide), the web site that you linked to suggests that a K6-2 at 450/100 is just as fast, if not slightly faster, than a Pentium II 450/100. This is hard to accept when even the K6-2 at 350/100 is halfway between a Mendocino 300/100 and a Mendocino 350/100 (again, courtesy of Tom's Hardware Guide). It's even harder to accept that the K6-2's performance will scale well with increases in the clock speed, given that the L2 cache is still on the motherboard. That's why the K6-3 with its on-die L2 cache is necessary for AMD to scale to the higher clock speeds and challenge Intel in its sweet spot.
Tenchusatsu |
