Benchmarks
So, bring on the benchmarks. I mean, what is a review with out pointless bar charts and tables? In this test, I want to show you just what the little' old K6-2s are capable of, vs the supposedly mighty PII-core Celeron. My favorite test is Quake II, Software, 640x480. I use software mode to cancel out the abysmal performance of my Savage 4 vs his Voodoo3, and place extra stress on the FPU unit. This shows just what 3DNow! can do, even on top of the frankly, Pentium-class FPU on the K6-2.Quake II Software 640x480:
K6-2 350 Non-3DNow! (3.5x100): 15 fps
K6-2 350 3DNow! (3.5x100): 21.2 fps
K6-2 500 Non-3DNow! (5x100): 20.1 fps
K6-2 500 3DNow! (5x100): 24.5 fps
Celeron 400 (6x66): 16.8 fps
I paid all that money for a downgrade? Huh? My kick ass Celeron with its monster Pentium-II grade FPU is only 80% of the speed of a K6-2 350? Good job I don't actually play too much Quake II software mode. These benchmarks, however, cheered me up.
Distributed.Net RC5 Crunch Rate:
K6-2 350 (3.5x100): 590 Kkeys/sec
K6-2 500 (5x100): 840 Kkeys/sec
Celeron 400 (6x66): 1120 Kkeys/sec
Well, a different story here. While Quake II makes good use of 3DNow! acceleration, Distributed.net relies heavily on rotating binary numbers. This is something the K6 has to do with several assembler instructions, while the Celeron does it in one go. Good news is I always have distributed.net running :)
So, I could leave it there. The K6-2 seems to be better at games, but it won't come first everywhere (esp. when the 100MHz L2 cache starts to drag things back). But that wouldn't be too good an article would it? I was, slightly annoyed maybe, but to be honest, other things kept me amused so I was able to ignore the CPU issue (see Making the Ultimate Hardware Modem, out soon).
