The speed (ns) of the individual chips comes into play too.
I only took a quick look and could have missed it but I didn't see that mentioned in the video.

My SS7 preference has been to use 2x256MB Low Density (double sided - 16 chip) PC133 modules at 6 or 7ns.
I don't remember any slow downs.

(I'm snoozey and probably scrambling terms here.)
My thinking when I started doing that was that two double sided modules presented 4 'sides' to the system and that would have a better overall bandwidth.
.

I have read before that many 486, P, P2 and p3 boards (prior to dual channel) have slightly slower scores when 3 or more slots are used versus 2 due to the way the north bridge or memory controllers scan and address the ram banks. A lot of it was speculation but real world results typically show what you are seeing.

Well, I tried a bunch of different configurations, but on this machine it really doesn't seem to make any difference. There's a slight trend towards lower benchmarks with less RAM, but only in a few tests. In games it doesn't amount to anything, as you'd expect. The DIMMs I have are a little mismatched, but aside from one of the 128MB ones, they're all using the same MT 48LC16M8A2-75 chips.

BTW it's worth mentioning this is a later revision P5A-B which does not suffer the >128MB caching bug (but does have the K6+ slowdown bug, which I'm going to have to deal with later.) It seems to be truly 512MB "clean."

I ended up just sticking the 2x256MB back in. I didn't bother to try 768MB (3x256) since it's completely pointless and causes problems with Win98.

This is getting strange. The above results were done with a K6-2/500AFX with a 2000 date code (0021) running at its rated 2.2V. I swapped in a K6-2/500ADK with a 1999 date (9947) out of a laptop, set the voltage to its indicated 2.1V, and the results were atrocious. Blatantly slower in just about everything. So I pulled the second RAM stick and ran it with 256MB and it sped up noticeably, but still didn't match the newer AFX. Finally, I bumped the voltage back up to 2.2V and it could nearly match the AFX results, but not quite.

I didn't run the full suite of benchmarks, but for reference:

1                          500AFX 512MB -- 500ADK 512MB -- 500ADK 256MB -- 500ADK 256MB 2.2V
2SuperPi 1M                      8:16.0          9:50.5          9:23.4          8:47.3
3ScienceMark SGEMM 3DNow   557.2 Mflops    528.8 Mflops                    553.5 Mflops
4ScienceMark SGEMM x87-ASM  71.0 Mflops    69.96 Mflops                     71.3 Mflops
57Zip all threads              359 MIPS        323 MIPS                        347 MIPS

(There's usually a little variability in these benchmarks, but nothing on this scale if the performance is the same. Most of these tests were run on a clean boot too.)

I know there were a couple architecture changes in the 2000 models, but I didn't expect them to behave that differently. What's going on?

"Fast" CPU:

"Slow" CPU:

Incidentally, the K6-2+ won't POST at all at its rated 2.0V, I have to set it to 2.1V to get it to boot. Others here have reported the same problem on this board.

I believe the 1.06 Asus P5A boards have problems with ANY of the latest cores. So some 500 and 550chips did not behave well and are probably from the same batches as the k6-2+ and k6-3+. The problem is not cache related as the normal k6-3 400/450mhz should work fine on that board (and is the best choice without any hassle).

There is a simple mod on here that makes the mobiles etc. work fine with the 1.06 (but you have probably seen it). Making the 1.06 boards the most desirable where it was the least before (imho).

Personally I run a 1.04 with 128mb and a k6-3+ with 98se as I see no reason for more ram. None of the games I play on that machine needs more and 64mb seems to be fine already. I also have the 1.03, 1.05 and 1.06 boards to play with. I did check the mod on the 1.06 and it works perfectly. I know a 1.04 with a k6-3+ can cache more ram but there is always that small penalty.

But I understand the game of ''maxing out hardware'' and it's just fun to do..