Notice the ~20% improvement in memory scores between 3-3-3 & 2-2-2. I am unable to set 5/7T with my Mushkin rev 1.5 memory. Largest perf increase comes from running RAM at 133MHz (compared to 100MHz). Next largest increase comes from running RAM at 2-2-2 (compared to 3-3-3). Smallest increase comes from running RAM at 5/7T (compared to 7/9T). 

Need great RAM to run 2-2-2 at 5/7T @ 133MHz or better. I cannot, but I hear it's possible with Mushkin rev2. Same RAM chips (as rev 1.5), with better PCB. If anyone else can run 2-2-2 with 5/7T at 133MHz with RAM other than Mushkin rev2, I'd like to know. 

Article here (TweakMax) says CAS-3 memory needs 112MHz to get same bandwidth as CAS-2 at 100MHz .. and CAS-3 memory needs 154MHz to get same bandwidth as CAS-2 at 133MHz. Benches are for BX chipset, but I'd expect similar numbers for 815e (not sure tho).

Brian Stone writes that, with Mushkin rev 2 (2x 128) set to 2-2-2 7/9T [P3 933/133, BIOS 102, WinME] he gets Sandra ALU/FPU of 386/454. When overclocked to 140/140/35 (980MHz), he gets 402/472. But anything above 140, his memory bench plummets to "well below" his 133/133/33 scores. Heard (from Daniele) that anything above 140 resets AGP to 2X. Anyone think this may be hurting memory perf >140 FSB?

Looks like BIOS 1003 improves RAM perf. After flashing, Brian reports these Sandra ALU/FPU memory scores:
395/461 (133/133/33, 2-2-2, 5/7T)
405/476 (140/140/35, 2-2-2, 7/9T - cannot get 5/7T at 140)

Paul from Connecticut reports marginal improvements with 5/7T timings, but notices 0.5 fps less in UT at 5/7T (P3-866, no o/c):
PC133 2-2-2 5/7T: 398/428 
PC133 2-2-2 7/9T: 389/414 

Philster's memory benches:
With 2 modules of Mushkin Rev2 at CAS 3-3-2 7,9 154/154/38 (1001 MHz) Sandra: 418/484
With 2 modeles of Mushkin Rev3 at CAS 2-2-2 5,7 154/154/38 (1001 MHz) Sandra: 448/526

Sandra HDD 75GXP on WinME at ATA66 protocol (23,883): here
Brian's Sandra HDD (IBM 75GXP): 23,618

BG reposts that a new version of Sandra - Sandra 2001 - has been released Nov 20, and that this will be the first version to completely sppt the CUSL2 & 815e chipset. There are some probs with the current version - like when to try to check out some mobo info, or the BIOS, the program will hang/lock. 

HD Tach graph of IBM 75GXP on WinME at ATA66 protocol: here
HD Tach graph of IBM 18LZX on WinME at U2W/LVD protocol here
HD Tach graph of Seagate Cheetah X15 (15Krpm) on WinME at U160 protocol here (compliments of Rooster)

Warning: If you run the HD Tach bench, do *not* put a check in the box that says Enable Write test. It has overwritten at least one person's hard drive (another reason why you want to use Norton Ghost, or similar imaging program).

Brian's 3DMark2000 v1.1 (Herc Prophet II GTS 64MB w/ Det3 drivers) Default bench: 6855

Quake III Arena (1024x768, 32-bit, highest everything, sound enabled, Radeon DDR 32MB @183 ala Power Strip) demo 001: 63fps. (the kids are Counterstrike junkies, but my fave is Q3A Instagibs - one shot, baby!)
Brian's Quake III Arena demo 001 1280x1024x32 all setting MAX'ed out (WinME)  59fps
Same Quake III Arena demo 001 as above (all settings MAX'ed out) except for 16-bit color (32-bit textures): 87fps

WCPUID for P3-700, cB0 stepping, retail @938MHz. WCPUID AGP for Radeon DDR 32MB. Can get WCPUID here.  

Closing comments:

Many are using the CUSL2 for their first system build. I recall my first, and yes, the mobo was most intimidating part - the BIOS, a no-mans land, where nothing made sense. Glad this guide has helped some first-time builders get up & running. The CUSL2's BIOS interface is one of its strengths. I'm used to Abit boards. The BH6 (w/ C300a) was my cherry build. Abits have great BIOS'es. If you like Abit boards, or appreciate a nice BIOS, you'll love the CUSL2. 

My logic for buying the CUSL2 went something like this: My BH6 (original v1.0) didn't have (official) sppt for Coppermine CPUs - something about lacking necessary 8.4 voltage spec. I wrote to Abit, and they said the v1.1 boards would run Cumines, but v1.0 boards would not. Some claimed to be running Cumines on their v1.0 BH6's fine, but others reported probs. I didn't want to be in the position that, every time I had a prob, I wondered, "Is this cuz my BH6 doesn't have the necessary voltage support? .. or is it cuz of something else?" 

Also, my AGP card at the time (V3-3K) wouldn't dance at 89MHz, and I wanted to run my RAM at 133Mhz. The BX chipset lacks a 1/2 AGP divider. Celeron2's with the BH6 was also an option I considered, but noticed that Celeron2's had poor perf compared to P3s (4-way vs, 8-way set associative). Built my 1st system Dec '98, w/ Cel300a @464MHz (pre-tested at 450 @ default voltage). 

That CPU (US$125) lasted me until the CUSL2, w/ P3-700 (cB0, retail) @938 (1.8v). Almost 2 yrs. Good deal, and I woulda went with a Cel again, if they offered perf comparable to P3's. I'm too chicken to try a Via chipset, especially since I heard here that Via has probs with Win2K (incomplete AGP sppt), and I spend most of my time in W2K. Always heard that Intel made the best chipsets .. (altho their reputation is steadily declining, ever since the 820/840 fiascos).  

At time, CUSL2 was only 815e board available. MSI was talking, but Pricewatch had not a single 815e Pro listed. Have no prob recommending one to a friend [the acid test, cuz you know he'll be calling you all hours of the night, if he has probs]. So the CUSL2 seemed like the best solution. I very happy with my choice.