It was 15ns from a PCI UMC motherboard.

I have the Lucky Star LS486e Rev D mobo, which does take EDO ram, and remember that there is a setting in the bios that drastically reduces the memory speed as You
have shown in the first post.
Maybe try adjusting the L2 cache tag bit setting to 7 and see if that increases the speed.

I'll check with my motherboard tomorrow after work.

Dirkmirk, those results are really weird. Cachechk reports your memory speed as equal to your L2 speed when L2 is disabled or removed. I have seen similar behavior with a PC Chips M918 motherboard which also uses this ALI chipset. I am speculating, but I think board manufacturers didn't figure out the nuances with the ALi chipset in comparison to the more common UMC and SiS variants.

I tested speedsys memory speed with Cache tag bits at 7 then 8, here are the results.

I've always observed odd behavior if you do not set Write-back L2 cache to a 7-bit TAG and write-through L2 cache to an 8-bit TAG.

I think I might need some help with setting up the cache perhaps that's why the results aren't right.

The board has 5 slots, 1 for tag & 4 cache slots, the motherboard has settings for 128,256,512kb of cache, the memory I installed were.

Tag chip - UM61256FK-15 (28pin)
Cache - 4 X W24512AK-15(32 pin)

Could I install 4 X UM61256FK-15(28 pin) for 256 or 128K?

Don't really know what I'm doing or what the numbers for cache means.

I've done some extensive testing and as you would expect does not show up in the benchmarks 😵 .

Something is seriously up with this motherboard, I'll summarize the facts.

#60ns edo ram with cache disabled yields the best results.
#60ns fpm ram with cache enabled is equal to 60ns edo with cache disabled

#60ns edo ram bandwidth in speedsys dropped from 238mb/s to 119mbs with cache enabled.
#70ns fpm ram bandwidth in speedsys dropped from 148mb/s to 132mb/s with cache enabled, I figured out that 60ns fpm yielded the same results so I stuck with that.

#I was able to enable 128kb of cache memory on the off chance something was wrong with my 256kb, no performance gain, settled back on 256kb L2 cache.
#When I had 128kb cache installed the second boot screen showed 128kb but when I disabled cache in the bios, that 2nd screen showed "256k cache" 😐
#I could not get any stick of fpm to work together, ie two sticks of 60ns 16meg fpm = incompatible as well as every other combination of 16meg/60ns/fpm + 4/8 meg fpm/70ns memory.
#Currently running a single stick of 60ns 16meg fpm,
# 0 wait states for L2 cache works perfectly, setting to 1 wait state freezes every time.......
#Every stick of 72 pin ram tested booted perfectly.
#Memory throughput/bandwidth highly important to overall benchmarking performance, I would try speedsys everytime and if the bandwidth at the first test screen showed no improvement it was pointless testing any other software.

For example, EDO with cache disabled = memory bandwidth of 238mb/s = best performance, enabled=119mb/s equal to fpm cache enabled = 132mb/s BUT cache disabled = 148mb/s which resulted in the worst benchmark performance, not entirely consistant.

I've settled on 16meg/60ns FPM ram for the moment but will seek out a single 32meg stick, I presume that it must result in better performance with L2 in windows 95 and im willing to sacrifice a little performance in dos for that.

Still cant understand why the cachechk effective ram access times are 70ns/77ns(read/write) with cache enabled, 44/51ns without cache.

I initially tried to slow down the dram settings in bios in case their were incompatibilities with the cache but that would see bandwidth from 119mb/s to 85mb/s and only slower the system, same thing with the cache ram if I slowed the settings it would lower overall performance, I figured if I got the best performance with cache disabled what If I slowed the cache right down would that increase the dram performance?

Not at all 😐 .

If you are seeing "256K cache" when L2 is disabled, I think your motherboard manufacurer tweaked the BIOS to make the user think that the fake cache which came with the board is real. PC Chips did this scam with displaying text indicating that there is cache when there is, in fact, none. I assume you are not using those cache chips which say "WRITE BACK".

I suppose it is possible, but unlikely, that EDO RAM on the ALI board is "so fast", perhaps due to automatic wait states being reduced, that the L2 cache is of the same speed as the RAM. With FPM, some additional wait states are added, making this confirmation need the cache to get up to EDO speed. All of this is far reaching speculation - ALI PCI 486 boards are oddballs and I think they weren't all that common for a reason.

I would not use the Speedsys bandwidth speed as any kind of performance indicator. I've come to learn that it is not at all reliable whatsoever. Speedsys has issues with less common CPUs and chipsets. You should rerun your tests and hardware configurations using Doom timedemo3.

I need the manual for SMT4321-22B 😢

So that is the best setup for these Chipsets?

What types of ram ? and at what speed ?
What types of cache ?

this old post looks very interesting, is there anyone else to verify on the ali1489 boards:
1. do they yield huge boost with edo than with fpm?
2. are their running without onboard cache as fast as with?

Does the L2 cache speed make much of a difference or is it minor and not worth the upgrade ?
I usually use 512kb. cache @ -20n or -15n

From what I recall looking at, the EDO speed with an ALi board is about the same as a non-ALi board (namely, UMC & SiS) with FPM. And the L2 results on the ALi were pretty sub-par. I have two of these ALi boards, but haven't published quantitative results vs. that of SiS & UMC. I remember being underwhelmed with the results and threw the board back into the motherboard bin.

feipoa wrote:

From what I recall looking at, the EDO speed with an ALi board is about the same as a non-ALi board (namely, UMC & SiS) with FPM. And the L2 results on the ALi were pretty sub-par. I have two of these ALi boards, but haven't published quantitative results vs. that of SiS & UMC. I remember being underwhelmed with the results and threw the board back into the motherboard bin.

i really want to see a detailed comparision between ali1489/sis496/umc8881, especially how they perform with amd5x86-133, cyrix5x86-100 and pod83, and the difference between edo and fpm, and with or without cache.
i also suggest using 33fsb as a baseline, because not everything can run stable at 40fsb with tightest timings.

One thing I'm surprised no one picked up on is that most of the boards being talked about here only have single bank cache. The SiS chipset supports bank interleaving, as do most other 486 chipsets. There is certainly a performance benefit with interleaved cache, although how much remains to be seen. I have a SiS 496 board on the way with two cache banks. I will be benchmarking all possible configurations to answer some lingering questions.