Are you sure about that 5MHz ISA setting...?

I usually opt for 16MHz if possible yet the raw VGA speed results are similar - if Landmark gives a comparable result to the vspeed 16-bit one.

All in all, I found that UMC chipset interesting. Seems to perform very well and I've never met it in the flesh. I have OPTi 495SLC/XLC and ALI M1429, though.

alvaro84 wrote:

Are you sure about that 5MHz ISA setting...?

I usually opt for 16MHz if possible yet the raw VGA speed results are similar - if Landmark gives a comparable result to the vspeed 16-bit one.

Setting ISA to CPUCLK/5 instead of CPUCLK/8 on the Biostar motherboard gave 10% boost in 3D Bench. Not all of my motherboards had this setting though and I wanted the results to be comparable (I was mostly interested in the chipset performance obviously, as other things were the same).

alvaro84 wrote:

All in all, I found that UMC chipset interesting. Seems to perform very well and I've never met it in the flesh. I have OPTi 495SLC/XLC and ALI M1429, though.

I recently got a PC with another UMC491F motherboard. It was working fine (for most of the time) until I pulled it from the case and cleaned the battery dammage. It's dead now and won't POST 🙁 even after repairing one of the traces...

Thanks for the benchmarks! You did lots of work there.

I've got a Jaguar V as well and it gives 12058 Dhrystones in NSSI. I haven't done any other benchmarks but I'm using the Mr. BIOS instead of the original BIOS so I guess that might be where the performance increase comes from. For comparison my HOT 307 with 256kb cache gives a score of 11617 in NSSI.

Great stuff there. Thanks for the effort.

The results are too low. Wonder what is causing it ?
For example, Doom readings here are almost double yours.

I used 70ns memory and left the timings alone (not all BIOSes even had an option to set them). Graphics card was TSENG ET4000AX - not the fastest, but not the slowest either. Also like stated above I left the ISA bus clock at 5MHz (again - not all motherboard supported clock settings). There is some room for optimization, but I just wanted to set the baseline so I will have some reference for future tests If I get some more Am386DX-40 motherboards, and I wanted these tests to be easly repeatable by other users - without looking for highest performance graphics card and memory modules... Also with higher ISA clocks I had well over 12000 Dhrystones in NSSI (more like 12200) on the Biostar motherboard. Anyway, thanks to these test I know that Abit and Shuttle boards have to go, and I will stick with a Biostar, TMC (EISA!) and Octek (because I like Octek stuff 😉 ).
I will definitely look at Mr BIOS.

@pshipkov what's your system configuration? Twice as high readings in Doom? It's more like 486 performance...

Got it. Thanks.

PC Chips M321 rev 2.3 (i believe it is an octek chipset)
ET4000AX or GD-5426 (the CL one is better overall, but in Doom they perform equally)
I made a post on the system specs forum with more details.

* EDIT: octek = opti

Just a though, by not using the same memory + cache timing in all motherboards this is more that a "BIOS defaults" benchmark than a motherboard benchmark. For example safer default values will cripple a fast motherboard and make it appear slower that another with more aggressive defaults. I understand this for motherboards that don't have timing settings, but for those that have you would have got more accurate results about the motherboard speed by setting the timing to the same values. Although to be honest your results don't show any motherboard using terrible default values, I only wanted to mention this.

konc wrote:

Just a though, by not using the same memory + cache timing in all motherboards this is more that a "BIOS defaults" benchmark than a motherboard benchmark. For example safer default values will cripple a fast motherboard and make it appear slower that another with more aggressive defaults. I understand this for motherboards that don't have timing settings, but for those that have you would have got more accurate results about the motherboard speed by setting the timing to the same values. Although to be honest your results don't show any motherboard using terrible default values, I only wanted to mention this.

This was described in the methodology section of the spreadsheet - same memory modules were used and same cache chips (only on BIOSTAR I left 15ns chips because I didn't wanted to ruin a sticker on them).
Timings (which I believe could be set only on two motherboards) and waitstates were set to the best value that felt stable although I didn't do any extensive tests of multiple configurations. Waitstate values differed among the motherboards because on some setting W/S to 0 caused errors even during boot (ie. himem.sys couldn't load) and on other it was perfectly stable - I supposed it may be related to how chipset and/or cache handles waitstates... Also some motherboard didn't allowed to set W/S to 0 at all.

HanJammer wrote:

This was described in the methodology section of the spreadsheet - same memory modules were used and same cache chips (only on BIOSTAR I left 15ns chips because I didn't wanted to ruin a sticker on them).
Timings (which I believe could be set only on two motherboards) and waitstates were set to the best value that felt stable although I didn't do any extensive tests of multiple configurations. Waitstate values differed among the motherboards because on some setting W/S to 0 caused errors even during boot (ie. himem.sys couldn't load) and on other it was perfectly stable - I supposed it may be related to how chipset and/or cache handles waitstates... Also some motherboard didn't allowed to set W/S to 0 at all.

I meant "using the same timing for the memory and cache", it's clear that you used the same chips. I only mentioned this because I read earlier on that

HanJammer wrote:

I used 70ns memory and left the timings alone (not all BIOSes even had an option to set them).

But I see you tweaked it when it was possible so it's ok, you got indicative results.

HanJammer wrote:

Setting ISA to CPUCLK/5 instead of CPUCLK/8 on the Biostar motherboard gave 10% boost in 3D Bench.

As I expected. What divisors are there in the setup? I suspect that CPUCLK refers to the crystal or clkgen output, not the actual CPU clock which is usually halved. Fixed clock 386DX40s tend to have a 80MHz crystal. So that CPUCLK/8 should be 10MHz while CPUCLK/5 is 16MHz. I think.

Yeah... something like this... problem is these boards don't seem to use the same CLK source or at least - use different designations in BIOS:

A bit like:

CPUCLK and CPUCLK2/2 - 386DX-FA3

CPUCLK/8 /6 /5 /4 /3 /2 - Biostar

SCLK/5 /2 /3 /4 - Jaguar V

CLK2IN/4 /5 /6 /8 - Puma200

CLKIN/4 /3 /2 SCLK/2 - Shuttle

(this is based on BIOS strings, not actual settings in setup - but it looked pretty much like this) - SCLK is system clock I guess, and CLKIN? same as CPUCLK?

If looking for the ultimate 386, try finding a PANDA 386V - https://stason.org/TULARC/pc/motherboards/U/U … -PANDA386V.html
With VLB cards, it should easily outrace other 386 boards in most tasks.

HanJammer wrote:

TMC (EISA!)

Are you sure it's EISA?
Looks more like Opti Local Bus to me - Re: EISA vs OPTi Local Bus cards - how to tell
Not electrically compatible!

I've updated the benchmark with 3 motherboards:
Soyo SY-019I/H - strong competitor, a lot of options to tweak in BIOS, but not very stable (at least not with 70ns DRAM and 20ns SRAM) even with moderate changes to default values.
FIC LEO 386-25/33/40SC-HQ (386-SC-HQ) Rev. A - very disappointing, perhaps slowest of the motherboards I tested.
PCChips (HTC, Elpina) M321 Rev. 2.2 - suprisingly (it's PCChips after all) perhaps the fastest motherboard so far, not much options to tweak in BIOS but runs flawlessly with default 0W/S settings.

I also dropped some tests as they are redundant or (like Doom) seem to be too much video-card dependent and since the first tests I done I got much better (faster) ISA cards than ET4000AX (45ns memory WDC90C030 in particular) which is visibly faster than ET4000AX so my assumption it's not choking the DX40 is no longer valid.