Here's my machine. Trying to make it the best possible 386 while still being a real 386. (So 486DLC, which i do have, doesn't count)
CPU: Am386DX-40
FPU: Cx83D87-40GP
Motherboard: M-321
Cache: 256kB + WriteBack chip installed
BIOS: Tightest settings available + AMISETUP used for some hidden settings
Memory: 8x4MB
HDD: CompactFlash 4GB
Also using XTIDE Universal Bios on a DE-250CT NIC. (NE2000 compatible card)
Attachments
Last edited by GigAHerZ on 2020-08-19, 14:20. Edited 1 time in total.
"640K ought to be enough for anybody." - And i intend to get every last bit out of it even after loading every damn driver!
Going by what GigAHerz and stalk3r posted, and my own benchmark found in the previous page (Re: Post your 386 Speedsys results here) we can probably asume that we hit the limit of the "real" 386DX-40 at 31 mb/s cache and 20-21 mb/s ram.
Stalk3r is using the same motherboard as I am, and the results are astonishingly close.
I didn't realise there was this kind of thread here before. I'll post my speedsys result here as well then.
Here is the thread where you can see all the parts: Early 90s 386 build as close to period correct as possible.
There are no other timing settings in bios except dram waitstates have been set to 0 W/S
I also have this extra result from SI of pctools 9.0 that I thought was bit strange because it puts my 386 above that 486.
Memory bandwidth is great, cache/memory speed is ok, CPU score is low since NPU is installed. This BIOS doesn't have a lot of the more standard options, and some I'm not really clear what they are ("TURBO Cache Function", "Back to Back Memory Cycles", etc). Any ideas what those are? Suggestions for tweaking performance welcome too!
Been playing around with some of my 386 motherboards a bit so here are some results:
CPU: Am386DX-40
FPU: None (won't post with fpu installed, tried IIT and Cyrix, and various bioses with the same result)
Motherboard: Seritech SER-386ADIII
Chipset: Ali M1429
Cache: 128K Write Back and Write Through
BIOS: AMIBIOS
Memory: 8x 1M 70ns
It's interesting how with WB, L1 gains performance but plain memory loses performance... weird
Write Back
Write Through
CPU: Am386DX-40
FPU: None and FasMath 40
Motherboard: Octek Jaguar V
Chipset: Macronix MX83C305FC and MX83C306FC
Cache: 8KB Inside chipset (WB or WT?)
BIOS: AMIBIOS and MR-BIOS
Memory: 8x 1M 70ns
Interesting how the shape of the write line is different just at the end of the cached 8kb with original AMIBIOS (which matches this post) and with MR-BIOS which doesn't show that dip. Even if MR-BIOS numbers are a bit worse in speedsys, it performs better in 3DBench, Doom, Wolf3D, Topbench, and even in Quake.
GigAHerZwrote on 2020-08-15, 13:03:Here's my machine. Trying to make it the best possible 386 while still being a real 386. (So 486DLC, which i do have, doesn't co […] Show full quote
Here's my machine. Trying to make it the best possible 386 while still being a real 386. (So 486DLC, which i do have, doesn't count)
CPU: Am386DX-40
FPU: Cx83D87-40GP
Motherboard: M-321
Cache: 256kB + WriteBack chip installed
BIOS: Tightest settings available + AMISETUP used for some hidden settings
Memory: 8x4MB
HDD: CompactFlash 4GB
Also using XTIDE Universal Bios on a DE-250CT NIC. (NE2000 compatible card)
So, in the journey of making it the best 386 ever, i took the motherboard out and desoldered the crystal oscillator. Now i can replace it as i wish.
I bought 90MHz and 100MHz oscillators.
The 100MHz one doesn't even show clock being present on debug card. So i might have gotten 3.3V variant accidentally.
But the 90MHz crystal works! So it's now a 386DX-45! 😁 (Speedsys thinks, it's 50MHz, but MR-BIOS and chkcpu both confirm it to be 45MHz)
Also, in the mean time, i've gotten MR-BIOS on it. With tightest settings, it was equal to AMI bios with hidden settings adjusted.
That's a lovely M321. I have 3 of them, but none works. 2 have severe corrosion and the 3rd one is in mint condition, looks like new, but still won't post. Weird.
That's a lovely M321. I have 3 of them, but none works. 2 have severe corrosion and the 3rd one is in mint condition, looks like new, but still won't post. Weird.
Where did you get the MR-Bios?
Yeah. That board has had quite a bit of work put into it. All cache jumpers were soldered with bridges, cache chips were soldered. Crystal oscillator (as mentioned above) was soldered. Everything had to be desoldered and socketed. (and i also added few decoupling capacitors that were missing) Plus the ordinary CMOS battery replacement, new BIOS chip and i had to even replace keyboard controller to fully satisfy the MR-BIOS.
I think he's got the hots for the Symphony board. I still haven't even turned mine on. lol. I'd like to compare the SiS Rabbit with the Symphony one day.
Plan your life wisely, you'll be dead before you know it.
Gigahertz was talking about TRUE 386.
Symphony Haydn is a hybrid 386/486 chipset so it is outside the defined scope.
I was thinking more about ASUS 386/33-64k and Octek Jaguar V, for example.
They are clock-to-clock slower than M321 but scale better - can reach 50MHz and outdo M321 there.
As for the Symphony Haydn chipset - it is not a magical by itself.
Tried at least 5-6 386/486 mobos by different vendors and they are ok at best.
It is DTK's magic touch in PEM-4036Y that made it stand out.
That package is ahead of any other classic 386 mobo and majority of VLB ones.
It takes late hybrid systems like Alaris Leopard/Cougar with BL2/3 CPUs running at 80+ MHz to partially outdo PEM-4036Y running at 50MHz.
Different story - we looked and talked at length about that stuff in another thread nearby.
M321 is a great board itself.
It was my entry point to the retro 386 game.
I picked it up after long research and it holds special status for me. 😀