DTK PKM-5031Y
With AMD 386 CPU at 50MHz mem bandwidth is at 105 Mb/s.
With SXL2 at 50/55 MHz is much higher.
But memory bandwidth is meaningless metric really.

So after I got that other DRx2 33/66 working this is my newest project: overclocking my Ti SXL2 to 33/66 MHz

CPU: Ti SXL2 25/50 running at 33/66 with added heatsink and fan
Motherboard: PC-Chips M326 G.W386/486 (like this one here https://www.ultimateretro.net/de/motherboards/5265)
Chipset: it says SARC but I found out that it's actually a UMC491-F
Cache: 128kB 15ns
BIOS: Swapped the Award with a MR BIOS that was designed for another UMC491-F board
Memory: 4x4MB 60ns
HDD: CompactFlash 512MB
Video: ISA 512kb AVGA2 (Cirrus Logic 5402)

Sidenote: it's interesting to see how the SXL2 outperforms the DRx2 by about 12% (in PCP, 3DB and Doom) at the same clock frequency due to the larger 8 kb L1 cache. (now if I only could get a IBM Blue Lightning 33/66 CPU)

Is your 5V SXL2 at 66 MHz able to complete Quake and load Windows?

feipoa wrote on 2022-03-29, 22:35:

Is your 5V SXL2 at 66 MHz able to complete Quake and load Windows?

I'll pop in an FPU and try the Quake test. Win 95 installed without problems, but I ran into some exceptions when I ran some Winamp tests. Not quite sure it it's the CPU or the ISA bus that runs at 16 MHz, thou. Need to test this further.

Edit: ran the Quake benchmark and had no problems

CPU Intel 80386 SX 20 MHz
BOARD Topcat Chipset
BIOS QuadTel
RAM 20 MB RAM (16 MB + 4 MB EMS)
SCSI Adaptec 1542B
HDD IBM DPES-31080
NET Intel EtherExpress PRO/10
SOUND SB16 CSP
VGA TSENG ET4000 W32 ISA

I'm sorry, I could not find a working VESA driver but tlivesa.com for vanilla ET4000 with VESA 1.2 support.

Windows 95 is running quite smooth for a slow clocked 386SX. .WAV system sounds are converted from 8-bit PCM via 16-bit PCM to 4-bit CT-ADPCM and decoded by CSP chip. GUI acceleration is done by Tseng ET 4000 W32. HDD access is done by busmaster Adaptec 1542B.

Wanted to share some notes about SXL2 CPUs based on past findings.

SXL2-50 has very high chance of running at 60-66 MHz but only on relatively slow motherboards.
For example this one - native 60MHz (no clock doubling), yet performance is lacking.

On fast clock-to-clock motherboards SXL2-50 does not climb much higher than its specification - tends to max-out at around 55MHz and even that is not very stable.
Below is a Quake 1 score from such system running at 50MHz natively (no clock doubling). Completely stable.
Number can be 3.6 fps if push things to 55MHz at compromised stability - mostly attributed to FPUs getting flaky in the context of Quake 1 and other apps demanding co-procs. So far haven't seen 40MHz rated FPU handling well 55MHz natively on fast board.

I too have noted a performance ceiling around 55 MHz on the SXL2 w.r.t. slow motherboards, e.g. VIA 481/495 based. JohnBourno, did you notice much performance gain between the SXL at 55 MHz compared to 66.6 MHz?

Unfortunately, the M326 mainboard doesn't have a dedicated crystal that I could swap out, so I'm limited to 25/33/40 MHz CPU clock.

My other 386 mainboard has a crystal socket, but it doesn't like to go faster than 45 MHz system clock, so hard to test the 55 MHz non-doubled clock with the SXL2.

pshipkov wrote on 2022-03-31, 05:25:

Wanted to share some notes about SXL2 CPUs based on past findings.

SXL2-50 has very high chance of running at 60-66 MHz but only on relatively slow motherboards.

I'm curious, what's the fastest results you got with an ISA bus based386 board? So in terms of 3D Bench and PC Player.

Wow, it looks like I did pretty much every single benchmark I know of except Speedsys on my 386SX-25 🤣

System Specs:
AT Mid Tower (No-Name)
TASK 230W AT
Scamp 386SX
AMD Am386SX/SXL-25
4MB (4x1MB FPM30)
Mitsumi 3.5" DS-HD
512MB CF/IDE (Internal)
CF/IDE (External)
ASUS 50x CD-ROM IDE
Cirrus Logic CL-GD5402 ISA 512KB
Edison Gold-16 (ES688)
MS-DOS 5.0/Windows 3.0

JohnBourno wrote on 2022-04-08, 09:05:

I'm curious, what's the fastest results you got with an ISA bus based386 board? So in terms of 3D Bench and PC Player.

Superscape - 50 fps
PC Player Benchmark - 11.2 fps
Wolf3D - 75.8 fps
Doom - 27.2 fps
Quake1 - 4.1 fps
more details here

Attached my 386SX25@27,5MHz. ISA and DMA @ 12MHz

CPU: AM386DX-40
FPU: IIT 3C87-40
Motherboard: FIC 4386-VC-HD
Chipset: VIA VT82C495 (Venus)
Cache: 256KB 20ns with 20ns TAG RAM and 15ns Alter / Dirty TAG RAM for write-back
BIOS: Award 4.20 version 1.15K with "Fast" DRAM Timings and "Turbo" Cache Timings
Memory: 20MB (4x1MB 70ns + 4x4MB 60ns, both Hitachi)
HDD: Maxtor 7540AV 504MB (jumpered for 1024 cyl)
Video: ISA 1024KB Cirrus Logic 5429
no overclocking (ISA / VGA or CPU)

Even with the "Turbo" setting, cache timings appear to be rather conservative, so I don't believe faster chips would change anything.

Other Dosbench scores
PC-Player VGA Bench: 4.0
Superscape: 14.7 / 14.5 (slower / faster systems)
Doom slow PCs: 28.34fps (2635 realtics)
Topbench: 79
Chris' Bench score: 13.2

CPU: AM386DX-40
FPU: None
Motherboard: ABit AB-FS3 https://theretroweb.com/motherboards/s/abit-ab-fs3
Chipset: SiS Rabbit
Cache: 256KB
Memory: 16MB (4x4MB 60ns)
HDD: QUANTUM MAVERICK 270A
Video: ISA 1024KB OAK OTI077
no overclocking (ISA / VGA or CPU)

CPU: Cx486DLC-40GP
FPU: Gray Cx83D87-40-GP
Motherboard: Digicom 386H (this one, it's my actual board): https://theretroweb.com/motherboards/s/digicom-386h
Chipset:OPTi 495SX
Cache: 128KB 20ns with 20ns TAG
BIOS: Mr.BIOS OPTI4BH, made for OPTi 495XLC, but seems to work beautifully here somehow.
Memory: 8x 1MB, (two different 4MB sets, both 70ns, Fastest 0 or 1 ws timings where available in BIOS)
HDD: Caviar 7200,
Video: NDI Volante Warp 10 ISA (good luck finding any info on that). 1MB RAM
40 MHz, 13.3 MHz ISA bus (CLK2/3 set in BIOS)

There is SO MUCH that's SO WEIRD about this motherboard. I'm almost beginning to suspect that DMAs are broken based on some behaviors, but everything else is amazing. One ISA slot even has pads to replace it with a OPTI/EISA slot (it's probably OPTI).

@iyatemu, the CPU's internal cache is also disabled. 486DLC should have 1kB of it and it will make a big difference, when enabled. (I would say in practice, the internal cache is the biggest differentiator between 386 and 486)

NB! on 386 and 486 boards, you can freely mix and match BIOSes between boards - the chipset is what matters.

GigAHerZ wrote on 2023-12-23, 10:30:

@iyatemu, the CPU's internal cache is also disabled.

Oh, it's enabled. Speedsys just can't see something that small. Here's the same test done literally just now with it off (only the CPU test and nothing else).

Here is also a much earlier test closer to completely stock from awhile ago using the onboard Am386DX-40.

The DLC on its own is a massive boost, the cache only bumps it up a little from there.
Stock confix (appx): - 6.82
486DLC cache off: - 9.39
486DLC cache on: - 10.20

The CPU score on speedsys is very unreliable. I get a rougly 7,5 with FPU installed and IIRC around 9 with FPU removed with my 386DX. It's good only for testing the same machine's speed difference when hardware has not changed.

I did expect to see higher cache speed in the beginning of memory speed graph. Though, i don't remember, if i saw it with my 486DLC... maybe speedsys is unable to measure a datapoint inside one kilobyte.

iyatemu wrote on 2023-12-23, 09:18:

CPU: Cx486DLC-40GP FPU: Gray Cx83D87-40-GP Motherboard: Digicom 386H (this one, it's my actual board): https://theretroweb.com/m […]
Show full quote

CPU: Cx486DLC-40GP
FPU: Gray Cx83D87-40-GP
Motherboard: Digicom 386H (this one, it's my actual board): https://theretroweb.com/motherboards/s/digicom-386h
Chipset:OPTi 495SX
Cache: 128KB 20ns with 20ns TAG
BIOS: Mr.BIOS OPTI4BH, made for OPTi 495XLC, but seems to work beautifully here somehow.
Memory: 8x 1MB, (two different 4MB sets, both 70ns, Fastest 0 or 1 ws timings where available in BIOS)
HDD: Caviar 7200,
Video: NDI Volante Warp 10 ISA (good luck finding any info on that). 1MB RAM
40 MHz, 13.3 MHz ISA bus (CLK2/3 set in BIOS)
Ophelia1.gif

There is SO MUCH that's SO WEIRD about this motherboard. I'm almost beginning to suspect that DMAs are broken based on some behaviors, but everything else is amazing. One ISA slot even has pads to replace it with a OPTI/EISA slot (it's probably OPTI).

I have the same board with a green PCB.
One thing weird I noticed about it is that it doesn't really appear to support Cyrix chips out of the box even though the jumper settings are silk screened on the board. Infact, I think one of the Cyrix specific jumpers even causes it to fail POST. In order to get things working properly, you need to make sure internal cache is disabled in the BIOS, and enable it using the software utility setting some cache exclusion areas.

Also, you may have noticed that the socket for your dirty bit is not soldered in. Without the dirty bit, if you enabled external cache to write back mode, your cache will always be dirty and you'll get a noticeable drop in performance. With dirty bit chip present, you should get a slight speed boost vs running in write-through mode.

I also noticed the solder pads for OLB, but as I don't own any OLB cards I opted not to bother installing the slot. It looks like a time-consuming nightmare anyway.

Anonymous Coward wrote on 2023-12-24, 09:42:

I have the same board with a green PCB.
One thing weird I noticed about it is that it doesn't really appear to support Cyrix chips out of the box even though the jumper settings are silk screened on the board. Infact, I think one of the Cyrix specific jumpers even causes it to fail POST.

I'm gonna dump a ton of info on this just because it's so interesting to me, so apologies if it's not welcome in this thread, but I actually did a lot of RE on this board when I got it because the behavior is so strange.

I actually found that the Cyrix-specific jumpers on mine (and yours by extension) are actually completely redundant if the onboard CPU is populated. Two of them are the cache flush (JP5) and Keyboard A20 signals (JP20 on mine, which is NOT present on any other revision, was wired incorrectly, and electrically reversed), and they only go to the PGA socket, those jumpers would only need to be removed if the CPU you're using is a PGA 386. JP13 wasn't present on my board AT ALL, that jumper shorts the /FLT pin of the SMD 386 to ground, but it also works double duty and shorts the Cyrix /KEN input to ground. Thankfully the state is ignored on reset so it doesn't seem to cause any issues, but it's still a strange decision to short both of them directly to GND. JP1 is also mentioned in the silk screen on some versions, but totally missing on others, That jumper shorts one of the KBC pins to ground, and I have no idea what it does. It's also unpopulated on mine and therefore open (but its "Cyrix" mode is supposedly "closed").

I think my board is a later "corrected" revision after looking at all of the similar versions that are on TRW, but obviously I have no basis for that assumption. There's a sheet of notes I typed up in the Downloads section of the TRW page for it, but I'm not sure how much of it applies for your board simply because of all of the minor differences between the other revisions.

I've been thinking of getting all the parts together to "restore" some of the missing features on mine, specifically the dirty bit and the clock generator. I've already added back the Cyrix compatibility (mostly) by adding and documenting the missing jumpers.

Anonymous Coward wrote on 2023-12-24, 09:42:

In order to get things working properly, you need to make sure internal cache is disabled in the BIOS, and enable it using the software utility setting some cache exclusion areas.

The AMIBIOS it had originally corrupted itself after first boot, so I replaced it with one from one of those similar boards that had a few options. I'm running it with a MR.BIOS now and that seems to work as well, but it sets totally different exclusion zones. It also doesn't work with the MEGA-KB-F this board came with originally, so I bought a JETkey 5.0 to replace it.

I let this BIOS enable the cache, but then I override it with Cyrix -xA000, 128 -xC000,256 -i3 -i4

I'll try disabling the cache completely and enabling it through software to see what happens. I've noticed things like DOOM work perfectly fine, but when trying to play Tyrian with SB16 sound effects, the game crashes within 2-4 minutes depending on the cache setting, but it always crashes. The only way the game doesn't crash is if I disable DMA sound effects, and I'm hoping it's not a faulty 82C206.

Anonymous Coward wrote on 2023-12-24, 09:42:

Also, you may have noticed that the socket for your dirty bit is not soldered in. Without the dirty bit, if you enabled external cache to write back mode, your cache will always be dirty and you'll get a noticeable drop in performance. With dirty bit chip present, you should get a slight speed boost vs running in write-through mode.

Like I mentioned, I'm set on restoring the missing features on this board, so the dirty TAG is coming. How do you choose the cache mode? I don't remember seeing an option for than in either the AMIBIOS or Mr. BIOS, and I'm not sure where the jumper for that is.

Anonymous Coward wrote on 2023-12-24, 09:42:

I also noticed the solder pads for OLB, but as I don't own any OLB cards I opted not to bother installing the slot. It looks like a time-consuming nightmare anyway.

I've been debating it just because that would make this a much more interesting board. It's my only retro PC-compatible motherboard though, so I don't want to stress it too much. I've already had to add some of the Cyrix jumpers to get it to work, and I've had to replace the KBC with a socket. Removing an entire ISA slot and installing an OLB/EISA one is a lot of heat and stress on the pads. It's not that I'm worried about the amount of work, I'm worried about the board itself.

The point stands, this is an extremely weird series of boards, where it has almost every feature you want from a 386 board, but it's all done half-assed and needed a look over by an engineer later, like they didn't have proper hardware to test them. I love it so much though, it's a really really fun project.