I run ISA at 16MHz on my Acer based 386 machine and it gave very substantial performance increase in graphically intensive games such as Doom (not that you should really play it on a 386 🤣). I did have to cherry pick cards to get a functional state as most cards i have don't run reliably at 16MHz while they did at 12.

I don't suppose you know offhand exactly which cards worked and which didn't?

I could theoretically buy any osc, so intermediate speeds like 14MHz would be on the table too

Most of the chips that are used on VLB cards that are also used on ISA cards can handle it. The bigger issue is can your HDD controller handle it.

I'm using :
* WD90C30 video card, it won't show random artifacts like my second fastest card TVGA8900D does.
* 3C509B-C LAN with XTIDE, the older 3com cards are not stable at 16MHz (ones that don't have "parallel tasking" written on the chip)
* IO card with bus buffers on IDE data lines. Cards without these buffers were very unstable and often showed data corruption.
* HDD is some modern Seagate, older drives I had didn't cope. No clue about CF cards and SD adaptors.

I do have a spare WD90C31 card if needed. And IIRC my 3C509B is a -C (it does say parallel tasking)

IDE is only being used for CDROM, HDD is SCSI through the AHA-1542B. When you say bus buffers, do you mean like 74 series logic ICs on the IDE card? The only chips on the card are the UMC ASICs IIRC.

When I tried most (not all) VGA cards could handle 16MHz ISA in a 386DX-40 with ALI chipset. Strangely even old 8-bit ones could take it. The full list of cards/chipsets is:

ATI Graphics Ultra
C&T F82C451
Cirrus Logic GD5401, 5402, 5422, 5426, 5429
Diamond Stealth VRAM (S3 924)
Elsa Winner S3 805
Intel Kama (very old, very slow 8-bit relic)
Oak OTI037C
Realtek RTG3105, 3105i
Trident 8900C, 8900D, 9000C, 9000i
WDC 90C31

The only anomaly I remember is CL-GD5426 exhibiting palette glitches in Wolfenstein. I'm not sure it doesn't do it anyway. Either I'm wrong and the 386 board divides the ISA clock from the CPU clock instead of the oscillator (I don't think so, even though the results could be slightly beaten by my QDI OPTi895 at 13.3MHz) or in lucky circumstances these cards can really deal with this high clock.

On the other hand, Anonymous Coward is right, the HDD controller is a bigger issue. I mostly use a Winbond W83757F based I/O card and it seemed to work but I've met some that couldn't take such high ISA clock and if they don't the consequences can be far worse than a screen corruption, data corruption up to the level of fully garbled directories or losing entire partitions in a fraction of a second. Best case scenario: the computer politely stays on the POST screen and won't boot.

Edit. Oh, you just said 286? My Octek Fox II (20MHz, 0WS capable) has a BIOS option called "Fast bus" or something similar. And apparently no HDD controller can deal with it. But it's nice fast even without it, reaching VGA speeds like 5MB/s in Landmark.

At some point I compared my ISA and vlb cards on one of my 486. I used a ISA tseng et4000 and then I raised the ISA speed up to 20MHz with no issue with it.

maxtherabbit wrote:

I do have a spare WD90C31 card if needed. And IIRC my 3C509B is a -C (it does say parallel tasking)

IDE is only being used for CDROM, HDD is SCSI through the AHA-1542B. When you say bus buffers, do you mean like 74 series logic ICs on the IDE card? The only chips on the card are the UMC ASICs IIRC.

Personally, I think a busmastering SCSI card over 12MHz is not a good idea. I don't run mine faster than 10MHz. If you're going to do it, at least use a 1542CF or CP. The 1542B is pretty ancient.

ISA bus at 1.2GHz!

Back in the day with the mca royalty in effect
I often wondered why no one made a specific ISA standard centered around having one ISA slot (lets call it sisa) that could run at higher frequencies than the rest, sort of like EISA but without the hassle

I would think that would have been much more simplistic and straightforward as standards go than what actually happened between 1988-1993. Further minimal investment would have been required so popularity would have likely met critical mass rather than failed standards.

I have everything on the disk backed up. If I do try this, I will flog it thoroughly to make sure it's not corrupting data before putting the machine back in service, and be fully repaired to rebuild the partition if necessary.

rmay635703 wrote:

Back in the day with the mca royalty in effect
I often wondered why no one made a specific ISA standard centered around having one ISA slot (lets call it sisa) that could run at higher frequencies than the rest, sort of like EISA but without the hassle

I would think that would have been much more simplistic and straightforward as standards go than what actually happened between 1988-1993. Further minimal investment would have been required so popularity would have likely met critical mass rather than failed standards.

They did. It's called local OPTi Local Bus.

Anonymous Coward wrote:

rmay635703 wrote:

Back in the day with the mca royalty in effect
I often wondered why no one made a specific ISA standard centered around having one ISA slot (lets call it sisa) that could run at higher frequencies than the rest, sort of like EISA but without the hassle

I would think that would have been much more simplistic and straightforward as standards go than what actually happened between 1988-1993. Further minimal investment would have been required so popularity would have likely met critical mass rather than failed standards.

They did. It's called local OPTi Local Bus.

My AST had onboard opti local bus video, never seen it in slot form

Spent the better part of the day sunday pulling the system all apart, socketing the oscillator, and reassembling. Turned out to be mostly a waste of time but I did discover something odd.

I stuck a scope on the ISA clock pin B20 and put it in freq counter mode. The stock oscillator is 48MHz. The CPU is 12MHz. With "turbo" mode off, the CPU runs at 8MHz and the B20 pin shows a steady 8MHz. However, with "turbo" on, the freq on B20 actually changes between 12MHz and 8MHz depending on what the system is doing. It seems like the mainboard is slowing down the system speed during bus cycles and clocking it back up during on-board memory and CPU cycles.

Upon discovering this, I figured that sticking a 60MHz oscillator in there with "turbo" off should produce a system with a synchronized 10MHz clock during all activity, and remain within safe limits of the 12MHz CPU. The results were very disappointing. The system would not even POST with my Tseng ET4000 VGA card and Adaptec 1542B. I pulled them out and temporarily switched to the WD90C31 with a CF card on the IDE. System booted just fine in this configuration, but after running DISKTEST a few times on the CF card, the IDE interface locked up and didn't start working again until a cold boot. Is 10MHz ISA really that much to ask??

10MHz is pretty easy to reach for most systems. Even an original AT can do 10MHz. Something else must be going on.

Anonymous Coward wrote:

10MHz is pretty easy to reach for most systems. Even an original AT can do 10MHz. Something else must be going on.

My thoughts as well, the system still runs 100% normal with the stock oscillator so I don't think it's related to my work though.

Not sure how to interpret the results - could the clock generation circuit on the board be overwhelmed by the 60MHz xtal? Something else becoming unstable at 16MHz even with turbo off?

Your board has a clockgen? It's usually either/or, not both.

The oscillator output goes to an IC labeled "Acer 013-11-1BS"