I'm not sure your data matches your conclusion. I've seen ~4800 char/msec on a 386sx-16 which is only slightly faster than your 286-12. So I'd say yes the video card is still being bottlenecked

Is that the unshadowed bios routines performance?

maxtherabbit wrote on 2025-03-31, 20:17:

I'm not sure your data matches your conclusion. I've seen ~4800 char/sec on a 386sx-16 which is only slightly faster than your 286-12. So I'd say yes the video card is still being bottlenecked

Well, I'm talking "bottleneck" in terms of "upgrading won't gain you any performance", which is the effect of the advice people tend to give about 286 video cards - "don't bother with the fast ones, the 286 isn't worth it!" - but what I'm finding is you can gain significant performance benefits anyway, even though it's a 286.

Also, fwiw, Topbench matches my system with a 386SX-16.

vstrakh wrote on 2025-03-31, 20:21:

Is that the unshadowed bios routines performance?

I don't think those tools are using the BIOS routines. But I think they're shadowed in any case.

Considering the character size in text mode is 2 bytes (a symbol + attributes) the throughput of 6kb/sec is abysmal. Whatever that rating is - it does not represent the ability of the cpu to saturate the ISA bus.

I don't know what those numbers actually represent. All I know is they were higher with the better card and I experienced better real world performance in games. That was the only point of this thread.

https://www.target-earth.net/wiki/doku.php?id … t_vlsi_vl82c201

286-16 (VLSI based) w/Et4000: ~7000chr/sec
286-25 (VLSI) w/ET4000: ~ 10000chr/sec

https://www.target-earth.net/wiki/doku.php?id … t_eurosoft-ht12

286-16 (HT12) w/ET4000: ~5200chr/sec
286-20 (HT12) w/ET3000: ~7500chr/sec

For reference on the same boards a TVGA9000 was around 30-40% of the figures above.

A CL-GD5428 is on par or slightly faster than the ET4000.

I misread the units! It's char/ms, not char/s.

~1800 char/ms for the WD90C00 and ~2800 char/ms for the WD90C30.

Just tried an ET4000 and it performs identically to the WD90C30.

Fwiw Landmark thinks my system performs like an "18 MHz AT". Also I have 0 wait states enabled in the BIOS.

keenmaster486 wrote on 2025-04-01, 02:33:

I misread the units! It's char/ms, not char/s.

~1800 char/ms for the WD90C00 and ~2800 char/ms for the WD90C30.

That's a different story now. 2800 char/ms would be around the max theoretical throughput of the ISA bus (~5.2MB/s at 8 MHz clock), so the CPU did bottleneck the card/bus after all.

vstrakh wrote on 2025-03-31, 20:49:

Considering the character size in text mode is 2 bytes (a symbol + attributes) the throughput of 6kb/sec is abysmal. Whatever that rating is - it does not represent the ability of the cpu to saturate the ISA bus.

I recommend running "TheFast" or older "Speedy3".
These are utilities that essentially change MS-DOS text-output from BIOS to direct.
On a PC with an 8-Bit bus @4,77 MHz this makes some difference.

vstrakh wrote on 2025-04-01, 07:59:

keenmaster486 wrote on 2025-04-01, 02:33:

I misread the units! It's char/ms, not char/s.

~1800 char/ms for the WD90C00 and ~2800 char/ms for the WD90C30.

That's a different story now. 2800 char/ms would be around the max theoretical throughput of the ISA bus (~5.2MB/s at 8 MHz clock), so the CPU did bottleneck the card/bus after all.

regardless of what the units are or what the testing methodology is, I have seen ISA only systems score up to about 8000 on this benchmark before

Well, so there are two things at play here. One is what I was originally trying to convey with this thread, that you shouldn't knock putting a faster VGA card in your 286 just because "it's a 286, it's already pushing pixels as fast as it can" or something like that. You will get better performance and it will matter in games. And the second thing is: you'll also get better performance out of that faster VGA card if you have a faster CPU or bus. But that's no reason not to use the faster card with the 286.

There's always a "bottleneck" somewhere, but there's nothing wrong with leapfrogging one limitation with another.
Your WD90C00 was obviously holding the system back, so upgrading the card was still a benefit.

I haven't gotten a chance to use a real 286 yet (unless I was too young to notice), but threads like these make make me excited to build and test one. It's fun to work on the bottlenecks one by one and improve overall performance while learning and feeling the different levels of parts, so thanks for the report keenmaster486.

It's also cool to see how far the high-end of one generation can meet the beginning of the next generation, like how a fast 386 can meet or exceed an early 486-33. Can a fast 286 match a slow 386, granted it'll never run 32-bit code?

Cosmic wrote on 2025-04-02, 00:08:

Can a fast 286 match a slow 386, granted it'll never run 32-bit code?

Yes, it's possible to match or exceed the speed of a 386SX with a 286 clock for clock running 16 bit code, since the 386SX only has a 16 bit data bus and takes more clocks for certain instructions, or so I've heard.

in my testing a 286 is about 5% faster than a 386sx, clock for clock on the exact same chipset

perhaps you should have titled the thread "yes a 286-12 can benefit from a better graphics card"

maxtherabbit wrote on 2025-04-02, 13:09:

perhaps you should have titled the thread "yes a 286-12 can benefit from a better graphics card"

Yes that would have been a more accurate title lol.

keenmaster486 wrote on 2025-04-02, 15:06:

maxtherabbit wrote on 2025-04-02, 13:09:

perhaps you should have titled the thread "yes a 286-12 can benefit from a better graphics card"

Yes that would have been a more accurate title lol.

The commentary that a 286 is too slow to care has always felt a bit peculiar considering even historically we knew EGA slowed 286 performance compared to VGA.

It’s only logical that the video speed should affect any cpu, because we have always known some vga setups are extraordinarily slow even in the era they released.

My guess is people were probably figuring 6% on a 286-6 AT didn’t matter much and went with it even though 286s came in a wide array of configurations, even with certain chipsets crippling performance (you would have thought that was only a pentium problem but nope already in the 286 generation you had garbage chipsets and memory controllers)

Sort of like saying a voodoo GPU is wasted on pentiums because it isn’t maxed out on a p60

rmay635703 wrote on 2025-04-02, 15:56:

keenmaster486 wrote on 2025-04-02, 15:06:

maxtherabbit wrote on 2025-04-02, 13:09:

perhaps you should have titled the thread "yes a 286-12 can benefit from a better graphics card"

Yes that would have been a more accurate title lol.

Sort of like saying a voodoo GPU is wasted on pentiums because it isn’t maxed out on a p60

I always thought that was utter BS even back then. Of course now we've even that a 486 benefits AND that it benefits from a Voodoo II as well. Granted I have a long history of staged builds and have done "odd" things like adding a Voodoo II to a 233 Mhz K6 build I was using back then, which then was totally bottlenecking an Athlon a little later, but there's no such thing as a totally balanced and optimal build.

I think it all comes down to real world performance.

Does the addition of something like a faster VGA card or a Voodoo card make it possible to play X, Y, and Z games smoothly, or does it just make you want a faster CPU to match?