I have this 286 mobo:

and these fast ISA VGAs:

and also a couple really old 8/16bit Realtek and UMC ones (that are really slow) plus Sergey's ISA Super VGA

what do I bench? 😎

I would try anything up to and including the Wolfenstein 3D benchmark included here : http://www.philscomputerlab.com/dos-benchmark-pack.html

so pick for example the WDC vga and do a wolf3d bench with VGA BIOS Shadowing On and Off?
Or you also want results from vgas with single and double bios roms?

I remember that shadow ROM would greatly improve performance in textmode. But I wouldn't expect it to do much for games, if at all, since games generally don't use BIOS functions, other than setting a graphics mode on startup.
Perhaps the exception is when games use the BIOS ROM font (although that is stored in the system ROM).

I don't need VGA ROM BIOS Shadowing, although it is conceivable that a 286 chipset could support such a thing in a custom way. I only need a 286 benchmark from roughly equivalent VGA cards.

OK, so I will Wolf3D-bench my VGAs and report back tomorrow - easy for me to do 😀

Allright, I found 2 Realtek RTG3105 cards - one with single BIOS chip and the other with dual BIOS. Here are the cards:

I used Phil's benchmark suite - Wolf3D for FPS and Landmark for characters/ms. The motherboard's BIOS allowed for several shadowing options:
none/system/vga/system+vga . So I benchmarked with either system+vga shadowing or none.

SINGLE BIOS VGA
with shadowing : 9,4fps - 1109,53 chr/ms
without shadowing: 9,4fps - 1109,53 chr/ms

DUAL BIOS VGA
with shadowing: 9,3fps - 1108,28 chr/ms
without shadowing: 9,3fps - 1108,28 chr/ms

So not really a difference in performance... is this what you were looking for GH?

well odd/even just meant the mfg used cheaper lower capacity eeprom than one larger one back in the day. easy to use the lower address line as a chip enable toggle and do odd/even. the performance going from an odd/even two chip to a single chip should be zero, the lowest address line on the card goes to the chip enable, the other lines go to both chips. it should be transparent as electrically its no different. only time youd see a neglible difference is if the two chip implementation had faster access times than the single larger chip implementation, but the isa bus speed + wait states are fare longer than the access times on the eeproms.

with shadowing bios on 486 up, at that point your getting into heavier weight memory controllers and cache. if you turn shadowing off, you have to round trip the isa bus, deal with abysmally slow eeproms and isa bus wait states. shadowing on takes advantage of caching and using ram directly hanging off the cpu/mem controller rather than the slow isa bus.

keropi wrote:

Allright, I found 2 Realtek RTG3105 cards - one with single BIOS chip and the other with dual BIOS. Here are the cards: […]
Show full quote

Allright, I found 2 Realtek RTG3105 cards - one with single BIOS chip and the other with dual BIOS. Here are the cards:

I used Phil's benchmark suite - Wolf3D for FPS and Landmark for characters/ms. The motherboard's BIOS allowed for several shadowing options:
none/system/vga/system+vga . So I benchmarked with either system+vga shadowing or none.

SINGLE BIOS VGA
with shadowing : 9,4fps - 1109,53 chr/ms
without shadowing: 9,4fps - 1109,53 chr/ms

DUAL BIOS VGA
with shadowing: 9,3fps - 1108,28 chr/ms
without shadowing: 9,3fps - 1108,28 chr/ms

So not really a difference in performance... is this what you were looking for GH?

Yes, I think that settles the issue. Is that BIOS screen for your 286? I was not aware that 286 chipsets supported shadowing. If so, then I guess that shadowing makes no difference and the split vs. the single ROM makes no appreciable difference.

BIOS shadowing improves performance the most when BIOS functions are heavily used. VGA BIOSes are most frequently used in text modes and to set up graphics modes, but graphics routines themselves typically don't require much in the way of BIOS functions (which are pretty basic in graphics modes). Of the ROM BIOS, the disk drive functions are by far the most frequently used after bootup. If the Int 13h functions are contained in the main BIOS, then it would have a performance boost in that machine. If on a card (extended BIOS support), then you are one generation behind.

Yep that's the Bios of my 286 board... It's one of the later 286/16 with Headland chipset
If you notice the landmark chrs/ms results there is no difference with shadowing options - and this is a text-mode benchmark as far as I can tell. Maybe the bios option doesn't do anything 🤣

keropi wrote:

Yep that's the Bios of my 286 board... It's one of the later 286/16 with Headland chipset
If you notice the landmark chrs/ms results there is no difference with shadowing options - and this is a text-mode benchmark as far as I can tell. Maybe the bios option doesn't do anything 🤣

It's possible that the Landmark benchmark renders text directly to video memory, rather than using BIOS functions for it. In which case, shadowing has no effect.
Do you see any difference if you just get a directory listing from your HDD or such? That's done with BIOS functions, and that's the thing where I noticed a big difference on my machines at the time.

I will check the directory listing speed - I just use a lotech cf adapter with a 256MB cf for testing/benchmarks so it's always kinda slow... maybe CheckIt uses BIOS functions for the VGA speed test?

I will also add that back in the day, my Trident video card that came with my 286 also included a TVGABIO.SYS that you could stick in your CONFIG.SYS and it would shadow the video BIOS into RAM even if your hardware didn't support it. My guess is that it just copied the C800 segment to low memory and pointed int10 there instead.

I remember copying this driver over to my XT with an EGA card and running it there, and not only did it work, but it made a *huge* speed difference for most text-mode apps - they were at least twice as fast.

But as others have said, games and other apps using direct hardware access weren't affected.