rasz_pl wrote on 2026-08-25, 18:19:

wow, this is great 😮
is CL5434 wired with support for Vesa Linear Framebuffer?

what exactly would this require?
is this anything different than the standard way to do it on the ISA bus and it's limitations?
the vga portion in essence is no different than the Speedstar64/Nitro64 cards for example but if we missed something we want to hear about it!

rasz_pl wrote on 2026-08-25, 18:19:

wow, this is great 😮
is CL5434 wired with support for Vesa Linear Framebuffer?

With UniVBE I think it can be used but in ISA bus there is no much advantage. Design doesn't differ much from other CL-GD5434. It's based on datasheet drawing with a few extras.

Wow, unexpected weirdness again 😀

I might just jump for this one too - at work we have a tiny 486SLC system with 1 16b ISA slot and 1 8b ISA slot, and a regular(ish) half-length bAT board with nothing onboard. Right now I have a combo VGA+IDE card in the 16b slot and an Aztech BXII in the 8b slot. But I'm missing LAN and a BXII's SB2.0 compatibility isn't the best sound ever. This card with an XTIDE would be slower on I/O, but better in everything else...

Edit:
One little bit of nit-picking though - the GD5434 doesn't support VRAM, only DRAM, so that's 2MB of DRAM on the card. These days people have long forgotten that distinction, but back in the days when these parts were current it mattered.

Oh, this is very interesting! I have switched to using smaller pizza box -format retro PCs to fit limited space, but that means I'm always running out of ISA slots if I want audio / MPU-401 / network.

How is the CGA compatibility on that chipset, IIRC it was one of the better ones at that?

keropi wrote on 2026-08-25, 18:45:

rasz_pl wrote on 2026-08-25, 18:19:

wow, this is great 😮
is CL5434 wired with support for Vesa Linear Framebuffer?

what exactly would this require?
is this anything different than the standard way to do it on the ISA bus and it's limitations?

1MEMR* I MEMORY READ*: This active-low input is used to indicate that a memory read is occurring. If linear addressing is being used, this pin must
2be connected to ISA signal MEMR*. If linear addressing is not being
3used, this pin can be connected to ISA signal SMEMR*.

a lot of brand cards still wired to SMEMR# and ignored LFB because
- not a lot of Mobos even had 15-16MB hole Bios option.
- still wanted that 8bit support

rasz_pl wrote on 2026-08-25, 23:36:

a lot of brand cards still wired to SMEMR# and ignored LFB because - not a lot of Mobos even had 15-16MB hole Bios option. - sti […]
Show full quote

Copy code to clipboard

1MEMR* I MEMORY READ*: This active-low input is used to indicate that a memory read is occurring. If linear addressing is being used, this pin must
2be connected to ISA signal MEMR*. If linear addressing is not being
3used, this pin can be connected to ISA signal SMEMR*.

a lot of brand cards still wired to SMEMR# and ignored LFB because
- not a lot of Mobos even had 15-16MB hole Bios option.
- still wanted that 8bit support

nope, we do not have this
will look into the possibility to make this a jumper so it covers all cases
8bit slot is not possible with the 5434 even on 386+ , none of the cards I have work in one (and it's video bios also needs a 386 , at least the original ones from back in the day)

edit:
will this actually make any difference?
I just checked Pine64 and they also do not use the MEMR* option, I wonder if this was also discussed back then

LFB is certainly useful on later titles that support VESA modes. Having done a little bit of VGA/VESA programming there's certainly less overhead (and therefore faster) in accessing that linear address range than swapping in those 64kb windows.

Most of the basic, unbranded CL cards I've owned over the years have supported it.

megatron-uk wrote on 2026-08-26, 08:31:

LFB is certainly useful on later titles that support VESA modes. Having done a little bit of VGA/VESA programming there's certainly less overhead (and therefore faster) in accessing that linear address range than swapping in those 64kb windows.

Most of the basic, unbranded CL cards I've owned over the years have supported it.

is there a benchmark that can use a LFB mode and we can actually get some data?

I have been researching this: the Cirrus design docs , the Pine card and a few random vga pics from the net show cards that do not use the LFB pins

My STB 5429, the speedstar64 and the nitro64 (I was mistaken before) and some random internet pics use the LFB pins

in both cases UniVBE5.3a does not report a LFB existence or show anything different when it loads
this is interesting but we have to decide how to handle this, the mention above of mobo/bios that do not support the memory hole is kinda alarming, what will happen if such a system is used and the card is configured for lfb? only vesa modes will be affected?

keropi wrote on 2026-08-26, 10:12:

is there a benchmark that can use a LFB mode and we can actually get some data?

PCPBENCH will try to use 640x400 LFB by default and you can use /NOLINEAR to tell it not to use the LFB.

IMO it's mostly a wishlist checkbox to see if it runs. All the games I can think of that support VESA modes that would use an LFB are late DOS games that struggle on a fast 486 and are thus pushing out of the ISA era. Typically they have a Mode X default res that is way faster than the VESA modes and are just about playable in that. So it makes slow resolutions less slow up to about P166 where they become playable, but then why the hell are you using ISA graphics on that instead of PCI? Even on PCI I think Quake VESA modes only just start to get playable on a fast PCI S3 on about Pentium 133.

I look forward to seeing how well this proves the viability of this combination of disparate features in a single card. While the SonicVision Pro as-is does not have the combination of [compatible] chipsets I prefer, if it proves a success I wonder if that means I could commission a custom combination making the best possible use of my remaining ISA slot.

The Linear Frame Buffer was introduced in VBE 2.0, so without being configured you should still be able to use VBE 1.0/1.2 and normal banked VBE 2.0 modes. As stated it is mostly later DOS games that added support, as well as demos.

There's not much advantage to the the linear framebuffer access for 320x200, since the entire region of course fits within the regular VGA window address space - it's when you go to higher resolutions that it becomes more efficient.

It's not a deal breaker, but I'd put it towards the top of a 'nice to have' list.

BTW, is there a PC-speaker input?

BitWrangler wrote on 2026-08-26, 12:11:

All the games I can think of that support VESA modes that would use an LFB are late DOS games that struggle on a fast 486 and are thus pushing out of the ISA era. Typically they have a Mode X default res that is way faster than the VESA modes and are just about playable in that. So it makes slow resolutions less slow up to about P166 where they become playable, but then why the hell are you using ISA graphics on that instead of PCI?

The only use case I can think of would be intentionally throttling performance via the slower ISA bus for something that is speed sensitive.

I've done this running Descent in 320x200 on a Pentium 4. With an ISA video card, it throttles video performance to the speed of a Pentium and is perfectly playable without CPU throttling methods.

No idea if there is such a case for games using VESA modes / LFB, since I imagine most if not all wouldn't be speed sensitive in the first place.

Arkku wrote on 2026-08-26, 17:19:

BTW, is there a PC-speaker input?

there is no speaker input on the card , main reason is lack of space

regarding the LFB addressing, the idea now is to use them and be similar to STB and Diamond cards
if anyone can think of any reason this would not work - please write about it but keep in mind that the 5434 will not work on a 8bit slot no matter where the MEM R/W pins are connected and that the vbios needs a 386+ system to work , so a 286 for example would not be able to use the vga portion no matter what we do (assuming stock/old vbioses are used)

No "turbo" button?
You should be hanged in the public square!

How fast of an overclocked ISA bus could it handle without artifacts (Yamaha OPL3-related issues)?

I am also curious about LFB with it. As mentioned, PCPBench is good for benchmarking this.

Another more useful check for LFB performance would be Fastdoom. If the 5434 on the ISA bus can do VESA (VBE 2.0) 320x200 w/ LFB, Fastdoom VBD should be able to use triple buffering and increase performance. You could try VBR as well in addition to base testing Mode Y and Mode 13h.

VESA 320x200 LFB is definitely worth it on VLB, but VLB is VLB.

keropi wrote on 2026-08-26, 18:15:

keep in mind that the 5434 will not work on a 8bit slot no matter where the MEM R/W pins are connected and that the vbios needs a 386+ system to work , so a 286 for example would not be able to use the vga portion no matter what we do (assuming stock/old vbioses are used)

Perhaps I and documentation wasnt clear. LFB support requires 16bit MEMR/W# wiring. 8bit support which you say is impossible anyway requires 8bit SMEMR/W# wiring.

This chip requires 16bit data . If a system does not have linear addressing it can still work with normal 16bit addressing. But it works with limitations. It's can only address 64kb at the time instead of 16mb. But since this is rare to happen. It only needs the memr# and memw#. Because we have access to linear addressing 98% of situations.

If you set bits 4-7 of SR7 to a value other than 0, does the LFB appear at that MB boundary?

Edit: from the pics in the first post, MEMR# and MEMW# are definitely not connected. So no memory access beyond 1MB is going to work.