An interesting fact to know about the Virge/VX is that it requires the framebuffer to be in VRAM, as it uses the VRAM serial output to read the framebuffer contents during video display. On the other hand, offscreen sprites that are only used to be copied into the framebuffer and 3D textures are read using the parallel port which is provided by any kind of DRAM, whether VRAM or not. This is exploited on the ELSA card: You see 8 Chips of VRAM (for 4MB of framebuffer-capable memory) and 4 chips of EDO RAM (for 2MB of offscreen/texture only memory). The parallel port of the VRAM chips (used for writing framebuffer data into them) is EDO, so just looking at the parallel port, that card has 3 banks of EDO RAM of 2MB each. The first two banks also provide VRAM features for displaying the framebuffer contents.

Before EDO RAM was common, the parallel port of VRAM usually was FPM, just as the non-VRAM DRAM chips.

The Virge VX supports up to 4 banks of 2MB (i.e. 8MB) RAM, with only the first two banks being usable for framebuffer data, so someone could build a card with 2MB more EDO DRAM for more texture space. Furthermore, as the VX datasheet explains, if you use an external RAMDAC with an 128-bit interface (expensive!), that external RAMDAC can be supplied from 8MB VRAM (kinda expensive), thus using more than 4MB of framebuffer memory, allowing e.g. a 1600x1200 truecolor mode. You might find professional CAD cards implementing that variant, but I wouldn't expect that configuration in the consumer market.

So even on Virge/VX (without external RAMDAC), 8MB is pretty much useless? Virge/325 is not considered a particularly good 3D card, and the Virge/VX is the slowest of the bunch (in 3D) if I recall, so it seems the texture memory would be completely wasted.

mkarcher wrote on 2024-08-01, 18:40:

An interesting fact to know about the Virge/VX is that it requires the framebuffer to be in VRAM, as it uses the VRAM serial output to read the framebuffer contents during video display. On the other hand, offscreen sprites that are only used to be copied into the framebuffer and 3D textures are read using the parallel port which is provided by any kind of DRAM, whether VRAM or not. This is exploited on the ELSA card: You see 8 Chips of VRAM (for 4MB of framebuffer-capable memory) and 4 chips of EDO RAM (for 2MB of offscreen/texture only memory). The parallel port of the VRAM chips (used for writing framebuffer data into them) is EDO, so just looking at the parallel port, that card has 3 banks of EDO RAM of 2MB each. The first two banks also provide VRAM features for displaying the framebuffer contents.

Before EDO RAM was common, the parallel port of VRAM usually was FPM, just as the non-VRAM DRAM chips.

The Virge VX supports up to 4 banks of 2MB (i.e. 8MB) RAM, with only the first two banks being usable for framebuffer data, so someone could build a card with 2MB more EDO DRAM for more texture space. Furthermore, as the VX datasheet explains, if you use an external RAMDAC with an 128-bit interface (expensive!), that external RAMDAC can be supplied from 8MB VRAM (kinda expensive), thus using more than 4MB of framebuffer memory, allowing e.g. a 1600x1200 truecolor mode. You might find professional CAD cards implementing that variant, but I wouldn't expect that configuration in the consumer market.

Thanks for the informative post it has been enlightening and I've got a lot to learn.

So that means that even though the Japanese Buffalo Melco VX8 has a combined 8MB of VRAM it can't actually make use of it due to not having an external 128 bit ramdac.
Instead its split like the other cards.

Well dang , makes me wonder why they even bothered and didn't do split vram+edo like the other two cards to save cost.

The STB is the one I've always seen, didn't know about the others.
It's been a long time but I remember thinking the STB 8MB VX card had better image quality than the other Virge cards I've used.

AppleSauce wrote on 2024-08-02, 03:23:

So that means that even though the Japanese Buffalo Melco VX8 has a combined 8MB of VRAM it can't actually make use of it due to not having an external 128 bit ramdac.
Instead its split like the other cards.

Well dang , makes me wonder why they even bothered and didn't do split vram+edo like the other two cards to save cost.

See chapter 7.7 "Block Write support" in the Virge/VX data sheet: VRAM enables a performance optimization for certain 2D drawing operations. I don't know how often these kind of operations happen in offscreen memory, but there is at least a theroretical advantage. Even if most block-writes hit the framebuffer part, not needing to care about the destination address simplifies the driver. Possibly the primary use case of VRAM for the offscreen memory is the possibility to put the Z-buffer into that part of memory. Clearing the Z-Buffer is an operation that greatly profits from being able to use block writes.