The claim "due to how they work...didn't require any special support" is false.
Every PCI card in your system is a separate device, and gets assigned its own unique address space. The driver needs to be aware of this. So at the very least, there must be support in the driver for multi-GPU support, as the driver has to expose multiple hardware devices as a single virtual graphics card to the graphics API.

I am not aware of any PowerVR drivers that have such support, although I do concur that the tile-based deferred rendering will make it relatively easy for multiple PowerVC cards to work together.
After all, they basically just copy their drawn tiles into a framebuffer on the main video card, via DMA transfer over the PCI bus. By definition, each tile is an isolated rectangle of pixels in the framebuffer, no overlap with any other tiles. So you can render as many tiles in parallel as you want.
The driver already has to batch draw calls and sort/clip polygons to the tiles. It would be trivial to distribute the tiles over multiple videocards.

So yes, in theory it is plausible, but I've never heard of this being done in practice.

I think that user just doesn't understand how the PCX2 card works. It's not that the cards are working together, it's that those early PowerVR literally render to the framebuffer of the primary card, in this case his Voodoo Rush. The curious part to me is that that I would have assumed the performance difference between a Rush and a PCX2 would have hard to notice by eye on most titles. The most likely scenario is either rose-tinted glasses of nostalgia or they were comparing software rendering to the PowerVR, not the Rush. The waaaaaaaaaaaaay out there theory is that maybe they had a Rush with the 3d capable Alliance chip and were comparing that steaming pile to the PCX2, in which case it absolutely would look like a night and day difference.

Apparently the drivers have been made open source:
https://github.com/powervr-graphics/PowerVR-Series1
The constant:
#define MAX_BOARDS 16
seems hopeful.

But a quick scan seems to show that this is only for the raw PCI devices.
The driver parts appear to use a single global HLDEVICE gHLogicalDev variable, so you select one 'logical device' that maps on one of the up to 16 PCI devices.
So it doesn't look like any part of the driver code can work with more than one card at a time.

There's also that SEGA fandom claim of NAOMI 2 having two PowerVR2 chips in "SLI", but I believe that's more of a split ISP/TSP situation akin to the earlier Midas cards

Also a theory of a SLI PowerVR (not SLi'd but more tiled) sounds like an extreme nightmare when you do any screen-based texture tricks (i.e. bloom, gta3 motion trails). A single Kyro drops to 3fps with that going on.

Well, based on their documentation here: https://github.com/powervr-graphics/PowerVR-S … re_for_CGDC.pdf
The architecture is actually designed to be 'scalable', and they show an example diagram of an 'arcade design' which uses 4 ISPs and one TSP.
However, the PCX chip combines everything in a single chip, and the documentation doesn't mention any kind of 'expansion bus' or scalability with multiple PCX chips.
So it's unclear whether PCX chips can be wired up similar to their example of the ISP (or whether you could use multiple TSPs for that matter).
Based on this documentation, I would think that scalability is ONLY possible at the ISP-level, and only with separate ISP and TSP chips, not with PCX ones.

A board with multiple first gen. PowerVR chips exists, it's the Namco System 23 from 1997 https://en.wikipedia.org/wiki/Namco_System_23

Not sure about which chip is what.
Crisis Zone, a late game on the most powerful version of the system, looks pretty good.
The PowerVR technology was definitely created with scalability in mind, but maybe not the PCX PCI cards.

JustJulião wrote on 2023-07-02, 13:12:

A board with multiple first gen. PowerVR chips exists, it's the Namco System 23 from 1997
Crisis Zone, a late game on the most powerful version of the system, looks pretty good.
The PowerVR technology was definitely created with scalability in mind, but maybe not the PCX PCI cards.

Do you see the word PowerVR anywhere here?

Crisis Zone shows way too much poly density, bilinear filtering, smooth alpha modulation and the kind of sorting a PCX1 can't do. Also, PowerVR cannot do stippled alpha like these jailbars on glass

leileilol wrote on 2023-07-02, 14:53:

JustJulião wrote on 2023-07-02, 13:12:

A board with multiple first gen. PowerVR chips exists, it's the Namco System 23 from 1997
Crisis Zone, a late game on the most powerful version of the system, looks pretty good.
The PowerVR technology was definitely created with scalability in mind, but maybe not the PCX PCI cards.

Do you see the word PowerVR anywhere here?

Crisis Zone shows way too much poly density, bilinear filtering, smooth alpha modulation and the kind of sorting a PCX1 can't do. Also, PowerVR cannot do stippled alpha like these jailbars on glass

Thank you for the detailed reply.
Checking MAME devs was the first step but "custom chips", especially on a board they struggle a lot with, doesn't imply it's not PowerVR based to me.
I didn't assume that the late version of the board running Crisis Zone was PCX1 based, System 16 even states that
it has additional graphical hardware : "System Super 23 games have texture filtering that standard System 23 games don't have".
At the end of the day, you are still right, it's not based on it as explained by a dev in the comment section (the claim that it's PowerVR based is quite popular)
https://www.youtube.com/watch?v=LYauMieaAPA
The early games displayed in the video looked more likely to be so, but I'll obviously trust a dev working on it more than a press claim of the the 90's.