Page 7 of 17
Re: Let's build a CTPCI!
Posted: 23 Mar 2023 10:52
by Badwolf
This is very cool. I don't have a CT60, but it's nice to see how these can perform when unleashed.
Technically, software-wise, how does this work?
I'd imagine even with a full 32 bit bus and an 060 running at goodness-knows MHz, a simple framebuffer with CPU rendering wouldn't cut the mustard.
So I'm assuming this is done with a patch to the VDI such that primitives and whathaveyou are translated into native graphics card commands in order to get the speed up? Is this all secret-sauce closed binary stuff or is this described, explained or published somewhere?
Speaking as someone who's played a bit with fVDI but only with a framebuffer and already one layer removed from what the OS is up to & would like to understand it better, if not actually be able to do it.
BW
Re: Let's build a CTPCI!
Posted: 23 Mar 2023 11:03
by alexh
Badwolf wrote: 23 Mar 2023 10:52
I'd imagine even with a full 32 bit bus and an 060 running at goodness-knows MHz, a simple framebuffer with CPU rendering wouldn't cut the mustard.
Isn't that how PiSTorm currently works? They have no hardware accelerated rendering at all.
Re: Let's build a CTPCI!
Posted: 23 Mar 2023 11:07
by mikro
BW, you are basically right. The bus itself isn't shockingly fast (33 MHz) but as you correctly pointed out, it's fVDI which makes the difference. Not sure how many primitives were patched / accelerated but obviously enough to make a difference.
The fVDI part is open so it's not such a secret actually:
https://github.com/mikrosk/ct60tos/tree/1.01
Naturally, the biggest downside is that it is really just a graphics card with separate Videl and PCI output (SuperVidel in comparison got this right from the beginning, I think this is its killer feature).
Re: Let's build a CTPCI!
Posted: 23 Mar 2023 14:00
by Badwolf
mikro wrote: 23 Mar 2023 11:07
BW, you are basically right. The bus itself isn't shockingly fast (33 MHz) but as you correctly pointed out, it's fVDI which makes the difference. Not sure how many primitives were patched / accelerated but obviously enough to make a difference.
The fVDI part is open so it's not such a secret actually:
https://github.com/mikrosk/ct60tos/tree/1.01
Oh, that's cool, thanks Mikro. Looks like a lot of good stuff in there. :)
BW
Re: Let's build a CTPCI!
Posted: 23 Mar 2023 15:32
by marss
foft wrote: 22 Mar 2023 20:35
marss wrote: 22 Mar 2023 09:01
I buy one if you have one left ;)
Well I managed to order 4 extra PCBs and 9 extra chips. I might have to put some of them together at some point...
P.S. If anyone wants to swap a completed one for a supervidel let me know.
I buy your Supervidel, too !!! How much?
Re: Let's build a CTPCI!
Posted: 23 Mar 2023 19:07
by foft
marss wrote: 23 Mar 2023 15:32
I buy your Supervidel, too !!! How much?
The other way round! I was asking if anyone wants to swap their Supervidel for my CTPCI.
Re: Let's build a CTPCI!
Posted: 23 Mar 2023 19:09
by foft
HigashiJun wrote: 23 Mar 2023 10:41
Better then to set a queue for interested folks ?
Sure, send me a PM
Re: Let's build a CTPCI!
Posted: 23 Mar 2023 19:09
by foft
I found a bunch of GPU docs here. No R200 docs though, only R300. At least unless it just has a weird name.
https://github.com/olvaffe/gpu-docs.git
They wanted to release them, but did it happen?
https://www.phoronix.com/forums/forum/l ... #post81315
Re: Let's build a CTPCI!
Posted: 24 Mar 2023 07:23
by mikro
foft, I think you are better off to study the open source Linux drivers (AMD did release some documentation back then IIRC so the community could implement it from scratch) -- they have pretty good support even for older graphics cards:
https://www.x.org/wiki/RadeonFeature
Btw, for us is good either R100, R200 or R300 (Radeon 7xxx - 9xxx).
Re: Let's build a CTPCI!
Posted: 24 Mar 2023 14:51
by Rustynutt
foft wrote: 22 Mar 2023 07:47
The Atari ‘green bar’ dma test passes. Is that accurate enough?
Honestly don't know.
Have a CT63 with the bus accelerator feature, never used that method.
Way back when, recall Rodolphe stating a clock buffer is not needed as the CT63 handles it. It got off into a "civilized" discussion where some users stated their setup yet required it, with him caving and saying if it works it works.
Thinking about the direction Exxos took his clock patches with additional gates, it's possible using the CT63 "with wires", as well as an F04 patch produced similar results.
Anyway, it's just that I haven't tried it setup that way.