Couple of interesting things about the 128.

It follows the same pattern as the Matrox engine, where batches return while drawing and the core drains at Lock, Flip, Blt, DestroySurface, and CPU clears.

The 128 register guide confirms CRTC_OFFSET is double-buffered via vblank latching, with bit 30 serving as a hardware flip-pending flag similar to Haswell’s.

I've got 2 systems - one with an XGI Volari Z9 and one with a silicon image SM750.

Both of these chips don't support 3D acceleration even with their official drivers, but I wonder, if I sent you the V9XSURV output of these chips do you think 3D acceleration could be worked in?

Duffman wrote on Today, 07:43:

I've got 2 systems - one with an XGI Volari Z9 and one with a silicon image SM750.

Both of these chips don't support 3D acceleration even with their official drivers, but I wonder, if I sent you the V9XSURV output of these chips do you think 3D acceleration could be worked in?

Oh wow would love to work on those, survey will not be enough, I will need your machine on internet and you to install https://github.com/michaeldale/V9x-Remote-Agent and enable Remote Help from the tray icon, then I can drive the machine remotely and run scripts/tests on it.

Duffman wrote on Today, 07:43:

I've got 2 systems - one with an XGI Volari Z9 and one with a silicon image SM750.

Both of these chips don't support 3D acceleration even with their official drivers, but I wonder, if I sent you the V9XSURV output of these chips do you think 3D acceleration could be worked in?

ChatGPT thinks it is very unlikely they have any hardware 3D but I'd be happy to try.

Just curious, how many machines are currently part of your test-pig-orchestra?

myne wrote on Today, 08:08:

Just curious, how many machines are currently part of your test-pig-orchestra?

I've got 6 computers from people who have lent them. I have 3 main physical machines: P3 1G 440BX (this one is on test bech so gets all the cards), netbook w/GMA 950 and Dell HD 4600 machine. I then have 7 86box VMs all setup with various GPUs as well.

Rage 128 support is out, performance is very bad but it works! https://github.com/michaeldale/velocity9x/rel … ses/tag/v0.28.0

I should be able to check the i3 tonight.
btw, it was headless, hence no edid.

Well, it works which is frankly amazing by itself.
Performance is a tad shy of what the hardware should be able to do.
1200 3dmark 2000 is brutal 🤣
Still, it didn't look obviously wrong that I saw so it definitely has potential there

https://forums.tomshardware.com/threads/lowes … -scores.321145/

Duron700+tnt beats it quite a lot for reference
I do understand some tests didn't run, and I am extremely grateful and surprised that this even works.

myne wrote on Today, 11:01:

Well, it works which is frankly amazing by itself. Performance is a tad shy of what the hardware should be able to do. 1200 3dma […]
Show full quote

Well, it works which is frankly amazing by itself.
Performance is a tad shy of what the hardware should be able to do.
1200 3dmark 2000 is brutal 🤣
Still, it didn't look obviously wrong that I saw so it definitely has potential there

Haha yeah, I'm not happy with performance yet, lots and lots todo, but will get there.

The HD 4600 machines with the GPU/CPU are fast enough to overcome driver limitations to be mostly fast enough (I mean HL is playable on HL1 in 640x480), but I want to get the best out of all hardware.

Yeah, I understand. Edited above and then got distracted before submitting.

I don't know if it's relevant to your code, but in glidulator I think some of the functions need to be split from one big function that checks the same thing (whether to run fog or not) every loop.
Those branches would be costing a lot if I'm right.

myne wrote on Today, 11:22:

Yeah, I understand. Edited above and then got distracted before submitting.

I don't know if it's relevant to your code, but in glidulator I think some of the functions need to be split from one big function that checks the same thing (whether to run fog or not) every loop.
Those branches would be costing a lot if I'm right.

Yes thanks I saw your post about that, I haven't review yet, but will 100% go over it. I haven't really started on real performance improvements yet, only small quick wins.

myne wrote on Today, 11:01:

Well, it works which is frankly amazing by itself. Performance is a tad shy of what the hardware should be able to do. 1200 3dma […]
Show full quote

Well, it works which is frankly amazing by itself.
Performance is a tad shy of what the hardware should be able to do.
1200 3dmark 2000 is brutal 🤣
Still, it didn't look obviously wrong that I saw so it definitely has potential there

https://forums.tomshardware.com/threads/lowes … -scores.321145/

Duron700+tnt beats it quite a lot for reference
I do understand some tests didn't run, and I am extremely grateful and surprised that this even works.

Sorry which GPU/CPU did you test with? Good to have some baseline scores.

Intel hd ironlake (gen1 core i3)

Rofl

Though... I just noticed it says software t&l

The earlier gens of Intel graphics that have official 98 and XP drivers, I swear they do a cheat when they see 3DMark2001SE is running. If you watch the demo, there's frames missing for several seconds, and parts that are literal slide shows, you can count the frames, then it will tell you it did 8 to 15 frames per second. Anyway, scores come up like 1100-1800 and it looks worse than DX6 or 7 cards that get an "honest" 300.

All I am saying is that it may be impossible to get close to scores seen on Intel drivers for XP or a slightly earlier similar architecture IGPU that had official 98 drivers, because of the cheats in those.

myne wrote on Today, 13:14:

Rofl

Though... I just noticed it says software t&l

The attachment 20261023_201214(0).jpg is no longer available

yeah I think something is wrong on that driver path, I'm getting better score on my Rage 128. I can run counters via the remote agent if you like.

Also yes I haven’t done hardware t&l yet! This driver is more 3dmark99/2000 era focused at the moment but I’ll get there !