I think we're good; tech demos, game demos, Win95 update, DirectX, IE5 and the drivers. Doesn't look like anything that'll get us in trouble.

Note to self: Sol_HSA pointed out that TIE95 can call PVRHAL32.DLL
http://www.gog.com/forum/star_wars_tie_fighte … rectstuff/page1

This isn't PowerSGL but more like optimized Direct3D, the kind that comes up in those "choose your primary display driver" menus if I recall correctly.

I have lost Scali's blog twice now, it will not happen again:
http://scalibq.wordpress.com/2012/12/18/just- … it-real-part-6/
Root: http://scalibq.wordpress.com/

I tried to examine the CPU and PCI dependancy of PCX1. The results are unexciting:

http://vintage3d.org/data/PCX1.jpg
I triple checked red values, don't know why it happens. The effect of clock increase of modern CPUs and their PCI bus has positive, but small effect. Without sound, but still...

I know PCX1/PCX2 were known to be CPU-/FPU-demanding but a Barton CPU probably has so much power (FPU faster than PPro/PII, high IPC, large L1 cache, large full-speed L2 cache, double-clocked FSB - also good PCI performance in newer mainboards) that it is equivalent to a medium-to-fast PII.
At the time of this review http://www.tomshardware.com/reviews/3d-accele … step,51-21.html
a PII 300 (Klamath) was the fastest you could get.

Could you do the same tests on the PCX2? 😀

Yes, I will do some for PCX2. When it has significantly better framerate than PCX1 CPU demand should be higher as well.
But first I have to correct myself. I forgot to force Quake into 32 bit depth, somehow I remembered it should use color depth of desktop, but it wasn't. So the picture is updated, now with real true color data and also sound, the integrated one. Default sampling rate of 11 khz was used, does the sampling affect PCI bandwidth?
I was kinda hoping the Barton at 66x3 will start where PII 300 would left. The consensus was PCX really needed best processors of it's time to show their strength, and I wanted to check how far it will go. I don't know weaker platform that would give me this kind of flexibility. Now I tested the 200 MHz Barton in software mode also, just like in the 686 comparison thread, but under Windows because DOS crashed. 24 FPS- it looks like too much compared to old Pentiums, I don't know.

You should also try to benchmark the performance difference between using a PCI host video card, AGP host video card and integrated host video card

I should get my PVR box going again and do 60fps PCX2 videos 😀

Putas wrote:

Yes, I will do some for PCX2. When it has significantly better framerate than PCX1 CPU demand should be higher as well.
But first I have to correct myself. I forgot to force Quake into 32 bit depth, somehow I remembered it should use color depth of desktop, but it wasn't. So the picture is updated, now with real true color data and also sound, the integrated one. Default sampling rate of 11 khz was used, does the sampling affect PCI bandwidth?
I was kinda hoping the Barton at 66x3 will start where PII 300 would left. The consensus was PCX really needed best processors of it's time to show their strength, and I wanted to check how far it will go. I don't know weaker platform that would give me this kind of flexibility. Now I tested the 200 MHz Barton in software mode also, just like in the 686 comparison thread, but under Windows because DOS crashed. 24 FPS- it looks like too much compared to old Pentiums, I don't know.

11 khz with 16 bit is 22 kB/sec so in theory negligible. The performance impact is probably from the CPU having to do additional work.

Flexibility: Probably not as flexible, but you could take eg. an S370 mainboard capable of FSB133, a Coppermine 533EB or 600EB and run it with 66, 100 and 133 MHz FSB, that might show some scaling without having to change the CPU. In your case there is nearly no scaling between 200 MHz and full speed.

The question is where the bottleneck really is. I remember something about texture uploads being slow and glQuake doing a lot of those for lighting. The RAM on those cards is only 32 bits wide. IIRC the PowerVR MiniGL had some optimizations to skip some texture uploads to gain performance.

Putas wrote:

Default sampling rate of 11 khz was used, does the sampling affect PCI bandwidth?

In theory yes, because all the samples are transferred over the PCI bus to the sound chip.
In practice however, even 44 KHz stereo 16-bit sound is only 176KB/s, so it's negligible.

hope this is 60fps

I should really do something to the shader stage loading function to parse blends first then upload texture after with some of the blending info to whack an alpha on

and also do the reverse by collapsing alphas on map textures that aren't for trans surfaces so there's no seethrough business on shiny walls

Yes, I should go with some proper age PCI VGA, Millennium maybe?
I thought sound requires lot more. What about the rumor of Sound Blaster Live hogging 1/3 of PCI, was that ever true?

idspispopd wrote:

Flexibility: Probably not as flexible, but you could take eg. an S370 mainboard capable of FSB133, a Coppermine 533EB or 600EB and run it with 66, 100 and 133 MHz FSB, that might show some scaling without having to change the CPU.

I don't see what news could such setup bring. Maybe B21ed Klamath or K6, with motherboard having many out of spec FSBs, but new rig is too much bother for me right now.

Thanks for the upload Putas. One thing I noticed from personal experience is that both on the PCX1 card I had(PC 3Dengine 1) and the PCX2 I have now (M3D) Tower 4 demo along with some of the other ones from the SDK would slowdown quite heaviliy when infinite plane shadows would come in till the point I was averaging just 26.9 on a single Barton MP2800 on a KT333 board with a V5 5500 AGP acting as host, whereas with just a P3 500 on a SOYO 6BA BX board and a 16mb Riva TNT the fps were quite more stable and higher than that?

Putas wrote:

I thought sound requires lot more. What about the rumor of Sound Blaster Live hogging 1/3 of PCI, was that ever true?

idspispopd wrote:

Flexibility: Probably not as flexible, but you could take eg. an S370 mainboard capable of FSB133, a Coppermine 533EB or 600EB and run it with 66, 100 and 133 MHz FSB, that might show some scaling without having to change the CPU.

I don't see what news could such setup bring. Maybe B21ed Klamath or K6, with motherboard having many out of spec FSBs, but new rig is too much bother for me right now.

Regarding SBLive: search for
VIA "sound blaster live" "bus parking"
for some explanations.

Regarding slower/faster CPUs: Maybe I don't really get what you are trying to measure. Your Barton@200MHz is already to fast to show much scaling at faster speeds. A 533EB@FSB66 would run at 266MHz, maybe this is slower than a 300MHz Klamath or Deschutes.
K6-2/-3+ is of course very flexible even without changing FSB, but than you'd have a slower FPU and probably no use for 3DNow! (not in glQuake, MiniGL, PowerVR drivers) so it might not scale high enough.

leileilol wrote:

hope this is 60fps

Playing in chrome it was 30 fps for me. If you right click on the video in youtube you can click stats for nerds it also says 720 x 480@30.
I think all of the 60fps youtube videos I've seen were at least 720p so its possible the resolution is the issue.

Correct, 60fps only supported in HD resolutions.

thanks for checking (not a chrome user here and never will be), looks like i'll have to upsample my videos to 720p from now on.

🙁

Screenshot time!

Dr. Brain requires KernelEx to run SGL, surprising for an earlier UE1 build. Also it asks "Do you want Unreal to use your PowerVR card?" 🤣 🤣 they missed that one

Star Trek Armada has a horrible clipping distance bug making the game completely unplayable. Kinda suprising given it's directly based off of Battlezone code which supports PCX2 just fine.

Midtown Madness 2 seemed to have undid some of the PCX2 workarounds MTM1 had.

I also hope this is 60fps

MORE!

Any test requests? 😀

- Must work on Win98SE at least, or probably through KernelEx
- Not Steam (no Win98 support for 9 years now so no go for that)
- Can't be an UnrealEngine game, i've think i've tested all of the ones with an SGLDrv. D3DDrv and OpenGLDrv is a no go on PCX2. UE2 is also a no-go
- Can't be a Glide game.
- If there's no demo available I probably can't try it.