VOGONS


Worst fastest early 3D cards

Topic actions

Reply 140 of 248, by 386SX

User metadata
Rank l33t
Rank
l33t
MSxyz wrote on 2024-03-26, 07:05:

My vote goes to the SiS 6326. It was slow, but image quality and feature support was pretty good. Can run Unreal at 320x240. A couple of site dedicated to vintage hardware have images of this board allegedly running Quake III using a MiniGL of sorts. So it's probably not the worst fastest but the best slowest 😁

The SiS card is surely on the list I'll buy. I hope to find the better versions considering how many ones it seems they've been built. I'll receive the Trident 9750 these days.

Reply 142 of 248, by 386SX

User metadata
Rank l33t
Rank
l33t

Had some problem with the PSU and configuring the i440 again to use the early slower P-III/II if this last CPU works and test more games and bench. Today I might even receive the Trident card still has to arrive.

Last edited by 386SX on 2024-03-31, 07:36. Edited 1 time in total.

Reply 143 of 248, by 386SX

User metadata
Rank l33t
Rank
l33t

I was testing the same card with the first mainboard I tried (i440 and early P-III) and I'm quite convinced that the card and even the audio card get worse depending on the chipet/board compared to the slower ALi chipset on Socket 7 with a simpler Pentium-MMX cpu. Even the rendering seems a bit less stable and the audio card (even tried a Creative Vibra ISA) have problems. Maybe the CPU is too fast or maybe this card it a bit more compatible in few mainboards only, I'll do more test on this but it may be interesting such idea. The ALi chipset board seems the "best" config found for now.

Last edited by 386SX on 2024-03-30, 17:39. Edited 1 time in total.

Reply 144 of 248, by 386SX

User metadata
Rank l33t
Rank
l33t

I was reading the old 1996 Alliance Semiconductor press releases and it's interesting the line about their texture filtering feature I was reading before in the vintage3d review

"Proprietary Pseudo-Bilinear Filtering (PBF) technology for faster, smoother images"

Meanwhile, the 3Dimage9750 AGP 4MB card has arrived.. had to fix four chip pins a bit pushed against the others, they forgot to say this I imagine before buying it, anyway classic cheaper version, 1998 and SGRAM probably 10ns. I'm going to test this card too. I also bought a SiS6326 (H0) AGP 8MB with main chip built in 2001 and SDRAM 6ns on usual cheap PCB but let's hope high clocks.

Last edited by 386SX on 2024-03-30, 19:04. Edited 3 times in total.

Reply 145 of 248, by 386SX

User metadata
Rank l33t
Rank
l33t

While maybe too much slow for 1997/98, the 3Dimage9750 seems enough "accurated" in old rendering which probably costs a lot in speed but still good. The driver probably make also the difference. 2D quality is indeed a bit better while 3D where it works, render quite well. Newer test like 3DMark99/2000 doesn't seem to run ok unfortunately. I'll try the usual games and benchmarks. Final Reality seems good and accurated for its time, maybe a bit slow. Quake based games seems having problems rendering, Thief Demo only works with Z-Buffer disabled @ 400x300 not bad. EDIT: Powerstrip report a memory clock (and only that) of 83 Mhz (overclock to 91Mhz seems ok). It doesn't seem to read AGP speed anyway, I try to reinstall the Intel INF drivers.

I've found a different driver (middle 1999) beside the last one on the Trident website. I'm going to test it to see if something improved. EDIT: The 1999 driver found is a bit slower than the last one (which seems having a date older than this one which may be some beta or whatever) and lack the Texture trilinear mipmapping in Final Reality check list when the last one has that! Final Realiy for that decrease the image quality score. Also the audio card seems having problem with this graphic driver.

EDIT! Interesting, a beta 9750/9850 driver can actually render 3DMark99 MAX in-game texturing and also have Bump Mapping test 1 enabled which before could not with final driver! While it's slow for such heavy bench, Final Reality scores is almost the same only a bit slower than the previous driver I tried. 3DMark99 score is 173 3DMarks.

Last edited by 386SX on 2024-03-30, 21:54. Edited 8 times in total.

Reply 147 of 248, by 386SX

User metadata
Rank l33t
Rank
l33t

As promised. Trident 3Dimage9750 AGP 4MB SGRAM @ 83Mhz default (up to 91Mhz easily) late 1998 chips on a Pentium II-350, i440, 256MB PC100, Win 98A latest updates.
Usually "last" driver (not really the last) can't render the in game texures after few seconds and I think also the Bump Mapping 3-pass first test if I'm correct. These series are the automatic Result Browser screenshots. The driver tree version is not the main/last one but this is probably a later beta of the previous version tree and compiled later in July 1999.

Attachments

Last edited by 386SX on 2024-03-30, 21:19. Edited 8 times in total.

Reply 148 of 248, by 386SX

User metadata
Rank l33t
Rank
l33t

Last Result Browser automatic screenshots with the Trident 9750 AGP (SGRAM 83Mhz)

Attachments

Reply 149 of 248, by 386SX

User metadata
Rank l33t
Rank
l33t

These ones instead are taken with Fraps (frame rate doesn't show in the corner) so to see it's really during the in-game scenes rendering it and the Bump Mapping test which also run into 3DMark2000 (with better quality too).

Attachments

Reply 150 of 248, by 386SX

User metadata
Rank l33t
Rank
l33t

Oh... it seems with this driver also Half Life is running REALLY well (Direct3D 400x300) with texturing and lights instead of disabling them before.. this is awesome! Too bad Thief game seems having instead the usual Z-Buffer problem but still working with it disabled and @ 400x300.

Screenshots soon... 😁

I only wish the aT3D had this type of driver support to show that maybe it could do something similar.. but anyway good job Trident! The fact that even 3DMark2000 runs is impressive and quite accurated too for that period!

Reply 151 of 248, by 386SX

User metadata
Rank l33t
Rank
l33t

Half Life and Trident 9750 AGP SGRAM (it has 64bit ram bandwidth with four modules in line, I've seen similar cards with two only modules and empty pads for the remaining one) and Direct3D, while not perfect I'd say I'm impressed considering no tweaks are enabled in console using this beta driver, final version could not do this.

Attachments

  • HL1_Trident_9750_1.jpg
    Filename
    HL1_Trident_9750_1.jpg
    File size
    134.74 KiB
    Views
    368 views
    File license
    Fair use/fair dealing exception
  • HL1_Trident_9750_2.jpg
    Filename
    HL1_Trident_9750_2.jpg
    File size
    113.88 KiB
    Views
    368 views
    File license
    Fair use/fair dealing exception
  • HL1_Trident_9750_3.jpg
    Filename
    HL1_Trident_9750_3.jpg
    File size
    107.24 KiB
    Views
    368 views
    File license
    Fair use/fair dealing exception
  • HL1_Trident_9750_4.jpg
    Filename
    HL1_Trident_9750_4.jpg
    File size
    122.66 KiB
    Views
    368 views
    File license
    Fair use/fair dealing exception
Last edited by 386SX on 2024-03-30, 22:32. Edited 3 times in total.

Reply 152 of 248, by 386SX

User metadata
Rank l33t
Rank
l33t

Trying the Trident card @ 95Mhz 100Mhz memory overclock... go! 😁 The video chip seems following the overclock with the heat increasing while SGRAM modules only warm of course. Final Reality 3D scores seems like 10% higher. Need to double test the real increase.

Last edited by 386SX on 2024-03-31, 07:35. Edited 3 times in total.

Reply 153 of 248, by Hoping

User metadata
Rank Oldbie
Rank
Oldbie

At least the Trident seems to have the necessary capabilities, the image quality seems good, it hurts about the lack of performance, it could have been really useful, but even if the overclock really works, the performance seems quite bad.
For me, playing a game at less than 640x480 with a "3D accelerator" is of no use to me.
So another 3D decelerator to the list in my opinion. 😀

Reply 154 of 248, by 386SX

User metadata
Rank l33t
Rank
l33t
Hoping wrote on 2024-03-30, 21:28:

At least the Trident seems to have the necessary capabilities, the image quality seems good, it hurts about the lack of performance, it could have been really useful, but even if the overclock really works, the performance seems quite bad.
For me, playing a game at less than 640x480 with a "3D accelerator" is of no use to me.
So another 3D decelerator to the list in my opinion. 😀

Well, in that time I wish I could have played at 400x300 with this level of quality beside frame rate; 3DMark is of course really too heavy but Half Life, Thief and similar games seems ok. 😀 I know it's not fast but for late 90 games I'd have played with this video accelerator. The Rage IIC may have better drivers but rendering quality I'd say is better here.

Anyway overclock to max value seems still stable @ 100Mhz Fill Rate in 3DMark2000 is 23.0 Mtexel/s, 400k Triangles with the P-II 350. It's running Final Reality now, I can already say frame rate is higher for sure! I wonder if there're newer bios for these cards. I'm reading the bios version is 03/06/1998.

Reply 155 of 248, by 386SX

User metadata
Rank l33t
Rank
l33t

Oh.. found few hidden tweak options in the driver registry values. EDIT: I was expecting more but there are few interesting ones, for disabling 32bit rendering, another that possibly might be for Motion Compensation but doesn't work of course in MPEG2 players, there's a Direct3D performance value that seems like could be related to some control panel fast setting, tried increasing the value but no changes, I'll try decreasing it. There're TV out interesting values and GUI related accelerations. Some values that are zero might be there for the 9850 and the Blade3D.

Reply 156 of 248, by 386SX

User metadata
Rank l33t
Rank
l33t

Installed a Pentium II 400Mhz into this i440BX board. Beside the boring L2 cache reading and disappearing at reboot problem and magically becoming full Celeron (the P-III doesn't have this problem) to see a bit of change, 0.15 Final Reality score increase. Now testing @ 100Mhz overclock for the video card memory and testing this benchmark again to see how much it increase with overclocking and always as before the lowest audio setting (8bit, 22Khz, mono).

I'm going to test this nice 9750 card for a while but don't worry Alliance aT3D, I'll be back!

EDIT: yes the overclock is really working, screenshots soon for the Final Reality 3D scores.

Reply 157 of 248, by 386SX

User metadata
Rank l33t
Rank
l33t

Trident 9750 AGP with 100Mhz SGRAM (overclocked) difference in Final Reality

Attachments

Reply 158 of 248, by 386SX

User metadata
Rank l33t
Rank
l33t

Trident 9750 and Final Fantasy 8 seems having more than a problem..

Attachments

Reply 159 of 248, by 386SX

User metadata
Rank l33t
Rank
l33t

Anyway, meanwhile going back for a while to the aT3D to do some more combination of tests. Did you know that the PCI Retries flag into the control panel was there specifically for the Sound Cards problems at first? It seems clear that it doesn't like PCI sound card at first and it's better to try different older ISA card I imagine and seems confirmed to work better with those. I'm trying without no other PCI and ISA cards so to have a better idea.