S3 Savage 4 - number of pipelines

Thaks to all discussants.

I have one more question (hopefully last): Whats the difference between Virge DX, GX, GX2 and Trio3D? All of them supports EDO RAM, SDRAM and SGRAM, full-speed trilinear. GX2 supports AGP and integrates tv-out encoder, but are there any technological differences? Why is GX2 faster than Trio3D?
 
It's been a LONG time, but I remember something funky about the Trio3d, where you couldn't use all the ram on them for somethings. If I'm remembering right, they were the ones that had 8meg usually onboard, but act like 4meg cards when in 3d modes.
 
no-X said:
I have one more question (hopefully last): Whats the difference between Virge DX, GX, GX2 and Trio3D? All of them supports EDO RAM, SDRAM and SGRAM, full-speed trilinear. GX2 supports AGP and integrates tv-out encoder, but are there any technological differences? Why is GX2 faster than Trio3D?

Web Archive - S3 Home Assuming the archive is working at the moment, you can read from the source.

Well I know that DX was supposed to be used with EDO and GX was to be used with SGRAM. GX2 was AGP. Trio3D is probably basically a refined Virge. I also know that Virge DX/GX were 2nd gen Virge chips, being a lot faster than the original Virge (EDO) and Virge VX (VRAM). I'd have to dig out my old Computer Gaming World mags to give you some real details.

Virge VX actually was hardly faster than Virge EDO. In fact it may have been slower in 3D. I think I've even read that the 64-bit Vision series was superior in 2D to the newer Virge. I personally like to dig thru Google Groups for info on older stuff like this. Usenet is just the ultimate archive.

I did some googling for some old reviews you may find interesting:
Diamond Stealth 3D 2000 Arguably the most widely known Virge board. Still around on eBay constantly.
STB Nitro 3D review Virge GX using EDO. I had one of these for a month but it didn't get along with my Supermicro (Megatrends FX83A) motherboard and would freeze up even in 2D.
Compilation of info on old video chipsets
 
Last edited by a moderator:
And before this thread sails into the sunset, are there any takers on why the single and multi texturing numbers for the Savage 4 look perculiar for 2001SE? I feel like I should know whay this is happening, but I'm drawing a total blank here.

I wanna say: because the drivers are weird and don't support 2001SE properly, but that is just so rude and dismissive.
 
wireframe said:
And before this thread sails into the sunset, are there any takers on why the single and multi texturing numbers for the Savage 4 look perculiar for 2001SE? I feel like I should know whay this is happening, but I'm drawing a total blank here.

I wanna say: because the drivers are weird and don't support 2001SE properly, but that is just so rude and dismissive.

I don't think that is all that rude. The Savage 4 has always gotten less attention from developers because it was never seen as a succesful card. 2001SE might just be triggering a bug in the drivers. I haven't worked on a Savage in so long I've forgotten a lot. I just remember one of my friends that was into Unreal had one.
 
swaaye said:
Web Archive - S3 Home Assuming the archive is working at the moment, you can read from the source.
Thanks, this is very helpful

swaaye said:
Well I know that DX was supposed to be used with EDO and GX was to be used with SGRAM.
It seems, that both cores supported both types of memory modules. (e.g. STB Virge GX + EDO, or this Virge DX with SGRAM). So it's possible that DX and GX is actually the same chip(?). The second version: both support EDO/SGRAM, but only Virge GX supports SGRAM operations (masked writes, block writes...). Official statemant is "Synchronous memory support in the ViRGE/GX". (DX ~ 66/50MHz, GX ~ 66/66MHz)

wireframe said:
And before this thread sails into the sunset, are there any takers on why the single and multi texturing numbers for the Savage 4 look perculiar for 2001SE? I feel like I should know whay this is happening, but I'm drawing a total blank here.

I wanna say: because the drivers are weird and don't support 2001SE properly, but that is just so rude and dismissive.
Fillrate test in 3DMark 2001 is strange. The results are sometimes higher then theoretical fillrate (I noticed this when using FX5900XT, 3Dfx cards - see V3 or Banshee result). Savage 4 numbers can be distorted because of bandwidth limitation (1999/2000 use 16bit rendering, 2001 32bit rendering).
 
no-X said:
Fillrate test in 3DMark 2001 is strange. The results are sometimes higher then theoretical fillrate (I noticed this when using FX5900XT, 3Dfx cards - see V3 or Banshee result).
I've never seen this myself, but I will gladly take your word for it.

Savage 4 numbers can be distorted because of bandwidth limitation (1999/2000 use 16bit rendering, 2001 32bit rendering).
Ahh. The 16 v 32-bit argument makes sense. It is almost a perfect drop to half the throughput.

I have a Savage 4 lying around here somewhere and I used it briefly for UT "back in the day," but I don't recall it having problems with 32-bit textures. In fact, I quite distinctly remember the reason I used it was because it looked so good and ran fast (compared to some other cards available at that time). This was partially due to S3TC, I'm sure, but I am fairly certain it was in 32-bit mode and all the bells and whistles.
 
the maddman said:
I don't think that is all that rude. The Savage 4 has always gotten less attention from developers because it was never seen as a succesful card. 2001SE might just be triggering a bug in the drivers.
I meant it half-jokingly. But really, I was wondering why, specifically, this would be happening, if anyone knew the exact "bug" or reason. I think No-X's comment about 16 v 32-bit may be the answer there.

Otherwise I like being rude about the Savage 4 :p. It's a POS and I'll never get over how bad the drivers were (ok, so I'm over it already). You know it stinks when you need a new driver for every new game and then you realize "hey...there is no new driver...". heh. But it was great for Unreal Tournament. (until the Geforce came along and stomped on everything in a most menacing manner)
 
wireframe said:
I meant it half-jokingly. But really, I was wondering why, specifically, this would be happening, if anyone knew the exact "bug" or reason. I think No-X's comment about 16 v 32-bit may be the answer there.

Otherwise I like being rude about the Savage 4 :p. It's a POS and I'll never get over how bad the drivers were (ok, so I'm over it already). You know it stinks when you need a new driver for every new game and then you realize "hey...there is no new driver...". heh. But it was great for Unreal Tournament. (until the Geforce came along and stomped on everything in a most menacing manner)

Hey, I know that feeling. I gamed on a Matrox G400Max for quite a while. I bought a TNT before it, but the Diamond card I got had the crappiest analog filtering I'd ever seen before. I took it back, and got the Max. Great hardware, drivers took forever to get decent though.

I'd really like to like S3 or XGI or someone else, but so far they have all promised so much and delivered to little. Heck I can't even say "Savage 2000" without snickering.
 
S3 Savage3D 8MB was a fine piece of hardware. Can't forget all the hours i was playing UT99, Deus Ex, Tomb Raider 3, NFS3/4/5, Quake 3 (with Vertex Lightning for higher res) etc with it.
Hell even Motocross Madness 2 worked better and faster than years later on GF2 MX 32MB. Don't ask me how it just did. Then i overclocked it's memory a bit and it performed even better. I also have S3 Savage4 Pro laying somewhere around but i have it only as backup card since you can't gain any reasonable performance on XP platform (lack of proper drivers).
 
no-X said:
Savage 4 numbers can be distorted because of bandwidth limitation (1999/2000 use 16bit rendering, 2001 32bit rendering).

That's my thought, too. The two chips I know of which were 1x1 yet did multitexturing were the Savage4 and the Rage Pro (and derived chips). I couldn't get the Savage4 I had to work in that era (though years later with updated drivers I did manage it), but the Rage Pro showed a significant speed 32 bit mode for multitexturing over single texturing, so I imagine the same will be true for the Savage4.
 
Back
Top