VOGONS


First post, by mbarszcz

User metadata
Rank Newbie
Rank
Newbie

I picked up a Riva TNT2 M64 as a PCI graphics card for my socket 7 system (no AGP) to replace my S3 Trio 64V2/DX. The S3 is a pretty reasonable 2D card, but without a Voodoo in the system and having no 3D capabilities at all, it is limiting in Windows for just about anything that uses Direct3D. I know the M64 doesn't set the world on fire in terms of performance (especially when compared to the full strength TNT2, but for a 233MHz socket 7 PC with no AGP, it is one of the few affordable and still readily available PCI graphics cards with a good output and 3D capabilities in Windows. Plus being a later card (1999), if anything it is a bit on the overkill side, but still has great DOS compatibility and built in VBE 2.0.

Phils computer lab had a good video on the TNTs, https://www.youtube.com/watch?v=hnT2hPClb-0

I was also considering one of the "new" ATI Rage XLs (that use the old chips), but ultimately decided to go with the Riva TNT2 M64 instead.

20230731_221232.jpg
Filename
20230731_221232.jpg
File size
1.73 MiB
Views
1370 views
File license
CC-BY-4.0

So I got it home, and put it in my Socket 7 system (P/I-P55T2P4 Rev 3.10), and..... nothing but a black screen and 1 Long, 3 Short beeps (no video). Nothing. Dead I thought, so I looked around on Vogons and found a very long thread that got me thinking perhaps my the socket 7 board was not providing 3.3V to the card at all. I also found a video by Mark Furneaux about modding the ATI Rage XL PCI to solve the same problem https://www.youtube.com/watch?v=Gbyv4wI0v0s. This thread was also interesting: PCI Video card works in one PC, not in another

So I popped in my POST card, and sure enough, there was no 3.3v (Bottom Left Light)

PXL_20230804_014924383.jpg
Filename
PXL_20230804_014924383.jpg
File size
826.86 KiB
Views
1370 views
File license
CC-BY-4.0

I tried the TNT2 M64 in a later Pentium 4 machine and the card worked perfectly. I probed around on the card looking for the 3.3V and 5V locations on a working system and found there was a spot for a capacitor on the 3.3V line near the bottom of the board that was unpopulated (because the 3.3V regulator at the top of the board was also unpopulated. I guess with this being a Dell OEM card and it going into Dell systems that have 3.3V on their PCI slots, they deemed the 3.3V regulation circuitry unnecessary, so now it was time for an experiment. Could I feed in 3.3V to the board and have it regulate down to the proper 1.8V to power up the GPU? Also, would this back feed into the board and potentially into the other slots? Just to confirm this was the problem, I put the card back in the socket 7 machine and sure enough, there was no 3.3V being delivered to the regulator, and therefore no 1.8V for the GPU. So that was definitely the problem.

I didn't want to modify the board or have a bunch of wires hanging all over the case to power the GPU externally, and wanted to keep the mods on the card. So to test the concept, I soldered some wires onto the card to feed in 3.3V from my bench power supply.

20230801_215609.jpg
Filename
20230801_215609.jpg
File size
1.71 MiB
Views
1370 views
File license
CC-BY-4.0

The card fired right up and I was very pleased to find that the 3.3V never left the slot (my post card was in the PCI slot next to it and the 3.3v never lit up). I measured the power draw at between 5-6W (~1.6A) under various load conditions.

So with the proof of concept complete, I looked for a suitable 3.3V regulator that was both small and could provide enough current. Initially I had hoped a simple linear regulator would work to regulate the 5V down to 3.3V, but that would have to dissipate 2.72W, which was going to get hot fast, so I needed a good switch mode option. There are lots of el-cheapo no name adjustable buck converters on ebay/amazon/aliexpress, but their adjustable nature as well as poor noise (100s of mV)/regulation characteristics concerned me for this use case of trying to run a graphics card. The chintzy pots wander all over the place, and I wanted something with a fixed output voltage anyway. I found this tiny little Adafruit buck module based on the TI TPS62827. This Adafruit module outputs a steady 3.3V with a tolerance of 1%, can supply up to 2+ amps, and had very good noise characteristics. Under load on my bench, I measured only ~30mV PP noise at the 2.2MHz switching frequency under a 2A load. The ATX spec allows for 50mV ripple on the 3.3V rail, so this looked like a good fit.

I didn't want to permanently modify the card so it could only be used in a 5V slot. I wanted a way to disable the regulator when I didn't need it, and it was nice that the Adafruit module broke out the output enable pin. Simply pull it to ground and the regulator shuts off.

So I got the module, soldered some wires to it, covered the back with Kapton tape, and used UV resin to secure it to the board. I took the 5V from before the polyfuse that carries 5V to pin 9 of the VGA port, and fed it back into the unpopulated capacitor pads at the bottom of the board. I added a small switch on the top of the board that would ground the enable pin so the regulator could be easily switched off. There is also a light on the module that shows when it is running, so there is visual confirmation with a green LED.

PXL_20230804_014700702.jpg
Filename
PXL_20230804_014700702.jpg
File size
1013.82 KiB
Views
1370 views
File license
CC-BY-4.0
Last edited by mbarszcz on 2023-08-04, 02:48. Edited 5 times in total.

Reply 1 of 13, by mbarszcz

User metadata
Rank Newbie
Rank
Newbie

The board still works in my P4 system with the regulator off

PXL_20230804_014817673.jpg
Filename
PXL_20230804_014817673.jpg
File size
1.14 MiB
Views
1369 views
File license
CC-BY-4.0

And in my Socket 7 (5V) system with the regulator enabled (note the green light)!

PXL_20230804_015010326.jpg
Filename
PXL_20230804_015010326.jpg
File size
1.12 MiB
Views
1369 views
File license
CC-BY-4.0
PXL_20230804_015021088.jpg
Filename
PXL_20230804_015021088.jpg
File size
1.87 MiB
Views
1369 views
File license
CC-BY-4.0

Reply 3 of 13, by analog_programmer

User metadata
Rank Oldbie
Rank
Oldbie

Good job!

mbarszcz wrote on 2023-08-04, 02:34:

I took the 5V from before the polyfuse that carries 5V to pin 9 of the VGA port, and fed it back into the unpopulated capacitor pads at the bottom of the board. I added a small switch on the top of the board that would ground the enable pin so the regulator could be easily switched off. There is also a light on the module that shows when it is running, so there is visual confirmation with a green LED.

So, the green wire is for buck converter on-off switch (output?) to ground, the black wire is for ground input, the "lower" red wire is 5V input voltage and the "upper" red wire must be 3.3V regulated output - is this correct? This is the essential part of the mod and it will be nice if you explain what pads on the videocard's PCB was used.

Edit: I think I figured it out on the second reading

Last edited by analog_programmer on 2023-08-04, 12:54. Edited 1 time in total.

from СМ630 to Ryzen gen. 3
engineer's five pennies: this world goes south since everything's run by financiers and economists
this isn't voice chat, yet some people, overusing online communications, "talk" and "hear voices"

Reply 5 of 13, by mbarszcz

User metadata
Rank Newbie
Rank
Newbie
analog_programmer wrote on 2023-08-04, 05:39:
Good job! […]
Show full quote

Good job!

mbarszcz wrote on 2023-08-04, 02:34:

I took the 5V from before the polyfuse that carries 5V to pin 9 of the VGA port, and fed it back into the unpopulated capacitor pads at the bottom of the board. I added a small switch on the top of the board that would ground the enable pin so the regulator could be easily switched off. There is also a light on the module that shows when it is running, so there is visual confirmation with a green LED.

So, the green wire is for buck converter on-off switch (output?) to ground, the black wire is for ground input, the "lower" red wire is 5V input voltage and the "upper" red wire must be 3.3V regulated output - is this correct? This is the essential part of the mod and it will be nice if you explain what pads on the videocard's PCB was used.

Edit: I think I figured it out on the second reading

Sorry for the confusion, when you ground the green wire through the switch (output enable), it disables the buck converter.

Here's a labeled picture:

labeled.jpg
Filename
labeled.jpg
File size
1011.74 KiB
Views
1263 views
File license
CC-BY-4.0
ala_borbe wrote on 2023-08-04, 09:24:

3.3 volts are available elsewhere on the motherboard. find a coil from 3.3v and run a wire from it to the correct pin on the PCI slot.

mbarszcz wrote on 2023-08-04, 02:34:

I didn't want to modify the board or have a bunch of wires hanging all over the case to power the GPU externally, and wanted to keep the mods on the card.

Reply 6 of 13, by analog_programmer

User metadata
Rank Oldbie
Rank
Oldbie
mbarszcz wrote on 2023-08-04, 15:49:

Sorry for the confusion, when you ground the green wire through the switch (output enable), it disables the buck converter.

Yes, I have swapped the purpose of the two red wires, because again I did not read carefully that

mbarszcz wrote on 2023-08-04, 02:34:

I took the 5V from before the polyfuse that carries 5V to pin 9 of the VGA port

But from the picture with the explanations, it's all clear now. Thanks!

from СМ630 to Ryzen gen. 3
engineer's five pennies: this world goes south since everything's run by financiers and economists
this isn't voice chat, yet some people, overusing online communications, "talk" and "hear voices"

Reply 7 of 13, by Sphere478

User metadata
Rank l33t++
Rank
l33t++

Many motherboards lacking 3.3v don’t even route those pins to other slots.

There are actually two group sets of pins with 3.3v on the pci specification

Some were 5v on older gen mobo slots but later those pins changed to 3.3v which sounds like a nightmare but in practice hasn’t been a road block. For the final version of the 3.3v adapter we just ignored them. And let the motherboard do what it wanted with them.

See all you have to do really is feed power to the pins that only ever were marked as 3.3v on the motherboard and vola! 3.3v compatible pci slot.

On your card, backfeeding or parallelling the mobo/psu 3.3v or to other cards would’t likely be an issue as the voltage would be the same, the only ussue would be if the system didn’t have 3.3v, and DID route signal and other cards were drawing too much from the mod you did. (Over load it with too much draw) this could likely be solved by using a beefier mod power supply. See there was once such a thing as a 3.3v pci power supply card. It took I believe 5v and backfeed 3.3 back to the motherboard as the motherboard lacked the converter. Some of these systems actually used this to run the processor/s

It is worth noting that many gfx cards are very picky with socket 7 and 3.3v isn’t always the problem or only problem. I would say you got lucky here that it was the only incompatibility. Don’t take that wrong though, not saying trying cards is dangerous, just that there are a heck of a lot of ones that won’t post properly 🤣.

So having the mod on in your p4 system shouldn’t cause a problem. At least not one that I am aware of. It would basically be the same as if the card had come populated with the converter. If wired correctly it should be sending 3.3v to pins that would or already do have 3.3v anyway. So there is no conflict if potential is the same.

You could get into concerns of line noise or power supply quality. But chances are that isn’t going to be a problem, and it’s all speculation without a scope. Chances are that your power supply will be just fine for what you are trying to do. It may even be better than the supply that the card would have been populated with.

If you wanted to add a smoothing cap, you would hook up the circuit to a scope and analyze the interference frequencies and size a capacitor that responds best to each frequency

My go to for socket 7 is radeon 7000-9xx0 I see the most of those working of any cards I have. Anything newer isn’t going to work because of pci spec/bios/ or drivers.

On a system like yours you could probably get 800-900 3dm2000 with a 7500

Curious what you are getting with that card. Could you share? 😀

Sphere's PCB projects.
-
Sphere’s socket 5/7 cpu collection.
-
SUCCESSFUL K6-2+ to K6-3+ Full Cache Enable Mod
-
Tyan S1564S to S1564D single to dual processor conversion (also s1563 and s1562)

Reply 8 of 13, by mbarszcz

User metadata
Rank Newbie
Rank
Newbie

Here's some benchmarks for you:

Sphere478 wrote on 2023-08-04, 17:27:
Many motherboards lacking 3.3v don’t even route those pins to other slots. […]
Show full quote

Many motherboards lacking 3.3v don’t even route those pins to other slots.

My go to for socket 7 is radeon 7000-9xx0 I see the most of those working of any cards I have. Anything newer isn’t going to work because of pci spec/bios/ or drivers.

On a system like yours you could probably get 800-900 3dm2000 with a 7500

Curious what you are getting with that card. Could you share? 😀

TNT2M64 / Pentium MMX 233MHz / Win98

3dm99tnt2-m64.jpg
Filename
3dm99tnt2-m64.jpg
File size
31.48 KiB
Views
1183 views
File license
CC-BY-4.0
3dm2ktnt2-m64.jpg
Filename
3dm2ktnt2-m64.jpg
File size
37.66 KiB
Views
1183 views
File license
CC-BY-4.0

Here's the same card on the Pentium 4 system removing the CPU bottleneck
TNT2M64 / Pentium 4 2.53GHz / WinXP

p4-3dm99tnt2-m64.jpg
Filename
p4-3dm99tnt2-m64.jpg
File size
32.81 KiB
Views
1183 views
File license
CC-BY-4.0
p4-3dm2ktnt2-m64.jpg
Filename
p4-3dm2ktnt2-m64.jpg
File size
37.25 KiB
Views
1183 views
File license
CC-BY-4.0

3DMark 2000 Results Comparison

3dmark2k.jpg
Filename
3dmark2k.jpg
File size
42.88 KiB
Views
1183 views
File license
CC-BY-4.0

Certainly not the most impressive performance under Windows, although that was never really the intention with my Socket 7 233MHz Machine. If it will run under Windows on a 233MHz, it will likely run better under Windows on a PIII with much more CPU power.

However, the DOS performance and compatibility on the TNT2 M64 is quite good. In most cases, there is a performance boost over the S3 Trio 64V2/DX.

+-----------------------------------------+------------+------------+--------------+
| Benchmark | S3 64V2/DX | TNT2 M64 | % Difference |
+-----------------------------------------+------------+------------+--------------+
| Speedsys VESA Memory | 2048 KB | 16MB | 800% |
| Speedsys VESA Memory Speed | 22337 KB/s | 73849 KB/s | 331% |
| 2) 3DBench 1.0c (Faster PCs) | 165.5 | 176.6 | 107% |
| 3) Chris's 3D Benchmark (fps) | 155.6 | 275.2 | 177% |
| 4) Chris's 3D Benchmark 640x480 (fps) | 44.3 | 80.6 | 182% |
| 5) PC Player Benchmark (320x240) (fps) | 58.1 | 58.8 | 101% |
| 6) PC Player Benchmark 640x480 | 23.0 | 24.4 | 106% |
| a) Doom min details (Slower PCs) - fps | 308.6 | 321.9 | 104% |
| b) Doom max. details (Faster PCs) - fps | 84.1 | 95.0 | 113% |
| c) quake timedemo - fps | 52.0 | 53.0 | 102% |
| d) Quake timedemo 360x480 | 22.0 | 23.8 | 108% |
+-----------------------------------------+------------+------------+--------------+

Reply 9 of 13, by analog_programmer

User metadata
Rank Oldbie
Rank
Oldbie
mbarszcz wrote on 2023-08-05, 05:52:
Certainly not the most impressive performance under Windows, although that was never really the intention with my Socket 7 233MH […]
Show full quote

Certainly not the most impressive performance under Windows, although that was never really the intention with my Socket 7 233MHz Machine. If it will run under Windows on a 233MHz, it will likely run better under Windows on a PIII with much more CPU power.

However, the DOS performance and compatibility on the TNT2 M64 is quite good. In most cases, there is a performance boost over the S3 Trio 64V2/DX.

+-----------------------------------------+------------+------------+--------------+
| Benchmark | S3 64V2/DX | TNT2 M64 | % Difference |
+-----------------------------------------+------------+------------+--------------+
| Speedsys VESA Memory | 2048 KB | 16MB | 800% |
| Speedsys VESA Memory Speed | 22337 KB/s | 73849 KB/s | 331% |
| 2) 3DBench 1.0c (Faster PCs) | 165.5 | 176.6 | 107% |
| 3) Chris's 3D Benchmark (fps) | 155.6 | 275.2 | 177% |
| 4) Chris's 3D Benchmark 640x480 (fps) | 44.3 | 80.6 | 182% |
| 5) PC Player Benchmark (320x240) (fps) | 58.1 | 58.8 | 101% |
| 6) PC Player Benchmark 640x480 | 23.0 | 24.4 | 106% |
| a) Doom min details (Slower PCs) - fps | 308.6 | 321.9 | 104% |
| b) Doom max. details (Faster PCs) - fps | 84.1 | 95.0 | 113% |
| c) quake timedemo - fps | 52.0 | 53.0 | 102% |
| d) Quake timedemo 360x480 | 22.0 | 23.8 | 108% |
+-----------------------------------------+------------+------------+--------------+

The biggest upgrade in this case is replacing old videocard (2D pre-Virge "3D deccelerator"model) with some decent for its time 3D accelerator videocard.

You have done nothing wrong with this mod. You will not put more PCI videocards demanding missing 3.3V PCI line in this mobo, thus overloading the buck converter. And the modification of this cheap videocard is not permanent. A way better solution than mangling the motherboard.

from СМ630 to Ryzen gen. 3
engineer's five pennies: this world goes south since everything's run by financiers and economists
this isn't voice chat, yet some people, overusing online communications, "talk" and "hear voices"

Reply 10 of 13, by Sphere478

User metadata
Rank l33t++
Rank
l33t++

It is interesting how much better that card does on the p4

Reason I say this is with a different card I can get like 400 more points in 3dm2000 on basically the same pmmx 233 system. So a faster card still will help. But your slower card will beat my faster card on a faster system.

I’m sure driver overhead is at play here but I have a hard time believing your old geforce driver is more demanding than a newer radeon 9k driver. Though idk for sure. But it all does seem to suggest the card its self being faster does have a few punches left to throw into the fight. “Bottlenecks” I hate the term as it lacks any nuance. Yet again prove to be more gradients than brick walls.

Facinating.

Pentium MMX 233 aiming for the stars!

Sphere's PCB projects.
-
Sphere’s socket 5/7 cpu collection.
-
SUCCESSFUL K6-2+ to K6-3+ Full Cache Enable Mod
-
Tyan S1564S to S1564D single to dual processor conversion (also s1563 and s1562)

Reply 11 of 13, by badmojo

User metadata
Rank l33t
Rank
l33t

Great work OP! I have a FX5500 card here with the same issue on my Socket 7 motherboard so I might attempt something similar one day.

An aside on the M64 - the VESA support is great but there are situations when UniVBE comes in handy. UniVBE won't see a M64 but this handy patch came out recently to fix that problem: NVPatch - making UniVBE work on NVidia cards

Life? Don't talk to me about life.

Reply 12 of 13, by Kahenraz

User metadata
Rank l33t
Rank
l33t
Sphere478 wrote on 2023-08-05, 08:11:

It is interesting how much better that card does on the p4

My guess is that because this card is pre-TnL (pre-GeForce), it relies heavily on the CPU for certain operations which are simply "faster" due to the raw horsepower behind the P4 compared to the Pentium 233. There may also be SSE optimizations from Coppermine optimizations, depending on which driver was used.

I've read that this can be a problem with the Voodoo 2 on fast CPUs causing the chips to overheat, since they had never been tested on such hardware since it didn't exist.

Reply 13 of 13, by Deivison

User metadata
Rank Newbie
Rank
Newbie

I have this PCI video card, just a different model, I wanted to ask if you have already adapted this video card?

It simply doesn't show an image on my motherboard model M748LMRT 1.3A, I've already made different settings in the BIOS and nothing shows an image, I've already updated the BIOS and it doesn't show any signs of life, the video card works on another more current computer, only Not on this motherboard.

Attachments

  • s-l1600.jpg
    Filename
    s-l1600.jpg
    File size
    362.49 KiB
    Views
    525 views
    File comment
    Motherboard model M748LMRT 1.3A
    File license
    Public domain
  • IMG_20231120_200717633.jpg
    Filename
    IMG_20231120_200717633.jpg
    File size
    35.24 KiB
    Views
    525 views
    File comment
    PCI Nvidia Riva TNT2 M64 32MB
    File license
    Public domain