douglar wrote on 2025-12-08, 12:27:

MattRocks wrote on 2025-12-08, 11:05:

Useful, but why test the DX6 cards under DX9c conditions? Depending on what the drivers do, I think that might be a handicap for […]
Show full quote

douglar wrote on 2025-12-08, 03:21:

It did some tests on the rv100 with a K6-3 400, P3-800 and an Athlon 2500+ recently.
Re: K6-III Socket 7 AGP vs PCI Tester The RV100 cards I had were not thrilling. I wouldn’t recommend them for DOS or Win31. There were some compatibility issues under DOS, didn’t have Win3.1 drivers, and they were all discount boards with at best average video quality. If you are going for the discount boards, I’d say the MX 4000 or an FX 5200 gives better bang for the retro $$ under Windows 95.

From my own personal experience, the ATI mach 32 with the fast ramdac was my card of choice from 1992 to 1995. Fast and had 8514/a hardware support for out of the box compatibility. Also had the best picture quality on the market at the time and you could save your refresh rate settings to the card.

Useful, but why test the DX6 cards under DX9c conditions? Depending on what the drivers do, I think that might be a handicap for some cards; FX5200 is notorious for being a DX8 chip sold as DX9 compliant due to driver software emulation.

In terms of capabilities the RV100 VPU aligns closely with TNT2 and Rage128Pro - definitely no T&L/DX7 stuff on any of them. Your test compared 128bit TNT2 and 128bit Rage128Pro against 64bit 7000.

I only did an anecdotal "test" with PC-BSD (an old operating system). I just booted them and shut down - so more a check than a test. The Radeon DDR/9550/9600/9700 all displayed beautifully; while the Rage128Pro/7000 both display the same software glitches around hardware mouse pointer, which I am led to believe is a known Rage128 driver issue (why the Rage128 driver decided it should drive the 7000 is not known - I just assumed the software knows what it's doing). I didn't test further than that as I was just sorting my cards, but it shows something worth further investigation. My 7000 is the 128bit variant, but VRAM behaviours would not be stressed on loading a desktop.

Radeon DDR was still R100 and when I was testing, the R100 family it still had a lot of the issues found in the Rage 128, like some of the 640x480 modes displayed incorrectly, etc.

That is really insightful:

Rage128 gets complaints in DOS.
R100 gets complaints in DOS.
So nobody bothers to test the RV100 in DOS.

That is rational, but inconclusive for one very simple reason: The RV100 is a simple architecture for OEMs that demanded stability, and our little investigation suggests it has no direct ancestor.

I just now checked Linux drivers and RV100 (Q1 2001), according to Linux devs, is the first ATI card to fully support multiple displays - not the R100 (feature disabled) and not the RV200 (came later). Whatever those devs found suggests the RV100 provides an evolutionary beginning, not an evolutionary dead end.

douglar wrote on 2025-12-08, 12:27:

I just tested a lot of cards that I got on the cheap, at first to see if PCI vs AGP mattered in the super socket 7 era. The answer was it didn't matter that much on Super 7 motherboards other than a 10-15% drop off in DOS pixel flinging.

That is kind of expected. Most first generation AGP cards were PCI native VPU and VRAM packaged onto an AGP PCB (the only "true AGP" card is the Intel i740). Intel patented AGP when memory was very expensive and Intel imagined using system RAM as VRAM would present big cost savings, but the price of memory dropped and the original AGP dream was redundant.

douglar wrote on 2025-12-08, 12:27:

But then I expanded the tests to AGP 4x and AGP 8x and by the time you get to AGP 8x motherboards, not only was there a difference in PCI vs AGP performance, but the PCI performance was actually worse on a 2005 motherboard than it was in the Super 7 motherboards.

Useful to know. PCI degrades even more with PCIe native motherboards and their "stuck on" PCI-to-PCIe adapter chips.

MattRocks wrote on 2025-12-08, 13:54:

That is rational, but inconclusive for one very simple reason: The RV100 is a simple architecture for OEMs that demanded stability, and our little investigation suggests it has no direct ancestor.

I just now checked Linux drivers and RV100 (Q1 2001), according to Linux devs, is the first ATI card to fully support multiple displays - not the R100 (feature disabled) and not the RV200 (came later). Whatever those devs found suggests the RV100 provides an evolutionary beginning, not an evolutionary dead end.

I don't know what investigation you're referring to... I thought we just determined that there hadn't been one? There is no evidence that the RV100 is anything but a cut down, budget oriented chip made after the R100. As for multiple displays... I don't think this is news either? Even the first paragraph of the Radeon wiki page says that the RV100 and RV200 were the first to support dual displays. They were made after the R100, so ATi had time to implement this feature, where it had previously not been a priority (Geforce 256 and I believe Geforce 2 GTS\Ultra also did not support dual displays).

Remember, RV100 was released 10-11 months after R100. That was a huge amount of development time back in those days. RV100 came out closer to the release of R200 than R100, so it's logical that they would implement some of the features that had been developed by that time. Also, from a budget perspective it makes sense for it to get additional features that were deemed to be relevant for value-oriented cards. For example, an office user may want dual displays while never needing something like hardware T&L or high 3D performance.

To summarize: ATi puts out their cutting edge R100 GPU to compete with Nvidia\3dfx and to show the world that they are relevant, focusing on performance and buzz-worthy features. Once that has been established in the enthusiast market, they cut apart and rearrange R100 to make a lower tier product to compete with the now 7-month-0ld Geforce 2 MX (that supports dual displays...). In doing this they cut seemingly unnecessary stuff to save die space and include improvements\features that customers will hopefully see as value-adds. Then they sell that product to a different market at a lower price point as the RV100.

There is no reason to think that there was some secret team locked in a Canadian basement with only access to DirectX 6, tasked with developing the best DX6 card ever, only to have it released as a budget DX7 card a few years later... looking suspiciously like a cut down but slightly more up-to-date version of the R100.

Sorry, this is just starting to sound a bit weird. Remember, this discussion started with a list of facts that we have determined were entirely made up. 😅

This is the direct quote from Wikipedia:

“ It is the basis for a variety of other succeeding products, including the entry-level RV100 (Radeon VE or Radeon 7000) and higher-performance refresh RV200 (Radeon 7500), the latter two having dual-monitor support.”

Wikipedia says “the latter” (RV200).
Linux drivers say, the former (RV100).

That is not consensus. That is a complete contradiction. Either all the facts align, or they don’t.

By our little investigation, I mean there is a smoking gun that the long held beliefs don’t match physical observations.

What we have done is turn over enough stones to show that nobody has published a comprehensive review: The Rage128 BSD drivers that claim the RV100, the RV100 die photo showing no ghosts of R100, the RV100 having features R100 did not.

And, that means common conceptions are guesswork. Isn’t that the real issue?

My physical card is a GeCube Radeon 7000 128bit VRAM (eight chips, four each side). I found an original retail listing on NewEgg, with discrepancy in PCB colour (red vs black) and mismatched marketing (128bit vs DDR) so the one sold by NewEgg is possibly a sibling rather than exact match.

I found AMD start with drivers from 2006, which reflects 5 years of lost data. I have not found the original driver CD, but there some possibles on Archive.org to check later (I’m writing on my iPhone).

All these observations suggest one thing - nobody cared enough to be thorough, and records are not 100% accurate.

MattRocks wrote on 2025-12-09, 03:16:

This is the direct quote from Wikipedia: […]
Show full quote

This is the direct quote from Wikipedia:

“ It is the basis for a variety of other succeeding products, including the entry-level RV100 (Radeon VE or Radeon 7000) and higher-performance refresh RV200 (Radeon 7500), the latter two having dual-monitor support.”

Wikipedia says “the latter” (RV200).
Linux drivers say, the former (RV100).

That is not consensus. That is a complete contradiction. Either all the facts align, or they don’t. You can’t have your cake and eat it.

It says the latter two . Which is either referring to RV100 and RV200 (rather than R100) using poor\unclear grammar (because it is wikipedia) or it is referring to Radeon 7000 and Radeon 7500 using unclear\inaccurate wording (because it is wikipedia). Since this is just wikipedia, one of us can easily just go and correct the wording to reflect more clearly that the RV100 (in either the Radeon VE or Radeon 7000) was always known to have multi-monitor support.

People knew that from the beginning and it was a notable feature of the card (search any of these links for monitor), since prior GPUs did not have it. Dual monitor support was a sign of it being a more modern design than previous generations, which it was, since it came out closer to the release of newer designs like the 8500. Also, notice that most of them have DVI and VGA ports, right from release day. It's not so much the sign of a conspiracy as it is open knowledge from 2001.

I think I'm going to have to bow out of this discussion though as I don't think it's going anywhere, even when I put time and effort into providing answers with citations, to the best of my ability. If you feel this is important enough to be worth investigating at all, it would be best served by creating a dedicated thread. That way if anyone is going to put in the time and effort to provide actual data, it can all be collected in one place to serve as a reference for anyone curious about this in the future. If you do make a thread, I would plainly state from the beginning what the goal of the thread is and why it demands investigating.

Fair point. It does say latter “two”, but RV100 and RV200 are not related and released at different times.

RV200 is a refined R100 - RV100 is not. I raised the RV100 as a wild card for DOS because I have seen BSD Rage128 drivers claim the RV100 and drive it.

Whether BSD was right or wrong, I don’t know. I then get dragged into quagmire over what the RV100 actually is and we find nobody bothered to document it!

Wikipedia is not a credible citation - it is by definition a secondary source that collates from other sources (often also secondary sources).

The Wikipedia article is glossing over the RV100, probably because every article they can cite also glosses over the RV100. They recognise RV100 existed, but they find RV200 a lot easier to cross reference.

I dug out the attached. It does not describe the random OEM variants shipped for budget PCs, but the Apple approved variant. Apple is always picky.

See anomalies? I’m not saying the document is right - I’m saying the record is lazy! We know more about what RV100 is not than what it actually is.

Whelp, I looked at the DOS Compatibility Chart and saw that Mario Shareware is one of the very few areas where Radeons tend to be obviously less compatible than a Rage 128. Personally, I don't find this super useful since no one plays Mario shareware, but since it'd be a very quick test and we have nothing else to work with I took a quick look through my stockpile and found a good assortment of cards to compare so we can add some actual new testing data. I have a hunch that no amount of data will truly "settle" whatever is going on here, but here is what I have taken time to add to the pit... err... pool!... of knowledge of whether the RV100 more closely resembles the Rage 128 or Radeons in its behavior. This is at least vaguely relevant to this thread so I don't think it's too out of place here.

(also, I love to do testing like this, and it gives me an excuse to dig through my bins of parts and type some DOS commands. 🤩)

Test system:
850Mhz PIII Coppermine (100x8.5)
Wintac 440BX
256MB PC-100
MS-DOS 7.1 (no notable drivers or TSRs running; very bare bones Windows 98SE "NO-GUI" freshly booted MS- DOS environment.)

(The test, as shown on a Rage 128 Pro)

Here's the data!!

Cards that did not give a black screen in Mario Shareware:

Starting from the top left and going across... any with no notes simply worked fine with no notable issues.

Radeon HD 5450 (uses a PCI bridge chip; possibly the last PCI Radeon; bottom half of screen would wobble a bit when moving but otherwise worked fine)
Dell Mobility Radeon PCI... thing. (By far the most surprising since everything online says the M6 is based on Rage 6... however, it is from 2003 and likely has many differences from the original Radeons)
Rage 128 Pro PCI
STB Powergraph 64V S3 Trio64v+
Nvidia Geforce2 MX 200 PCI
Rage 128 Pro AGP

Cards that did give a black screen in Mario Shareware:

Starting from the top left and going across... all of these would simply display a black\blank screen (not "no signal", just blank), while the PC speaker still played game sound effects. Escape returned to the DOS prompt as normal.

Radeon 7000 PCI (RV100)
Radeon 7000 PCI (RV100)
Radeon 7500 PCI (RV200)
Radeon 9800 Pro (R350)

Sadly, the beautiful original Radeon DDR (R100) that I have (DDR SDRAM model from 2000 with an aftermarket Zalman cooler) turned out to have a small ding in the PCB which had broken a trace, so I was unable to test that. Since all of this testing pretty much mirrors the results shown on the DOS charts, I would expect the R100 to also display a black screen.

Anyway, I'm not going to bother analyzing the results. Just dumping this information out into the universe.

Thank you for sharing, that is awesome detail that adds value. Completely ruins the RV100 as DOS dark horse but I’m glad it’s finally on record 😀

Ozzuneoj wrote on 2025-12-09, 07:03:

By far the most surprising..

Mobility Radeon, designed for laptops? Laptop chips need to work under all conditions or the whole laptop gets an RMA - mobility BIOS may contain critical fixes for VESA/DOS?

P.S. Something I noticed when looking at the R100/RV100 dies - they have different scaling: You can see this is in the pins around the outside that solder to the PCB, the RV100 has bigger pins. In fact the RV100 has many larger or courser components. And, feature wise, the RV100 adds multiple timing displays. Less, more, completely rearranged, and packaged differently? This implies complete redesign. I feel the words "cut-down" may have been marketing to shield RV100 from scrutiny, and given lack of records that appears to have succeeded. Comparing RV100 to R100 might be like comparing Windows 3.1 to Windows NT3.1!

Ozzuneoj wrote on 2025-12-08, 01:33:

Ah, yeah, the 320x240 version of the test I linked above has the Rage 128 Pro at the top. https://www.vgamuseum.info/images/vla […]
Show full quote

marxveix wrote on 2025-12-08, 01:00:

Rage128 Pro is one of the fastest cards for dos, someone has tested it, but its the speediest in lower resolutions as i remember.

Ah, yeah, the 320x240 version of the test I linked above has the Rage 128 Pro at the top.
https://www.vgamuseum.info/images/vlask/bench/quake320.png

Still, a 205fps vs 199fps lead over cards that seem to be significantly better in other situations isn't a huge win.

... it is still a win though. 😅

I havent seen these bencmarks before, first i saw Rage128Pro DOS results here: 10 Way DOS Graphic Card Benchmarks on Celeron 333

MattRocks wrote on 2025-12-09, 09:18:

P.S. Something I noticed when looking at the R100/RV100 dies - they have different scaling: You can see this is in the pins around the outside that solder to the PCB, the RV100 has bigger pins. In fact the RV100 has many larger or courser components. And, feature wise, the RV100 adds multiple timing displays. Less, more, completely rearranged, and packaged differently? This implies complete redesign. I feel the words "cut-down" may have been marketing to shield RV100 from scrutiny, and given lack of records that appears to have succeeded. Comparing RV100 to R100 might be like comparing Windows 3.1 to Windows NT3.1!

I just noticed this part of your post. There is no conspiracy here! The RV100 is a smaller die, that is why the pins look larger. I thought this was very clear from the beginning. Manufacturing silicon wafers on a cutting edge process is extremely expensive. Rearranging the chip and cutting\adding features to fit the market need is exactly how companies create a "cut down" product that requires less silicon for a lower price segment. Just let it go. 😅

I mixed up two things: Internal bond pads and external BGA pins.

I'm not interested in conspiracy. I'm interested in strategy.

I see no reason to entertain the idea of RV100 descending from Rage 128 instead of R100, mostly because of Environmental bump mapping.

Putas wrote on 2025-12-22, 06:21:

I see no reason to entertain the idea of RV100 descending from Rage 128 instead of R100, mostly because of Environmental bump mapping.

Oh boy, get ready. 😅

Feature overlap indicates reuse, not descent.

I arrived at the view that RV100 is mutually exclusive of R100 and Rage128 because the three dies show RV100 has a different implementation. RV100 satisfies the same DirectX contracts without the structural continuity of either architecture. Maybe some IP is reused, but where do you draw the line on reuse that is inherited from Rage128 or R100, and reuse that is inherited from wider industry?

In my experience, I have found the Matrox MGA cards to be fantastic in terms of DOS/Win3.X/Win95 compatibility if you have a pci slot. Another suggestion is the 3dfx Voodoo 3 2000 PCI

J-Tech95 wrote on 2026-01-26, 23:56:

In my experience, I have found the Matrox MGA cards to be fantastic in terms of DOS/Win3.X/Win95 compatibility if you have a pci slot. Another suggestion is the 3dfx Voodoo 3 2000 PCI

That's my experience as well. In fact I only came across older DOS titles that a Millennium or a Mystique has issues with by reading about these issues here.
Personally I've never played anything an a MGA-card that did not work fine. But there are some (more than for several other cards) issues, especially with older titles. But tbh: Tose are best played on a pre-PCI system anyway.

I have a Riva128 PCI in my Cx5X86 PCI system and it works flawlessly with anything DOS/Win16/Win32 I throw at it..

appiah4 wrote on 2026-02-03, 08:27:

I have a Riva128 PCI in my Cx5X86 PCI system and it works flawlessly with anything DOS/Win16/Win32 I throw at it..

You are lucky to have one. Not sure if it's because the Riva 128 cards are loved or if they are just getting scare but it seems like the value of those Riva PCI cards is approaching voodoo 3 levels. I was very happy to find a Diamond viper V330 PCI in a socket 7 Dell Dimension this year. I've also seen Dell branded Riva 128zx AGP cards in Dell 440BX systems.

Would Dell have sold the system new with a Viper v330 PCI?

I have mixed feelings about the Riva128 because it was one my major purchasing mistakes/regrets back in the day. I was trying to purchase an "STB Systems Velocity 100" for a build for a friend's budget gaming build in late 1999 and accidentally purchased an "STB Systems Velocity 128"

The Riva 128 might be good at some things, and the card might have looked a lot like the 100, but it didn't like the 3dfx drivers and certainly wasn't up for playing Half life 1.0 on Windows 98. The driver bundled with the card produced all sorts of washed out banding colors, was a real disappointment for me. Newer drivers would have likely fixed the issue though.

What is wrong with a Riva TNT 2 pci or agp?

Very strong cards , and way overpowered for dos / win95 era pc s.

For win 98 , either a Geforce 4 mx 440 , or a geforce 4 ti 4200.

bitzu101 wrote on 2026-02-05, 08:24:

What is wrong with a Riva TNT 2 pci or agp?

Very strong cards , and way overpowered for dos / win95 era pc s.

For win 98 , either a Geforce 4 mx 440 , or a geforce 4 ti 4200.

Nothing wrong with those cards at all. Certainly more than adequate. Do you feel that they are the best cards?

"The best card" really depends on what the rest of the setup is to be honest..