Awesome. Just stumbled on a Medion GeForce 4 Ti 4200 128Mb for about £40. It looks like they have about 3-4 of them. Much cheaper than the one I was watching, which has a starting price of £60 but is only 64Mb. 😄

DustyShinigami wrote on Yesterday, 19:31:

Okay. In my case the BUS type is AGP and the Memory Base Addresses are D4000000 - D4FFFFFF and D6000000 - 07000007. That last one is different to what it says in the Device Manager. I'll try changing it from hardware to BIOS next and see what happens.

EDIT: Setting it from Hardware Enumeration to DOS worked fine. No crashes. But still get the same results with all of those errors. I also tried manually adding the address in the extended options, but it always says 'can not set environment variable 'CMDLINE'. Environment full?' And it still proceeds to set it to D6000000 - D6FFFFFF.

Wait, but that's off. That second range should be D6000000 - D6FFFFFF (16 MB). But the upper boundary here (07000000) is actually lower than the starting offset. That's exactly what MATS complained about: Start location is greater than end location. So there's a chance that your BIOS is improperly assigning resources to the graphics card, or that the graphics card is requesting an abnormal amount of memory, and that's why the tools stumbled. To test if the BIOS or the card is to blame you'd need a second motherboard to try it on.

When Windows boots, it re-scans the PCI bus and does its own assignment, possibly correcting the BIOS's mistake, so you had the correct addresses there.

Now the interesting question: If you disable that re-assignment by setting the PCI bus in Windows to BIOS enumeration mode, does it retain that suspicious D6000000 - 07000007 in the TNT's device manager resources tab? Or does it correct it to D6000000 - D6FFFFFF?

Something else I found in VMCE: Boot to Windows or DOS, put in the VMCE floppy, and edit A:\autoexec.bat. At the locations where it calls "vmtce.exe", change that to "vmtce.exe /Verbose:1". This will make it output a couple of extra lines on startup, in particular "VBE Memory Range" and "Found PCI video device: X bytes reported, X bytes found". It would be interesting to see what it's saying there.

What about the Geforce 4 MX, does it also have the wrong upper boundary under DOS in Hwinfo?

About that Medion card, nice. Just be aware that this one is underclocked and 10% slower than a regular 4200. It's basically the lowest end Geforce 4 you can get, but the difference isn't large and you can overclock it. One user also mentioned that it has good signal quality: Re: Anyone here knows or owns a MEDION GEFORCE 4 ti4200?

That, and not performance, would be my biggest concern, since cards from that era could have vastly different picture quality between manufacturers. I have two Geforce 4's and the cheaper model has a noticeably dimmer output than the higher end model.

asdf53 wrote on Today, 00:10:

Wait, but that's off. That second range should be D6000000 - D6FFFFFF (16 MB). But the upper boundary here (07000000) is actuall […]
Show full quote

DustyShinigami wrote on Yesterday, 19:31:

Okay. In my case the BUS type is AGP and the Memory Base Addresses are D4000000 - D4FFFFFF and D6000000 - 07000007. That last one is different to what it says in the Device Manager. I'll try changing it from hardware to BIOS next and see what happens.

EDIT: Setting it from Hardware Enumeration to DOS worked fine. No crashes. But still get the same results with all of those errors. I also tried manually adding the address in the extended options, but it always says 'can not set environment variable 'CMDLINE'. Environment full?' And it still proceeds to set it to D6000000 - D6FFFFFF.

Wait, but that's off. That second range should be D6000000 - D6FFFFFF (16 MB). But the upper boundary here (07000000) is actually lower than the starting offset. That's exactly what MATS complained about: Start location is greater than end location. So there's a chance that your BIOS is improperly assigning resources to the graphics card, or that the graphics card is requesting an abnormal amount of memory, and that's why the tools stumbled. To test if the BIOS or the card is to blame you'd need a second motherboard to try it on.

When Windows boots, it re-scans the PCI bus and does its own assignment, possibly correcting the BIOS's mistake, so you had the correct addresses there.

Now the interesting question: If you disable that re-assignment by setting the PCI bus in Windows to BIOS enumeration mode, does it retain that suspicious D6000000 - 07000007 in the TNT's device manager resources tab? Or does it correct it to D6000000 - D6FFFFFF?

Something else I found in VMCE: Boot to Windows or DOS, put in the VMCE floppy, and edit A:\autoexec.bat. At the locations where it calls "vmtce.exe", change that to "vmtce.exe /Verbose:1". This will make it output a couple of extra lines on startup, in particular "VBE Memory Range" and "Found PCI video device: X bytes reported, X bytes found". It would be interesting to see what it's saying there.

What about the Geforce 4 MX, does it also have the wrong upper boundary under DOS in Hwinfo?

About that Medion card, nice. Just be aware that this one is underclocked and 10% slower than a regular 4200. It's basically the lowest end Geforce 4 you can get, but the difference isn't large and you can overclock it. One user also mentioned that it has good signal quality: Re: Anyone here knows or owns a MEDION GEFORCE 4 ti4200?

That, and not performance, would be my biggest concern, since cards from that era could have vastly different picture quality between manufacturers. I have two Geforce 4's and the cheaper model has a noticeably dimmer output than the higher end model.

Ah-haa. Interesting. Hmm. I do have a second motherboard - my MSI MS-6156. The snag though is I don't really have any space to set it up. Hmm. It's probably one of those things I'd have to look at playing around with in the future. Though the other motherboard's BIOS really doesn't have many options to adjust, unlike my current one, which is much more customisable.

I'll have to check and see what the Device Manager reports now. It's still left at BIOS from PCI. Thanks for the tip regarding the autoexec file for VMCE; I'll give that a try later and report back the results. I'll also give the Geforce 4 another test and see what its addresses are.

And yeah, I did briefly read something about it being slightly underclocked and having good signal quality in that thread. I'm not sure if I've added a utility to my collection to overclock GPUs as it wasn't something I figured I'd need. But bumping back up by 10% should be easy enough. I'll make a note of the default clock speeds for that particular model.

asdf53 wrote on Today, 00:10:

Wait, but that's off. That second range should be D6000000 - D6FFFFFF (16 MB). But the upper boundary here (07000000) is actuall […]
Show full quote

DustyShinigami wrote on Yesterday, 19:31:

Okay. In my case the BUS type is AGP and the Memory Base Addresses are D4000000 - D4FFFFFF and D6000000 - 07000007. That last one is different to what it says in the Device Manager. I'll try changing it from hardware to BIOS next and see what happens.

EDIT: Setting it from Hardware Enumeration to DOS worked fine. No crashes. But still get the same results with all of those errors. I also tried manually adding the address in the extended options, but it always says 'can not set environment variable 'CMDLINE'. Environment full?' And it still proceeds to set it to D6000000 - D6FFFFFF.

Wait, but that's off. That second range should be D6000000 - D6FFFFFF (16 MB). But the upper boundary here (07000000) is actually lower than the starting offset. That's exactly what MATS complained about: Start location is greater than end location. So there's a chance that your BIOS is improperly assigning resources to the graphics card, or that the graphics card is requesting an abnormal amount of memory, and that's why the tools stumbled. To test if the BIOS or the card is to blame you'd need a second motherboard to try it on.

When Windows boots, it re-scans the PCI bus and does its own assignment, possibly correcting the BIOS's mistake, so you had the correct addresses there.

Now the interesting question: If you disable that re-assignment by setting the PCI bus in Windows to BIOS enumeration mode, does it retain that suspicious D6000000 - 07000007 in the TNT's device manager resources tab? Or does it correct it to D6000000 - D6FFFFFF?

Something else I found in VMCE: Boot to Windows or DOS, put in the VMCE floppy, and edit A:\autoexec.bat. At the locations where it calls "vmtce.exe", change that to "vmtce.exe /Verbose:1". This will make it output a couple of extra lines on startup, in particular "VBE Memory Range" and "Found PCI video device: X bytes reported, X bytes found". It would be interesting to see what it's saying there.

What about the Geforce 4 MX, does it also have the wrong upper boundary under DOS in Hwinfo?

About that Medion card, nice. Just be aware that this one is underclocked and 10% slower than a regular 4200. It's basically the lowest end Geforce 4 you can get, but the difference isn't large and you can overclock it. One user also mentioned that it has good signal quality: Re: Anyone here knows or owns a MEDION GEFORCE 4 ti4200?

That, and not performance, would be my biggest concern, since cards from that era could have vastly different picture quality between manufacturers. I have two Geforce 4's and the cheaper model has a noticeably dimmer output than the higher end model.

Okay, so I've tried my Geforce 4. I can only check with HWINFO and not the Device Manager in Windows. According to that the three memory addresses are: E4000000 - E4FFFFFF, D0000000 - D8000007, and D8000000 - D8080007. VMCE lists it as being D0000000 - D7FFFFFF.

For the TNT, it's still set to BIOS and the address is still D6000000 - D6FFFFFF.

I can't edit the autoexec file in DOS as it complains the floppy disk is full. I have to do it in Windows. I tried adding the verbose:1 command, but nothing different appeared to happen. Not sure if I need to add it in other places.

DustyShinigami wrote on Today, 13:44:

Okay, so I've tried my Geforce 4. I can only check with HWINFO and not the Device Manager in Windows. According to that the three memory addresses are: E4000000 - E4FFFFFF, D0000000 - D8000007, and D8000000 - D8080007. VMCE lists it as being D0000000 - D7FFFFFF.

For the TNT, it's still set to BIOS and the address is still D6000000 - D6FFFFFF.

I can't edit the autoexec file in DOS as it complains the floppy disk is full. I have to do it in Windows. I tried adding the verbose:1 command, but nothing different appeared to happen. Not sure if I need to add it in other places.

Okay, the values for the Geforce 4 are all plausible, nothing unusual happening. But for the TNT, it'd be really interesting to get that verbose output to see what VMCE is doing. The base offset (the lower boundary) it gets correctly, because otherwise, you'd get a black screen. But then there's two interesting parts, the memory size that it detects and the upper boundary for the address area -"bytes found" and "bytes reported" on the last line.

asdf53 wrote on Today, 17:30:

DustyShinigami wrote on Today, 13:44:

Okay, so I've tried my Geforce 4. I can only check with HWINFO and not the Device Manager in Windows. According to that the three memory addresses are: E4000000 - E4FFFFFF, D0000000 - D8000007, and D8000000 - D8080007. VMCE lists it as being D0000000 - D7FFFFFF.

For the TNT, it's still set to BIOS and the address is still D6000000 - D6FFFFFF.

I can't edit the autoexec file in DOS as it complains the floppy disk is full. I have to do it in Windows. I tried adding the verbose:1 command, but nothing different appeared to happen. Not sure if I need to add it in other places.

Okay, the values for the Geforce 4 are all plausible, nothing unusual happening. But for the TNT, it'd be really interesting to get that verbose output to see what VMCE is doing. The base offset (the lower boundary) it gets correctly, because otherwise, you'd get a black screen. But then there's two interesting parts, the memory size that it detects and the upper boundary for the address area.

Yeah, I believe that's the same line I added it to. Unless it's case sensitive? I used a lower case 'v'. No, tried it with uppercase, too. I'm also booting the utility from a floppy, which I presume is fine...? Are there any specific options that should be enabled/disabled for these extra options to appear? I don't get anything different when I go through them.

EDIT: Ahh, wait... I'm supposed to let it run option 1/defaults...?

Okay, this is what I get:

Yeah, that last block is different. The VBE is only checking 4Mb...?

Interesting. That's all as it should be, the PCI sizes are correct and the smaller VBE size is normal because it might only expose the area that's needed to draw into the framebuffer (screen area). But I just realized something else, I mentioned that it was seemingly poking around in the wrong memory area, but I might have been wrong there - that address (00FF0000) is probably relative to the physical address range, so it's starting at 0, and FF0000 would be just 64kb below the end of the 16 MB memory area. Maybe the card doesn't actually allow to go past that last bit, and VMCE has an error in calculating the memory size of the card.

But then again, it's reporting 16,777,216 bytes, and that's exactly FFFFFF, the same size that Windows has assigned. Looks correct. I wonder if the memory test under Windows doesn't really test the full range. When you run VMCE under Windows, does it also report the 16,777,216 bytes in size?

asdf53 wrote on Today, 18:19:

Interesting. That's all as it should be, the PCI sizes are correct and the smaller VBE size is normal because it might only expose the area that's needed to draw into the framebuffer (screen area). But I just realized something else, I mentioned that it was seemingly poking around in the wrong memory area, but I might have been wrong there - that address (00FF0000) is probably relative to the physical address range, so it's starting at 0, and FF0000 would be just 64kb below the end of the 16 MB memory area. Maybe the card doesn't actually allow to go past that last bit, and VMCE has an error in calculating the memory size of the card.

But then again, it's reporting 16,777,216 bytes, and that's exactly FFFFFF, the same size that Windows has assigned. Looks correct. I wonder if the memory test under Windows doesn't really test the full range. When you run VMCE under Windows, does it also report the 16,777,216 bytes in size?

I'll have to check a bit later. I think it probably does. I take it there are no additional options/tests that can be done under the Windows test? I get the same kind of errors/results even with the Geforce 4, so whatever issues it has with the TNT appear to be the same with the Geforce 4. Not sure if you've read through this thread; I merely skimmed it. I think the OP had similar issues with VMTCE and what it was outputting.

Accurately troubleshooting video memory faults with VMTCE

I quickly glanced over VMCE's algorithm it uses to determine the size of the graphics memory, and from what I understand, it tries to write two bytes to the VRAM every 1 MB and tries to read them back, and once it fails, it knows that that's the upper limit.

Now if just the beginning of that final MB is writable, but not the end, it will calculate a wrong upper memory boundary, which might be what's happening with your TNT. This is why it might fail to test the final 64 KB - that area could be off limits beyond the actual video memory range.

The Nvidia driver under Windows could exclude that area, so you won't get that error.

asdf53 wrote on Today, 20:15:

I quickly glanced over VMCE's algorithm it uses to determine the size of the graphics memory, and from what I understand, it tries to write two bytes to the VRAM every 1 MB and tries to read them back, and once it fails, it knows that that's the upper limit.

Now if just the beginning of that final MB is writable, but not the end, it will calculate a wrong upper memory boundary, which might be what's happening with your TNT. This is why it might fail to test the final 64 KB - that area could be off limits beyond the actual video memory range.

The Nvidia driver under Windows could exclude that area, so you won't get that error.

I see. Where's that option found? In the Nvidia Control Panel?

I would not imagine that to be an option, that would be internal to the driver. It's just a theory - I don't know if that's a thing in practice. If you run the video memory test under Windows, it goes through the graphics driver to determine the available memory. So if the memory size reported there is smaller than 16,777,216 bytes, that could be a hint.

asdf53 wrote on Today, 20:23:

I would not imagine that to be an option, that would be internal to the driver. It's just a theory - I don't know if that's a thing in practice. If you run the video memory test under Windows, it goes through the graphics driver to determine the available memory. So if the memory size reported there is smaller than 16,777,216 bytes, that could be a hint.

Ah, gotcha. And yes, going through the test in Windows, it lists it as 16,035,840. It also decreases after the test. It's gone down to 15,690,240.