Joseph_Joestar wrote on 2023-09-23, 16:49:

The justification could be that my card is using the R430 chip (110 nm) while most X850 cards use the R480 chip (130 nm). Theoretically, the R430 should be slightly easier to cool, due to the improved production process, but who really knows.

Well, that particular card is also R430, fully unlocked.

Update: R430 chip works around 65C on X850XT clocks. So X850 Pro reference cooler is dramatically better, while still being one slot.
Update 2: Ok, no, that's R480. Apparently the card has R430 bios flashed.

thanks so much joseph_joestar! I had used your 2 ZS Win98 guide in the past and I was now having issues with a regular Audigy 2 (SB0240) in another build. My understanding was that the drivers weren't the same between 2 and 2 ZS, put this post of yours saying the same procedure would work with the SB0240 made me follow your guide again - and it's a success! So...thanks again 😀

Boohyaka wrote on 2023-10-18, 12:59:

thanks so much joseph_joestar! I had used your 2 ZS Win98 guide in the past and I was now having issues with a regular Audigy 2 (SB0240) in another build. My understanding was that the drivers weren't the same between 2 and 2 ZS, put this post of yours saying the same procedure would work with the SB0240 made me follow your guide again - and it's a success! So...thanks again 😀

You're welcome! 😀

And yeah, I have successfully used the steps from that guide on Audigy 1 (SB0090), Audigy 2 (SB0240) and Audigy 2 ZS (SB0350) cards. The only difference is that you don't need to add a second SB16 emulation device on Audigy 1 cards.

Joseph_Joestar wrote on 2023-09-13, 11:03:

Anyway, I haven't tested it, but the performance issues are most likely caused by the 4 GB of RAM, not by the dual channel operation. So 2 x 1GB sticks should work just as well.

My 2x1GB memory kit just arrived and I can confirm the findings made earlier by @mockingbird. It's the 4GB memory amount that lowers performance under Win9x, not dual-channel.

I ran some tests with an AMD Regor platform on an AMD chipset with Win9x. AMD seems unaffected by this oddity. The highest performance was with 2 x 1GB sticks, but performance was almost identical with 2 x 2GB sticks, and likewise with 2 x 4GB sticks, only declining by a few points (between 5-10fps) on Q3A DEMO001. I should point out that Win98 seems to max out at 3.2GB with RLOEW's Patchmem.

Regor seems on par with Wolfdale and the Regor platform doesn't have that odd behavior where Windows won't boot when a USB flash drive is plugged in. I think I may ultimately go with AMD on this build.

Update: changed the sound card to an Audigy 2 ZS (SB0350) and moved the system to a Chieftec Libra case. I've also added two Noctua 120mm case fans for better airflow. Pics in the first post.

GPU temperatures have dropped by 2-3 degrees, so my X800 GTO now peaks at 76°C under full load. The entire system is also a bit quieter, mostly because the PSU is now mounted at the bottom, and no longer pulls hot air from the case.

I've been experimenting with Windows ME on this system, and I think I found a use case where it might have a slight advantage over Win98SE - the ability to turn off mouse acceleration.

Under WinME, this can be done from Control Panel > Mouse > Pointer Options > Accelerate. I haven't found a similar setting under Win98SE, at least not without using third-party programs. This is definitively an edge case, and I don't think a lot of people need this functionality. However, if you're using a modern USB gaming mouse on your Win9x retro rig, this is very helpful. I currently have my Logitech G403 HERO hooked up to this system and it's working flawlessly with acceleration turned off under WinME.

BTW, setting up WinME on this system relies on the steps for installing Win98SE, save for a few minor differences. First, I couldn't run WININIT on WinME (it would freeze) so I needed to manually edit SYSTEM.INI and reduce the maximum available RAM there (since I'm using more than 1 GB) and apply R. Loew's PATCHMEM later on. Next, NUSB 3.3 cannot be installed on WinME, but it is needed for USB 2.0 functionality. My solution was to extract the contents of NUSB33.EXE to a folder and manually update the driver for "PCI Universal Serial Bus" from Device Manager by pointing it to that folder. Seems to be working fine.

bloodem wrote on 2023-09-11, 11:42:

You're welcome! :-) I was actually going to mention the X800 cards and their coolers... I have a few X800XT and X850XT cards, an […]
Show full quote

Joseph_Joestar wrote on 2023-09-11, 11:11:

Thanks man! Couldn't have done it without your help! 😀

BTW, do you have any advice on cooler maintenance for X800 cards? I'd like to re-paste mine and possibly also lubricate the fan, but this thing looks like it wasn't built to be easily disassembled.

You're welcome! 😀
I was actually going to mention the X800 cards and their coolers... I have a few X800XT and X850XT cards, and they are annoyingly LOUD and HOT.
I'm not familiar with the cooler on your particular card, my cards have the typical plastic shroud (as seen in these photos from TechPowerUp).
Those are fairly easy to clean: once you remove all the screws from the card's back, the whole cooler/heatsink assembly comes right off, and on its bottom side it has a bunch of tiny screws that hold the whole plastic enclosure together.
I imagine that yours should be similar, even if it has a different appearance on the surface.

The best option I can think of for these cards would be the Arctic Cooling ATI Silencer 4 & ATI Silencer 5 coolers, but sadly they are very hard to obtain. I actually have quite a few NOS ATI Silencer 3 rev.2 coolers (compatible with the Radeon 9800XT cards), and I was thinking of trying to exchange one of them for a new ATI Silencer 4 and/or 5 cooler.

Go with Zalmans (VF900 or Fatality version). Last revision of Silencer 4 has flat heatsink surface., not sure why they did that. Previous versions were outdented on the GPU core. Also, they all tend to rattle with time. (I've seen posts where it was blamed on actual fan controller on the card.)

BTW, great build.
If core is R430, then X800XL makes sense.

stef80 wrote on 2023-11-14, 11:26:

BTW, great build.
If core is R430, then X800XL makes sense.

Thanks! 😀

I was kinda surprised how well everything turned out. For example, I have no unknown entries or exclamation marks in Device Manager (screenshots here) despite using a motherboard based on the P35 chipset. Even DOS games work (from within Windows) thanks to the Audigy's SB16 emulation. Sure, it was a bit of a hassle to set everything up, but the system is rock solid now.

Dumb question: can Deus Ex be played on XP?

stef80 wrote on 2023-11-14, 14:18:

Dumb question: can Deus Ex be played on XP?

Yup, and it works great there too.

However, it did come out in June of 2000 (well before WinXP) so it's often used to demonstrate late-era Win9x game requirements. Also, if you want to experience Deus Ex with A3D 2.0 then you pretty much need Win9x. Glide mode tends to work better there as well.

I was reading through this thread on Splinter Cell: Pandora Tomorrow and wanted to test that game on this system. Surprisingly, it appears to display almost all shadows and light sources correctly. Here are a couple of screenshots for comparison:

Dynamic shadows seem to be slightly softer on the Nvidia FX card, and there is some additional (cosmetic) shadowing around the window frame. Otherwise, not much difference. To me, the important thing is that all light sources and dynamic shadows render correctly on the X800 since they are used by the game for stealth. From what I've seen in Phil's video, those elements become broken on cards which use the unified shader architecture, meaning Radeon HD 2xxx and Nvidia 8xxx series or later.

My X800 using Catalyst 7.11 drivers passes all the relevant checks in the game's config utility, and delivers 30-60 FPS at the 1280x1024 resolution. I consider that fairly playable for a stealth game.

While I would have preferred to play this game on my GeForce FX 5900XT, the frame rate is simply too low at 1280x1024. With maxed graphical options, it often drops to 15-20 FPS, which is not great even for a slow paced stealth game. However, this system with the Radeon X800 offers a viable alternative and plays the game reasonably well at those settings. Just thought I'd share that in case anyone else is planning to revisit Splinter Cell: Pandora Tomorrow.

Hi Joseph_Joestar, could you run Final Reality benchmark on this build under Win98 and post results here? Can you also try with 4GB of RAM?

Thanks in advance.

mln wrote on 2024-03-13, 19:19:

Hi Joseph_Joestar, could you run Final Reality benchmark on this build under Win98 and post results here?

I'm not that familiar with Final Reality, so I just downloaded version 1.01 and ran the default benchmark, if that's what you were looking for.

Can you also try with 4GB of RAM?

No, because that was already proven to reduce overall system performance under Win9x on this rig.

This is exactly what I am looking for, especially the "Bus transfer rate" tab. Thank you for the effort!

Joseph_Joestar wrote on 2024-03-15, 13:32:

Can you also try with 4GB of RAM?

No, because that was already proven to reduce overall system performance under Win9x on this rig.

I asked to check with 4GB of RAM to see how it affects 2d / 3d transfer rates of this benchmark vs. 2GB RAM. I would appreciate if you could give it a try. 😀

My rig (6-core Xeon, 48 GB RAM, Radeon X550) gives transfer rates 10x worse than yours... Since I don't have 2GB RAM modules I can't test if it makes a difference.

mln wrote on 2024-03-15, 16:07:

I asked to check with 4GB of RAM to see how it affects 2d / 3d transfer rates of this benchmark vs. 2GB RAM. I would appreciate if you could give it a try. 😀

I'm unable to switch to the 4GB memory kit at the moment.

But if it helps, you can see the performance boost from dropping down to 2GB in this post compared to my initial benchmarks at the start of that thread. Based on that, I suggest getting a single stick of 2GB for your system to see if it improves things. Dual channel 2x1GB didn't seem to make a huge difference vs. single channel 1x2GB back when I did my tests.