Jackal1983 wrote on 2023-07-19, 23:37:

his.lordship wrote on 2023-07-19, 13:59:

feipoa wrote on 2016-06-02, 09:27:

What speed MediaGXm are you using? Benchmark results for all GXm speeds are reported in the Ultimate 686 Benchmark Comparison. […]
Show full quote

What speed MediaGXm are you using? Benchmark results for all GXm speeds are reported in the Ultimate 686 Benchmark Comparison. For example, the FPU ability of a GXm-266 is about that of a Pentium 133, while the ALU ability is similar to that of a Pentium 166. On the other hand, if you look at Quake (software mode), the GXm-266 equates to a Pentium 100. For Quake II, it is similar to a Pentium 120. The DOOM results for the GXm-266 are similar to a Pentium 200 or a Pentium MMX 166.

I found the integrated graphics to have horrendous quality, so I went with a PCI Banshee card. The integrated audio also sounded terrible, so I went with an ISA sound card. There is an ultra DMA driver, which works in Win98SE, but it was a pain to get it working. Ultimately, I use a PCI Ultra100 card.

There are some performance tweaks for the GXm. I have outlined them here, Register settings for various CPUs
One such tweak is enabling FP_FAST, which helps the FPU results. For example, DOS Quake scores went from 33.1 fps to 36.2 fps after enabling FP_FAST on the GXm.

Basically, the GXm has speeds similar to the Cyrix 5x86, but are about 10% slower on account of not having any L2 cache. What the GXm has going for it is MMX. What I found surprising was that they didn't bother to fix branch prediction, not even with a GXm at 266 MHz. Yet branch prediction works even on the 6x86-P90.

Hi,

Regarding the UDMA driver, did you ever get it to work? I am testing an IGEL-4 thinclient with board MB-5BLGP its the same as a neoware eon. I am using a 44 pin IDE hdd. When i fisrt updated the driver it gave me conflict in device manager and hdd stop showing on device manager under the disks section. Then since this board only has 1 ide connector i set the controller driver settings to only use the primary IDE, rebooted and conflict went away, this disk showed up again and gave me the option to enable DMA. After enabling and rebooting the system hangs on the windows 98 boot screen, rebooting again and pressing escape there are no messages just hangs with blinking cursor. thanks.

I've had the same issue with a ECS P5GX-M. I'm guessing the reference drivers are hit and miss for DMA compatability and the OEMs modified them to work for their individual use case. I'd see if you could hunt down the driver package that shipped with your thin client.

Thanks. I dont think there are much different drivers around, unfortunately. Besides the official driver package i have a couple loose files but i think they are the same. Now that i think about it seems that the driver package doesnt even contain the udma driver, i think i grabbed it from Philscomputerlab website. Anyway i am now testing on this paricular machine, it seemed like an easier mod for fitting my gravis pnp but only now i am encountering some posts about gus incompatabilities as well, however didnt get there anyway, lets see how it does 😀 i am trying sort this dma thing now, however this is just to say that i have another less common thin client with the same cpu chipset combo, an esesix thintune, that i recall being able to use the dma without issues, strange. Other things that i would like to ask, is regaring the atto disk speed test, i made a test and the max speed i got was just below 1.7MB are these speeds normal even with not using DMA? i feel windows 98se a bit heavy for this machine, i dont remeber it being like that on the esesix machine. I intend to try now with a compact flash card to see if dma works or the atto speed test are different.

Just forget about the on-board graphics, it's terribly slow. If you install a dedicated video and sound card you should get about pentium 100-120 performance out of your GXm200. I have one such build using a new old stock ECS P5GX mainboard I found on ebay - I'm currently running a GXm 266, S3 Virge DX, SB16 and a Voodoo 2. Most 1997-1998 games are perfectly playable, especially if you manage to enable branch prediction. The only games I'm having minor issues with are Command and Conquer and Red Alert windows 95 versions, and I have a hunch the problem is related to the IDE controller. The games freeze for a millisecond when loading in sound or graphics assets, but once it's loaded everything into ram both games run fine.

his.lordship wrote on 2023-07-20, 08:04:

Jackal1983 wrote on 2023-07-19, 23:37:

his.lordship wrote on 2023-07-19, 13:59:

Hi,

Regarding the UDMA driver, did you ever get it to work? I am testing an IGEL-4 thinclient with board MB-5BLGP its the same as a neoware eon. I am using a 44 pin IDE hdd. When i fisrt updated the driver it gave me conflict in device manager and hdd stop showing on device manager under the disks section. Then since this board only has 1 ide connector i set the controller driver settings to only use the primary IDE, rebooted and conflict went away, this disk showed up again and gave me the option to enable DMA. After enabling and rebooting the system hangs on the windows 98 boot screen, rebooting again and pressing escape there are no messages just hangs with blinking cursor. thanks.

I've had the same issue with a ECS P5GX-M. I'm guessing the reference drivers are hit and miss for DMA compatability and the OEMs modified them to work for their individual use case. I'd see if you could hunt down the driver package that shipped with your thin client.

Thanks. I dont think there are much different drivers around, unfortunately. Besides the official driver package i have a couple loose files but i think they are the same. Now that i think about it seems that the driver package doesnt even contain the udma driver, i think i grabbed it from Philscomputerlab website. Anyway i am now testing on this paricular machine, it seemed like an easier mod for fitting my gravis pnp but only now i am encountering some posts about gus incompatabilities as well, however didnt get there anyway, lets see how it does 😀 i am trying sort this dma thing now, however this is just to say that i have another less common thin client with the same cpu chipset combo, an esesix thintune, that i recall being able to use the dma without issues, strange. Other things that i would like to ask, is regaring the atto disk speed test, i made a test and the max speed i got was just below 1.7MB are these speeds normal even with not using DMA? i feel windows 98se a bit heavy for this machine, i dont remeber it being like that on the esesix machine. I intend to try now with a compact flash card to see if dma works or the atto speed test are different.

The only thing I didn't do was follow the directions in the driver package (putting the UDMA driver files on a floppy and all) to the letter. I didn't do any speed tests, but W98se wasn't a dog by any means. I was running a 266mhz GXm at 300 mhz with a /4 ram speed divider (lets you run the full 256 megs with speed similar to a 266 mhz with a /3 ram speed divider) however. I agree with the above poster: yeet the internal graphics. They were a truly interesting glimpse of the future, but were barely adequate at the time for basic office shit. I'd recommend a Virge DX with some reasonably fast ram (so you can OC the core, it helps a lot) for good dos and 2D Windows compatability. Another option might be an Dell OEM Velocity 128, which would give you some appropriate 3D acceleration. The only reason I'd stick with the internal graphics would be if you had a Voodoo 1. It might be worth it to save a PCI slot.

Sorry for the delay; I've been on holiday.

From my recollection, the UDMA drivers that shipped with this motherboard weren't available or were not working properly. Attached are the updated drivers I tested years ago which worked with UDMA. I don't recall if I first installed the Cx5530 all-in-one driver first, or just went straight to the UDMA driver folder. I recall it gave me a little trouble, but ultimately worked once I determined how to make Windows 98 with respect to changing the IDE drivers.

Hi, I'm dealing with similar problem on Neoware 3000, which is also Geode GX1 platform. https://www.parkytowers.me.uk/thin/neoware/Eon/
I believe I tried all available UDMA drivers, including drivers from the last post, but none of them is able to start. I see the error of IRQ conflicts with USB chip on IRQ15. If I disable it, the UDMA driver is still unable to start but now complaining to Realtek NIC instead. There is no option in BIOS to disable it though.

Please, help if you saw a similar problem before.
Thanks!

Might help to provide photos of your BIOS setup screen. Look for manual configuration of IRQs, not auto.

Thank you for the response.
Not much to show, it's all loaded by default.

For the IRQ settings not much to choose from. It just toggles between manual assignment for ISA

You want to play around with the options in "Resources controlled by: Manual". Take it off Auto and adjust the resource availability until IRQ 14/15 are freed up for your IDE controller. You may also need to disable the USB Controller and Onboard Parallel Port before the BIOS reassigns IRQ's to where you need them. You may need to enable "Reset Configuration Data" when you've made these changes.

This is a very common problem, particularly with certain motherboards of this era. I ran into the same issue on my MediaGX board and a Tyan socket 7 board. If you continue to have grief, then you may also need to disable the serial ports and the onboard audio. Swapping PNP cards to a different slot also sometimes triggers the BIOS to reassign IRQ's.

Thank you. So far I tried following:
1. Tried various boot media, including CF cards (which I'm sure support UDMA) and real IDE HDD.
2. Disabled All unused peripherals, including built-in Audio, COM, LPT ports, USB. Can't disable NIC (no BIOS option) and even with COM ports off they are still detected.
3. Switched to the Manual IRQ mode and set IRQ15 to ISA
4. I use Reset Configuration Data, every time I change IRQ settings.
5. Not sure the correct settings of PCI IRQ Actived by Level or Edge, but I tried them both.
6. Also tried your MB BIOS, my system Neoware 3000 doesn't POST with it.

At this point the best result I could achieve is this - Driver for IDE controller initialized and channel drivers fail (no resource conflict shown).
Do you have any other ideas?

What error do you get under the yellow exclamation?

I have been at a similar point before and I do not quite remember what I did. Try going into safe mode and deleting the IDE controller. Then when you reboot Windows 98, be sure to select the drivers I shared in this thread.

From what I sort of recall, my Win98 installation was tainted with the wrong drivers I had used initially and I think I reinstalled Windows 98. Then I was sure to use the above attached drivers. I don't recall if I used the "all in one" drivers for UDMA+sound+vga, or the individual driver packages. You may need to try both with a fresh install of Win98.

On that screen there is no message, other than a driver couldn't run. No HW conflicts or other explanation, just said to reinstall the driver.

Still keep fighting. So far no good news. Reinstalled fresh Win98 several times, each time playing with various drivers and BIOS settings. This time it's vanilla Win98SE, (not a quick install image with patches that I used previously).
Despite many attempts, I wasn't able to find a magic combination. If you happen to have access to your PC, could we cross check your BIOS settings? so this way I'll narrow down my options.

Finally got around to this. Here's images of my BIOS on ECS P5GX-M.

And this is what your Device Manager should look like. You might need to manually install the National drivers once the files are extracted.

Thank you! Will get it another try!
What I found that default drivers at lease give me 2MB/s, while failed Cyrix drivers (with a bang) are 0.2MB/s. That's an improvement for sure, but it's still slow.

Is your BIOS detecting your hard drive correctly? Do you also have a CD-ROM connected to the IDE port?

Unfortunately, still no luck. My BIOS looks similar to yours, but not exactly. As I wrote previously, I tried to burn and use your image but it doesn't post with it.
I don't have a CD rom connected. It's a Neoware 3000 - thin PC, no room for CD rom. DOM is removed too.
I started looking for any alternative BIOS versions, and figured that Neoware 3000 is similar to IGEL WinNet III.
I found this thread IGEL WinNet III - need BIOS and got one of those BIOS images working. I don't see much changes, some options have new names, but this didn't fixe my problem.

Maybe the issue is related to the hardware implementation on the motherboard.

If there is a DiskOnChip chip, unplug it as it conflicts with IDE devices. And the performance just isn't great in general - the CPU is doing a lot less each clock cycle than a Pentium. Also there probably were small revisions between Cyrix -> NatSemi -> AMD Geode GX which might make thingy not 100% compatible. My Igel uses the NatSemi CPU

Hi Folks, after weeks of trying back and forth, I was able to find a working combination. I post my notes and screenshots here in case somebody else has to deal with this.

To recap - my platform is Neoware 3000, which is 300MHz GX1.

Summary:
1. You only need drivers from feipoa post: Re: Cyrix MediaGXM performance fix? and only the installer that does all at once. Individual drivers are not needed.
2. No BIOS modifications or changes required, at least, this is 100% default settings on my Neoware 3000.
3. All tricks made with how IDE drivers are configured.
4. I confirm this steps worked fine with IDE HDD, CF 2GB industrial card. I couldn't get it to work with 64GB IDE DOM by some reason (could be that it doesn't support all ATA commands)

Steps:
1. Stock Win98 install or patched USB Quick Install with integrated drivers.
2. Install all-in-one National drivers. During install it will ask for directory, point it to Program Files, National - see the screenshot

3. Reboot. IDE drivers won't start and you'll see a bang next to it. Now the trick is to open preferences, you'll see resource conflicts

4. This is not correct. Change it to manual configuration, - the only option that you can adjust is the memory range address at the very end. This is what we need! For my system I found (100s of experiments) that 940 starting address. DON'T RESTART.

5. Once you click Ok, two new child devices will appear with yellow bangs

6. Those are stock drivers, that we need to change to National ones. Click update the driver and manually select National driver.

7. Do that for both without restart. Then reboot.

Once you do that once, it's straightforward. But I still have no clue how to find I/O Range properly, 0x940 works for me but might be different for you.

That process feels somewhat familiar. It is really round about and difficult to remember those steps unless they get documented. Thank you!

Did you first try these steps with the separate National drivers, found it didn't work, then tried with the all-in-one driver pack?