- This is almost certainly a memory density issue. Single-sided 256MB has 256Mbit chips, double-sided has 128Mb chips. I very much doubt you need to use double-sided RAM, but if you want 256MB, you don't have much choice, as the only way to get 256MB with 128Mbit chips is double-sided.
- Manual info on memory is an exercise in guessing how the engineers writing it translated technical stuff to something roughly human-readable and then translating that back into the proper specs. It's entirely possible there were no 256MB unbuffered SDRAM modules available for testing at the time when the QVL was being composed - so desipite 256MB being supported in theory and in practice (your original DIMM does work after all), it's not listed.
- high. A single-sided 128MB DIMM uses the same 128Mbit chips as a 256MB double-sided one, just 8 of them instead of 16.

Thank you for your answers! Another question though: I see a lot of Kingston 128MB Single-sided sticks, with only 4 chips on them. Will memory density be an issue with those as well?

Yes, those are 256Mbit (16Mx16) chips as well. Other exotics to avoid:single-sided DIMMs with 16 chips (either 16Mx4 or 32Mx4). Only a few mem controllers support them (Via basically). No idea about the Geode's controller but unless you want to confirm either way, avoid.

From my experience with my Cyrix Media GXm-266 board, it needs to use single sticks of 128 MB (chips on both sides of the module - double sided).

For 256 MB, you need to use two 128 MB sticks. I am able to use 256 MB CL3 just fine, but if I want to use CL2, I needed to reduce the total memory size for the system to be stable. So I run the system with 128 MB + 64 MB. I prefer the faster RAM over the larger RAM quantity.

I tried several sticks of SDRAM, but I could not get CL2 at 256 MB working. It was very frusturating.

Thanks for the info. I think I will go with 128MB in this case. It's a DOS/WfWG 3.11 machine, so it's overkill, but I'm not able to find smaller sized sticks that are CL2. Any idea what causes the instability with 256MB?

It is fairly common when pushing memory timings to the limit that the larger size RAM's won't work. The more RAM the system/chipset has to deal with, the more resources are required (or so it seems). I've run into this time and time again. I can run my IBM 5x86c-133/2x, for example, with only 1 ws on the RAM when using 64 MB, but when I up the RAM to 128 MB and maintain the 1 ws, it isn't so stable. The effect is sometimes not immediately obvious. Add 256 MB to the IBM 5x86c-133/2x system, and it is immediately obvious.

Concerning finding CL2 RAM on a MediaGXm. The front-side bus on this system is 33 MHz, not the 133 MHz that PC133 is accustomed to. So you should be able to run any non-buffered, non-registered CL3 PC100 or PC133 at CL2 on this system.

So I tried some sticks of RAM, wanted to share my findings. First I tried a stick of RAM of 128 MB, PC133, CL2, useable memory density. Worked at CL2, but not with an divider of 3. Unstable, games crash or do funny things like upside down cars that fall through the road.

Then I got my hands on a 32 MB stick of PC133 memory made for printers. No specs on the stick but the datasheet of the memory chips was promising. Tested it and it runs at CL2 with a divider of 3! Stable and a noticable performance gain and more compatible (Mega-Em etc).

I wonder if the manual was correct about the 128 MB RAM spec... with my previous configurations going beyond 128 I was not able to apply more optimal settings that should be in spec with all the used components.

@feipoa With the new RAM and FP_FAST (and a PCI videocard instead of the integrated one) I went from 15 to 27 fps in Quake and from a performance index of 35 to 43 in MDK. I got to squeeze a lot more performance out of this system with your info on RAM and register enhancements.

Yeah, its not bad if you can use a proper PCI 3D card. So you followed the guide in this thread then? Register settings for various CPUs Besides me, you're the only other person I know of who has optimised register settings on a MediaGX.

It may just be that 32 MB stick worked because of its size, rather than the specs of the stick.

Do you know what the differences is between a divider of 2 and a divider of 3?

Which graphics card are you using? I found the integrated graphics to be quite crummy in comparison to almost anything you put into the PCI slot.

Yup, I followed that guide. I currently only use FP_FAST. Some settings already seem optimal according to CTChip34, but I need to properly research it so I can push it more. Do you know any tools that can set register set CCR7? I want to enable NMI.

It's probably indeed the size of the stick, but I don't have any other sticks to do a <= 128 config, so can't try it.

I can choose between a divider of 4 or 3. The mainboard manual didn't give any info but another mainboard's manual did. It divides the cpu speed and that is your memory speed. There is some overhead though. A 300 Mhz cpu divided by 3 gives you 100 Mhz memory speed, but according to the manual it will run around 111 Mhz in practice. In my case I'm running at 233 Mhz divided by 3 = 77 Mhz. That way it also makes sense that PC133 memory is advised.

I'm using a S3 Savage4 with 8 MB SGRAM, works beautifully in DOS and Win31 with official drivers. The integrated solution isn't fast and compatible. Mode X/Y is slow and has issues. Not sure if this because memory speed/double transactions or the cpu taking a hit.

Why do you want to enable NMI? Force IRQs? Or you looking for memory parity checking? I haven't played with this at all. If it is for memory parity, I find on some systems that soft reset can be enough to generate and NMI error.

I don't recall what my BIOS options are for memory clock divisions, however, the GX1 databook mentions that,

The SDRAM clock is a function of the core clock. The SDRAM bus can be run at speeds that range between 66 MHz and 100 MHz. The core clock can be divided down from two to five in half clock increments to generate the SDRAM clock. SDRAM frequencies between 79 MHz and 100 MHz are only supported for certain types of closed systems and strict design rules must be adhered to. For further details, please contact your local National Semiconductor technical support representative.

I also found it surprising that , "The maximum memory size is 1 GB." and "The processor directly addresses up to 4 GB of physical memory even though the memory controller addresses only 1 GB of DRAM." Especially when it says that the board only supports two 168-pin unbuffered DIMMs.It is almost a guarantee that 1 GB has never been tested on a Cyrix MediaGX/GXm board.

Savage4, wow! Never thought I'd see this combination with a MediaGX. Are you using the Cyrix built-in audio and UltraDMA33 controller?

I've also got a Gravis Ultrasound in this system and one of its utilities, Mega-Em, likes to have less than 64 MB of memory and NMI enabled for Roland emulation purposes. With NMI enabled it should work in protected mode games.

Max. 1 GB sounds not realistic. Perhaps with conservative settings but that will come with a severe performance hit on this platform.

Yeah the system doesn't make sense from a period-correct perspective but it's all DOS / Win31 compatible and the additional parts (Savage4, GUS) complement it well. I use the integrated audio as well. UDMA is enabled but access is through the BIOS.

If I manage to get NMI going and find a way to prevent high DMA conflicts it would be perfect.

If you feel like loading Vogons, then 64 MB may be preferred if using Win9x. Vogons seems to load with 32 MB in Win 3.11 w/IE5.

I just checked my BIOS. I have SDRAM divisors of 3 and 4. I have it set at 3, so 266.666/3 = 89 MHz for the RAM. I'm also using CL2. THe 89 MHz is probably why I cannot run the full 256 MB w/CL2, but am limited to 196 MB. If I set the divisor of 4, the RAM is running at 67 MHz. I bet I could run CL2 with the full 256 in that case, but then 89 MHz >> 67 MHz. I really wish the GX1-300 I sourced had worked.

I suspect the MediaGX is a platform that will surprisingly become valuable in the next decade or two, especially the boards with PCI and ISA expansion slots. It was the first mainstream consumer board with this level of integration into the CPU.

I use the onboard UDMA ports for CD-ROM access, while I use a Promise UltraDMA 100 controller for the HDD. I'm not sure if it makes much of a difference on this system.

Is the Savage4 not period correct with the GX1? You're using the GX1, right? I'm pretty sure the GX1 was released around the time of the Savage4 - 1999.

I guess it's more period correct than I thought. Must be the BIOS date: 2002. Last night I used CTCHIP34 to enable/disable some other register bits; got Quake to run at 30.1fps and MDK with performance index 45. The tools to set register bits are confusing though, some list other options per register than others, which not always matched the datasheet. I think I enabled BTB, which according to your other thread isn't the most stable option. Anyway to test this under DOS/WIN31? And it seems that enabling/disabling EMM386 influences the registers your see in CTCHIP.

Your GX1-300 isn't working? I had a search running for one, wanting to install it in the thin-client system I had before. But it seems most variants of the 300/333 CPU's are soldered onto the mainboard.

I'd leave EMM386 disabled until the registers are set.

I beleive CTCHIP34 is for the Cyrix 5x86 and 6x86. I don't recall there being a MediaGX specific option. From memory, I think I used the 5x86 config file for the MediaGXm. I would refer to the datasheet for determining which registers are being altered, rather than rely on the register headings/descriptions offered by CTCHIP.

Yes, the GX1-300 is was vary hard for me to find in PGA format. My GX1-300 works, but I couldn't get it running reliably on my MediaGXm board. I should test it again using slower settings, like 233 MHz. The issue may be the SDRAM speed. I know I tried it at 300 and 266 MHz, but don't recall if I went slower. The issue could still be with the BIOS for my particular board.

Did you say you have a GX1 running? COuld you upload a photo of your motherboard and your BIOS? If similar enough, I might try out your BIOS.

I've got a NS Geode GX1 - 233 soldered on an Advantech POS-7551 motherboard. There are some low res pics on the internet, will take a pic when I have time. Attached is the BIOS with flash tool from the CD it came with. It's the same version as is running on my board.

You might be right about the memory speed being the problem with the GX1 300. I would lower the memory (CPU) speed and size and test again.

Great. Hopefully I will get to this in the next few weeks. Need a quality photo of the MB first, so that I can ensure the hardware components are similar. If not, I won't test the BIOS due to risk.

I attached a pic, it's a low res press release picture. Might be useful, will take a photo of the mainboard when I have time.

I'm now posting from my system using the GX1-300P. I'm not sure what I did wrong previously, but its working now. This is a 2.0 V chip, so perhaps I forgot to decrease the voltage previously? The GXm chips are 2.9 V. My motherboard supports down to 2.1 V and the GX1 seems to be working at this voltage.

Sweet! Still running it with 192MB of RAM?