It's my understanding that selling anything on this forum is specifically barred... But were such a machine available for purchase somewhere for around 100-150$, I would purchase one...

No problem, the ones I built are not for sale on here. They were built for customers needing such new hardware to support DOS / Win 9x with old hardware ISA controllers. I was just curious if anyone was interested in such a system that doesn't use a 'standard' CPU and board. This stuff is available from time to time on EBay and to me the benefit of NOS outweighs several other factors. You could put together what I described for around the price you mentioned. I may take one of the systems I have tucked away and put in a PCI video card, ISA sound card and play with it. IMO this route might be interesting and productive. I can post pics or other info if there is interest.

Oh, yeah, this route would be wonderful given how many people want to put together hardware from around the 1997-era. That CPU speed is just about perfect for most DOS games, and fast enough for most 90's Windows games.

If the price is right, and with that microATX board size, yeah, quite a number of people should be interested.

How compatible are they with games, though? I thought Geode was descended from MediaGX which had all the peripherals on the chip to simplify motherboard design.

They are exactly that but... we have two ISA slots and two PCI slots so we just won't be using the onboard graphics or audio. As far as executing x86 code they are 100% compatible even doing MMX. They lack L2 cache and their FPU's are not as strong as Intels but I don't see needing an FPU. I mean I'm not going to play Quake on it.

Where could one buy this product?

Motherboard

http://www.ebay.com/itm/BRAND-NEW-RARE-ECS-P5 … D-/300957930626

CPU

http://www.ebay.com/itm/GEODE-GXM-200P-OEM-CPU-/111169245462

I would send them offers, not full price buyout.

One very important thing to remember about these, they are Socket 7 but they are NOT compatible with Intel CPUS and vice-versa.

I may have some time tomorrow to play with one, at least by Tuesday if all else fails.

I am willing to bet these vendors are sitting on top of several hundred units each. A group buy would result in cheapest price. Heck I might even be willing to loan a board and CPU to someone with lots of older hardware to really try the thing out properly.

For a performance comparison of this motherboard with MediaGX/Geode CPU's in the 133-300 Mhz range, refer to the Ultimate 686 Benchmark Comparison. The fastest easily obtainable CPU for this board is 266 Mhz. With this CPU installed the system's FPU capability is about as good as a Pentium 133, or ALU between a Pentium 166-200. Interesting is that the ALU is only as good as a Cyrix 6x86-133 Mhz. If slow is what you are after, you can run it at 150 Mhz to achieve the FPU equivalence of a P75. This is a pretty versitile board for a socket7-gxm - 2 PCI, 2 ISA, 256 MB SDRAM. I have a system setup around this board and use an ISA Sound Blaster, ISA network, PCI graphics, and PCI IDE ATA. Not using the chips integrated controller features makes this like a Cyrix 5x86 system with a faster clock, MMX instructions, and SDRAM support. The main loss is the lack of L2 cache. I do beleive I finally managed to get the onboard IDE drivers working in DMA-33 mode in Win98SE.

Can you disable the L1 cache in BIOS?

Yes, you can disable the L1 cache in BIOS.

Ok I had a few minutes today and here are the basics. I set the CPU to 150-MHz and tested with and without L1 cache enabled.

The board.

Test at 150-MHz with L1 cache enabled. It's hard to see but it scored a 74.51.

Test at 150-MHz with L1 cache disabled. Scored a 16.55, is that slow enough for the really old DOS games?

That looks very promising!

Pretty much all socket 7 CPUs I tried perform like a 386 with L1 cache disabled. Some chips are fast 386s like the Pentium MMX some more average 386s like the AMD K6.

For benchmarking gaming performance it's better to use 3dbench. You can find it in the first post here:

3DBENCH CPU Benchmark Database

Use the version 1.0 not 1.0c when you disable L1 cache please.

I am hoping for a score around 10 or so maybe faster up to 15 or slower down to 8.

More questions:

When did they make these boards?

Are there other companies / brands doing similar products?

I should have some time Monday or Tuesday to run a few tests using 3dbecnh for you.

I am assuming the boards are circa 1999. There were lots of these things made but most were special purpose for embedded applications like cash registers, slot machines, etc. I have another board with soldered on CPU, built in Compact Flash slot, onboard RAM, flat panel connector and more. Problem is it has only 1 PCI and 1 ISA slot. These particular boards I stumbled on several months ago and bought several. I have had really good luck with them. None of the boards were intended for end users, as far as I can tell. From what I have seen (and I am no expert on Geode-GXM boards) the ECS P5GX-M is the one for me but I am sure there are others out there.

I don't want to over promote this particular GXM board, but I did do quite a bit of searching for GXM boards a few years back and this particular board seems to have all the bells and whistles. There weren't a whole lot out there with socketed CPU, 2 PCI, and 2 ISA. I'd buy them while they are still NIB and affordable. I ran the whole gong show of 686 benchmarks on this board without as much as a hick-up.

It took me some time to find the proper working Win98 drivers, but they all panned out. Some of the drivers in the enclosed CD-ROM are not correct for this board, but can be found online. The onboard sound and graphics are pretty lousy compared to an ISA or PCI solution.

My main complaint about the board is that it has jumper settings for 4x33 (133 MHz) operation, but lists this as "testing only". The board didn't even turn on when set to 133 MHz. Perhaps there is some trick to getting 4x working?

feipoa wrote:

.

My main complaint about the board is that it has jumper settings for 4x33 (133 MHz) operation, but lists this as "testing only". The board didn't even turn on when set to 133 MHz. Perhaps there is some trick to getting 4x working?

Maybe Cyrix/AMD did something silly with the MediaGX/Geode like Nexgen did with the nx586 and put the multiplier on the chip itself. Maybe you have to match the jumper setting to a specific CPU that runs at 4x to get it to work.

Sliderider, that is a good point, though it would require some tricky verificaiton on our part. I thought all multipliers were on the CPUs themselves though, but it could be that none of the socketed GXM chips had a 4x multiplier option. However, when this is the case, they usually fall-back to a default multiplier - but the GXM's were an odd bunch, so who knows.

The slowest socketed version of the GXM CPU was the 180, which ran at 30x6. I tested a GXM-200 (33x6) at 4x, but it didn't work either. The 33x4 GXM-133 chips were BGA only, so it is possible that the motherboard testers had some kind of BGA-to-socket7 adpater to test the 33x4 operation. When I worked at a semiconductor company, we would use fujipoly as a contact means and a pressure latch to test such grid-based devices. The maker of this motherboard could have used a similar testing aparatus, that is, if a GXM-133/4x is the only chip to support the "test mode" jumper setting.

Yeah I wanted to run it at 30-MHz or 33-MHz x 4 too. I was in a hurry so I didn't look long but mine would not post using the 4x multiplier either. I will play with every single permutation of the jumper settings just to see if undocumented settings might have an effect. Might play with bus speed too. A data sheet for the Geode GXM series might be useful too, I will try to find one.

Now if we want to get crazy... look at the upper right corner of the board. Notice that J2 (PCI Bus Speed) has three undocumented pins. Also, there is a crystal (might replace) as part of the timing circuit. What if we were able to get the bus speed down a little? I imagine at some point we might lose the PCI slots working and only have the two ISA slots left.

Just thinking out loud.

Here is the data sheet. Read page 24 and look at CLKMODE[1.0] and CLKMODE[2.0]. I am just quickly looking but if we can use CLKMODE[2.0] then there is a 4x multiplier available.

EDIT: looks like I jumped the gun on this one but there may still be useful information in the datasheet.

From my experience, replacing the crystal on boards newer than a 386 leads to timing interrupt problems with the diskette drive, amongst other things.

There might be a program to soft change the multiplier on the CPU. I'd start by looking for this. I know the Evergreen program for the Cyrix 5x86 lets you change the multipliers in software via certain registers.