I'm not sure how you would get around the CD-ROM check to run those late era DOS games on the system. I guess you aren't running any late DOS games.

Thanks for the tabulated results. Big bummer that the GLQuake results aren't present as those were the primary results I was looking for. Is there any benefit to having DX9 installed on this system?

The Radeon 9250 showed poor game playback results compared to V3 and GF series. I remember when I was setting up my K6 system, I compared GF2, GF4, and FX5500. The GF4 showed slightly better framerates. I guess you'd need a Voodoo5 to match the GeForce FX on the MediaGX system?

feipoa wrote on 2025-02-13, 07:54:

Big bummer that the GLQuake results aren't present as those were the primary results I was looking for. Is there any benefit to having DX9 installed on this system?

Of course they are GLQuake results in the table (sadly not for the V3-2000 oc-ed to 3000)....
I don't think, the system benefits from Dx9 - I installed it just because the GF FX5200 can do it (as this will be the final VGA for the system).

About DOS-games: some time ago p_shipkov asked me the same question. To be honest - I never was a big gamer, so no one game is for me important enough to optimize a system for playing it faster. I am focusing more at the overall-performance or the possibilities offered by the hardware (not software).

BTW: yesterday I was able to overclock the GX1-CPU very high! Today there will be more info about this (some finetuning is still needed) 😉

gonzo wrote on 2025-02-13, 09:18:

Of course they are GLQuake results in the table (sadly not for the V3-2000 oc-ed to 3000)....

Of course I am referring to the GLQuake results from the Voodoo3 to compare with the Geforce results.

gonzo wrote on 2025-02-13, 09:18:

BTW: yesterday I was able to overclock the GX1-CPU very high! Today there will be more info about this (some finetuning is still needed) ;-)

Got Turbo working 'eh?

Very high? Sounds like someone wired pclk into the correct bclk

Well, I was able to get 37,5 MHz FSB (according to the datasheet for the turbomode), so the CPU runs now at 375 MHz (from 266 printed on it, so this is an overclock of 41 %!).
I start to think, this 266-MHz-CPU is maybe really a re-labeled 300- or 333-MHz-CPU.

Please be patient for some more hours, I am finishing the finetuning today, so I really hope to post the final results and some explanations soon (today or tomorow).

BTW, what is your own MediaGX-system (@feipoa and others)?

Ah I thought you might have got 66fsb

Well, let's have a MediaGX1-CPU running STABLE at 375 MHz!

1. Mainboard

To enable the turbomode of the clockgenerator PLL52C59-14T, the resisitor connected to pin 5 (it's a 4,7k one) has to be let connected to it, but the other end of the resistor must be set to ground (to a Vss-pin of the generator; I choosed pin 11). Simply removing this resisitor lets the turbomode disabled.
Now, a turbomode of 75 MHz FSB (= settings for 55 MHz FSB in non-turbo) is possible - so the datasheet of the clockgenerator has to be corrected (as the turbomode there is listed in the text for an FSB of 50, 60 and 66 MHz only).

Regardless of the FSB and regardless of the used PCI-cards, my board seems to have problems, if the upper PCI-slot (the next one to the CPU-socket) is used - this causes instability in WinME. So I have to let it empty. Maybe it's shared with some other hardware onboard - I just don't know.

The BIOS shows the CPU-frequency as 366 MHz (of course not true).

2. RAM-Frequency

It seems to be so, that, compared to the GXm-chip (it cames on the market in 1999), the GX1-chip (it cames on the market in 2000) has a better RAM-controller. With the GXm-chip, 120 MHz of RAM-frequency was not possible for my system. Under equal conditions (incl. RAM-sticks), the GX1-chip can do now STABLE 125 MHz at 375 MHz CPU-frequency (on the other hand, 133 MHz RAM-frequency at 333 MHz CPU-frequency is not possible - maybe 133 MHz are just too much for the GX1-RAM-controller).

3. VGA

Next to the GeForce FX 5200 (driver version 44.04), I tested a Geforce 6200 PCI (512 MB RAM, 64 bit; driver versions 77.72, 81.85 and 81.98, with no differences in the 3D-score).

In DOS, the 6200 outperforms the FX 5200 at all (GX1@375 MHz/ RAM@125 MHz):

- DOOM: 91,5 FPS (6200) / 72,6 FPS (FX 5200)
- Quake: 43,2 FPS (6200) / 41,2 FPS (FX 5200)
- PCPBench /vgamode: 60,5 (6200) / 56,3 (FX 5200)
- 3dBench 1.0c: 156,9 (6200) / 121,9 (FX 5200)

In Windows, maybe because of the driver, the FX 5200 outperforms the 6200 in some benchmarks (GX1@375 MHz/ RAM@125 MHz):

- 3DMark 99MAX: 860 3D marks (FX 5200) / 870 3D marks (6200)
- 3DMark 2000: 823 3D marks (FX 5200) /805 3D marks (6200)
- 3DMark 2001SE (still crashes in Car Chase and Lobby at high detail, so the 3D mark results are in fact NOT correct!): the FX 5200 is faster than the 6200
- Quake II: 29,0 FPS (FX 5200) / 32,8 FPS (6200)
- GLQuake: 55,1 FPS (GF 5200) / 47,9 (6200)

To not overload this thread, I uploaded not all pictures for some benchmarks.

and the last pics

Ah you had to deliberately pull it down. I was on the right track then.

The next trick will be changing the crystal. 16,18,and 20khz might be contenders

OK, so you were able to operate your system at 5x75 MHz = 375 MHz? But wouldn't your PCI slots also be running at 75 MHz? How fast are your ISA slots running?

If the MediaGX is capable of running up to 75 MHz FSB, why would they dumb it down to run at only 33 MHz? Something doesn't feel right here.

The motherboard I use is an Elitegroup ECS P5GX-M. It is one of the more common MediaGXm boards. It contains the Cyrix CX5530 companion chipset. This board also works with the GX1 and I am using a ceramic GX1-300, 2.0V. I have voltage adjustment ranges from 2.1 V to 3.5 V, in 0.1 V increments. Multiplier options are 5x thru 10x.

The PLL on my board isn't quite as versitle as yours. The PLL on my board was make specifically for the Geode line of CPUs. It's the ICS MK1491-06 https://mm.digikey.com/Volume0/opasdata/d2200 … 491-06_RevJ.pdf

The clock options on this PLL are 25, 30, 33.3, and 37.5 MHz. While my board manual only lists 30 and 33.3, 25 and 37.5 MHz should be available from the existing jumper header.

I haven't been able to run the system at 333 MHz with my ceramic GX1-300 chip. Perhaps it doesn't like the extra 0.1 V? I'm running it at 2.1 V instead of the rated 2.0 V. Could I get some speed boost by running the system at 8x37.5 rather than 9x33.3 ? I assume the only boost would be by way of PCI graphics speed.

myne wrote on 2025-02-14, 02:23:

The next trick will be changing the crystal. 16,18,and 20khz might be contenders

I doubt this chip itself can do much more than 375 MHz (even more so, as it is not possible to increase the Vcore-voltage on this board).

But, is there somewhere maybe any other information about overclocking of a MediaGX-CPU, how far can it goes?

As the RAM-frequency depends directly from the CPU-frequency AND 133 MHz are not possible (at least with this particular CPU), there will be no more significant benefit for the RAM, I'm afraid.

So the only once benefit will be for the CPU to work at a higher FSB ( but maybe at a lower ratio, maybe 9 or 8 ).

feipoa wrote on 2025-02-14, 08:09:

I am using a ceramic GX1-300, 2.0V. ...
......
I assume the only boost would be by way of PCI graphics speed.

You should try a CPU in plastic (not ceramic) manufacturing. I have some more ceramic variations here (200 to 266 MHz) at voltages from 2,2 to 2,9 V - only one of them was possible to run at 300 MHz (the ceramic NSC-266 shown in the pictures few days ago).

You are maybe right about the VGA-speed.

Isn't the ceramic GX1-300-2.0 produced on a smaller process compared to the ceramic 266 CPUs you are using? My GX1-300-2.0 should be the same process as your plastic GX1. I ran some quick tests at 9x37.5. DOS was OK, but Windows spat out an error on boot.

9x37.5/3 = 112.5 MHz. I noticed that I'm using PC100 CL2 chips. Maybe that's why it crashed. Perhaps I need PC133 here?

It feels like I ran all these tests 8 years ago and sort of recall PC133 did not help in achieving 333 MHz. I will need to test this again. No time now. However, I suspect your board assembly is better for running these higher SDRAM frequencies.

feipoa wrote on 2025-02-14, 08:55:

Isn't the ceramic GX1-300-2.0 produced on a smaller process compared to the ceramic 266 CPUs you are using? My GX1-300-2.0 should be the same process as your plastic GX1.

You are right. My (first tested) ceramic CPU is a GXm (0,35 µm); all GX1 are made in 0,18 µm.
The only difference is the Vcore: my GX1-CPU needs 1,8 V only (your GX1 needs 2,0 V) - exactly this was for me the main reason to try/use it

feipoa wrote on 2025-02-14, 08:55:

I'm using PC100 CL2 chips. Maybe that's why it crashed...However, I suspect your board assembly is better for running these higher SDRAM frequencies.

My SD-RAMs are both 128 MB each (2 identical ones), PC133, CL2 (7 ns), double-sided.
Like you, I do not think, the mainboard itself affects the RAM-frequency, as the RAM-controller is build-in into the CPU.

Datasheet mention anything about spd or sdram clocks?

SDRAM requires 4 clocks per stick.
Generally that means there will be another clockgen somewhere.

I think your BGA GX1-266 is 1.8 V because of the operating frequency. You'll see that if you BGA GX1-300 is 2.0 V, like my ceramic.

Next update!

Looking forward how to optimize the GX1, I found this thread (many thanks to feipoa for it!): Register settings for various CPUs

So, the GX1 can be optimized by using of the software "CTCHIP34".
This software increases the FPU-performance itself, as well the RAM-throughput to the FPU (as shown by SiSoft Sandra 99 and 2004 SP1).

WINDOWS ME

As I am using WinME without pre-installed DOS, AND as the Autoexec.bat of WinME is write-protected, a small batch-file is needed:

cmd.exe
C:\CTCHIP34 cx586 /20h:=%%xxxxx1x1
C:\CTCHIP34 cx586 /0E8h:=%%xx1xxxxx

Increasements (GX1 @ 375 MHz/RAM @ 125 MHz / GeForce FX 5200):

3DMark 99 MAX: 893 3DMarks (+ 3,8 %), 1381 CPU 3DMarks (+ 16,8 %)
3DMark 2000: 826 3DMarks (no increase), 59 CPU 3DMarks (no increase)
GLQuake: 56,5 FPS (+ 2,5 %)
Quake II: 29,8 FPS (+ 2,8 %)
CPU-Z, FPU-benchmark: 1412 (+ 7,1 %)

DOS

In DOS (booted from a floppy disk of Win98SE), on the prompt-line:

C:\CTCHIP34 cx586 /20h:=%xxxxx1x1
C:\CTCHIP34 cx586 /0E8h:=%xx1xxxxx

Speedsys 4.78: no increase at all points
DOOM: 72,6 FPS (no increase)
Quake: 41,3 FPS (no increase)
PCP Bench /vgamode: 56,3 (no increase)
3dBench 1.0c: 121,9 (no increase)

I'm wondering, why there is no increase at least for Quake, as this game is very FPU-sensitive...Is it maybe because of missing of a "true" pre-installed DOS?

feipoa wrote on 2025-02-14, 17:07:

I think your BGA GX1-266 is 1.8 V because of the operating frequency. You'll see that if you BGA GX1-300 is 2.0 V, like my ceramic.

Yes, you are right! Found today this link: https://www.cpushack.com/MediaGX.html

A GX1 @ 300 operates at 2,0 V; a GX1 @ 333 operates at 2,2 V. So I am really happy to have one CPU doing 375 MHz at 1,8 V 😀

What is the tolerance range for the Vcore? Could you bump it up to get an ever higher clock, or would that be too risky for a processor that for this level of overclocking is likely a golden chip, so to speak.

Did you actually measure to see if it is running at 1.8 V? Can you confirm this? Maybe it is being run ta 2.2 V.

I'd be weary about running a 266 Mhz chip at 375 MHz for too long. From my experience testing GXm and GX1 chips, I haven't had any overclocking room. I have a rare GXm-300 with Cyrix logo, I forget the voltage, but I couldn't get that chip running at 300 MHz.

I suspect there's either motherboard layout considerations, or binning of the companion chip are needed. Basically, I don't know if you are lucky with your CPU, lucky with your motherboard, or lucky with your CX5530 companion chip.