bertrammatrix wrote on 2025-07-09, 15:45:

I will have to check the bios revision to be certain, but, I never updated either of these boards so it should be whatever came […]
Show full quote

I will have to check the bios revision to be certain, but, I never updated either of these boards so it should be whatever came with them originally. I don't remember an issue with 3x/4x, however I mainly used the board with Cyrix so if it had an issue with AMD I may have not noticed it. Could it be a jumper configuration issue?

Yes, CPU-Z also displayed only half of the cache I had installed.

One more thing I remember- I think the jumper directly below the multiplier jumper did something strange with cyrix CPUs, like enable a fractional multiplier of 1.5 or something. It wasn't useful to me at the time, but definitely interesting.

Is that voltage regulator you used an adjustable DC-DC "buck converter"?

I have never been able to use the 4x multiplier on AMD with this board successfully. After activation, the L2 cache did not work at all or with a different BIOS version it was many times slower. Many users do not even notice it, because if there is EDO memory and the L2 is turned off, the speed is still very good. I have tried various combinations of jumpers and without success. But with the 3x multiplier everything always worked great.

I think there could have been more BIOSes, because some report cache even though there is none on the board (because some boards were manufactured with fake cache). Only benchmarks will reveal the truth.

I have now tried the jumper under the multiplier and unfortunately it does nothing for me. A 2.5x multiplier would be great for me, because I would try 2.5x60 or 2.5x66MHz 😜

and yes, it's a regular adjustable stepdown module with an attached LCD. The regulator on the board was not stable when overclocked to 150MHz. The voltage had big drops under load. I don't have an oscilloscope, but it was easy to see even with a multimeter.

Why are you running the Cyrix at 4.15 V now, that is, instead of 4.0x volts?

Do you have a link for this pre-built switching voltage regulator with voltage display?

From my experience with socket 3 boards, the speed rating on the DRAM module is less important than luck. Sometimes you might get a 60 ns module that handles your crazy timings, while even 45 ns modules won't work.

When you wrote "Quake is an unattainable goal", is this with the 3-1-1-1 timing, or with 2-1-1-1 timings?

douglar wrote on 2025-07-09, 15:09:

Do the three empty banks cause signal reflections? If that's the case, could there be anything like a "terminating simm" that can be installed to make a cleaner signal without creating additional loads?

This is something I too was curious about. I think issues with transmission line reflections is why DRAM modules only work well in select slots (with crazy speeds) - presumably in the slots which yield milder reflections.

RayeR wrote on 2025-07-09, 16:57:

BTW I recently designed a new interposer v2 with onboard VRM with fast response DC-DC switcher RT5789AGJ8F at bottom side. I hope it could improve Vcore clarity and maybe allows to lower Vcore a bit while keeping the stability.... maybe I could go from 3,7-3,8V to 3,5-3,6V @40MHz,

This would surprise me. But even if it does nothing in this regard, its nice to have a general purpose QFP-PGA interposer with VRM. The spec sheet for the RT5789AGJ8F shows it can handle 6 A, so wouldn't the Cyrix 5x86, even if more power hungry, still work on your v2 interposer? Do you have a photo of the bottom side of v2?

BobocoCZ wrote on 2025-07-10, 08:45:

I will have to check the bios revision to be certain, but, I never updated either of these boards so it should be whatever came with them originally. I don't remember an issue with 3x/4x, however I mainly used the board with Cyrix so if it had an issue with AMD I may have not noticed it. Could it be a jumper configuration issue?

I wasn't aware of this. I will have to do more testing with the M918 and the Cyrix 5x86-120 QFP. I haven't got around to putting another 3 of these on my Chinese interposers. I'm also putting 4 more B1 Am5x86 chips on these interposers, which I am hoping will work at 200 MHz.

feipoa wrote on 2025-07-10, 09:31:

Why are you running the Cyrix at 4.15 V now, that is, instead of 4.0x volts? […]
Show full quote

Why are you running the Cyrix at 4.15 V now, that is, instead of 4.0x volts?

Do you have a link for this pre-built switching voltage regulator with voltage display?

From my experience with socket 3 boards, the speed rating on the DRAM module is less important than luck. Sometimes you might get a 60 ns module that handles your crazy timings, while even 45 ns modules won't work.

When you wrote "Quake is an unattainable goal", is this with the 3-1-1-1 timing, or with 2-1-1-1 timings?

douglar wrote on 2025-07-09, 15:09:

Do the three empty banks cause signal reflections? If that's the case, could there be anything like a "terminating simm" that can be installed to make a cleaner signal without creating additional loads?

This is something I too was curious about. I think issues with transmission line reflections is why DRAM modules only work well in select slots (with crazy speeds) - presumably in the slots which yield milder reflections.

RayeR wrote on 2025-07-09, 16:57:

BTW I recently designed a new interposer v2 with onboard VRM with fast response DC-DC switcher RT5789AGJ8F at bottom side. I hope it could improve Vcore clarity and maybe allows to lower Vcore a bit while keeping the stability.... maybe I could go from 3,7-3,8V to 3,5-3,6V @40MHz,

This would surprise me. But even if it does nothing in this regard, its nice to have a general purpose QFP-PGA interposer with VRM. The spec sheet for the RT5789AGJ8F shows it can handle 6 A, so wouldn't the Cyrix 5x86, even if more power hungry, still work on your v2 interposer? Do you have a photo of the bottom side of v2?

BobocoCZ wrote on 2025-07-10, 08:45:

I will have to check the bios revision to be certain, but, I never updated either of these boards so it should be whatever came with them originally. I don't remember an issue with 3x/4x, however I mainly used the board with Cyrix so if it had an issue with AMD I may have not noticed it. Could it be a jumper configuration issue?

I wasn't aware of this. I will have to do more testing with the M918 and the Cyrix 5x86-120 QFP. I haven't got around to putting another 3 of these on my Chinese interposers. I'm also putting 4 more B1 Am5x86 chips on these interposers, which I am hoping will work at 200 MHz.

The voltage was slightly higher because it seemed to me that after replacing the voltage regulator the system was not fully stable. But now I know that it was caused by a very worn out socket for the TAG chip 😁 After my testing I will have to replace all the cache sockets on the board. This morning, after adjusting the socket for the TAG, I set the voltage there to 4.10V and everything seems to be fine 😀

The regulator is simple from AliExpress. The XL4015 chip was recommended to me, so anything with it will do. I attached the display there separately:

https://www.aliexpress.com/item/1005006634151432.html
https://www.aliexpress.com/item/1005003119613094.html

Neither I nor my friend managed to get the 4x multiplier working on this board and we have 3 different M918 boards. I would be very happy if the fault was mine and not the board 😀 But I remind you that I am talking about the 4x multiplier for AMD am5x86. This Cyrix should not support the 4x multiplier and I do not have any other that can.

feipoa wrote on 2025-07-10, 09:31:

When you wrote "Quake is an unattainable goal", is this with the 3-1-1-1 timing, or with 2-1-1-1 timings?

I meant 2-1-1-1 0WS. I wish it would be stable, but I don't believe it 🙁

Sorry, my English is not good, I translate everything, so sometimes there may be misunderstandings. With the cache setting 3-1-1-1 0WS everything is tested and stable. The setting 2-1-1-1 0WS is not stable, but I managed to do a few tests. I also included them here for comparison. Quake has 21.5fps with the stable setting 3-1-1-1 0WS.

If we managed to have a stable 2-1-1-1 0WS, the FPS in Quake would improve to 22.4 (Read RAM "Faster" or 22.6 (Read for RAM "Fastest")

Neat, that LCD display board is extremely cheap. I assume you connect the black wire to GND, the Red wire to VRM Vcc out, and then yellow wires goes where?

For the M918's original VRM, do you recall the pinouts? I probably will just use these tiny $1 bucks I have sitting in my bin.

With 3-1-1-1, DRAM read = fastest, DRAM write = 0ws, you got 21.5 fps in DOS Quake? If you disable L2, what Quake score do you get?

feipoa wrote on 2025-07-10, 13:07:

Neat, that LCD display board is extremely cheap. I assume you connect the black wire to GND, the Red wire to VRM Vcc out, and then yellow wires goes where?

For the M918's original VRM, do you recall the pinouts? I probably will just use these tiny $1 bucks I have sitting in my bin.

With 3-1-1-1, DRAM read = fastest, DRAM write = 0ws, you got 21.5 fps in DOS Quake? If you disable L2, what Quake score do you get?

Summary, let's all get to know each other 😀

Test 1:
DRAM Read Timing = Faster
DRAM Write Timing = Fastest
SRAM Read Timing = 3-1-1-1
SRAM Write Timing = 0 WS
L2 cache = Disabled
Quake Bench 320x240: 19,2fps (Stable!)

Test 2:
DRAM Read Timing = Faster
DRAM Write Timing = Fastest
SRAM Read Timing = 3-1-1-1
SRAM Write Timing = 0 WS
L2 cache = Enabled
Quake Bench 320x240: 21,5fps (Stable!)

Test 3:
DRAM Read Timing = Fastest
DRAM Write Timing = Fastest
SRAM Read Timing = 3-1-1-1
SRAM Write Timing = 0 WS
L2 cache = Enabled
Quake Bench 320x240: 21,7fps (Unstable!)

Test 4:
DRAM Read Timing = Faster
DRAM Write Timing = Fastest
SRAM Read Timing = 2-1-1-1
SRAM Write Timing = 0 WS
L2 cache = Enabled
Quake Bench 320x240: 22,4fps (Unstable!)

Test 5:
DRAM Read Timing = Fastest
DRAM Write Timing = Fastest
SRAM Read Timing = 2-1-1-1
SRAM Write Timing = 0 WS
L2 cache = Enabled
Quake Bench 320x240: 22,6fps (Unstable!)

and LCD - there is a trimmer, it can be easily calibrated with a precise multimeter.

black = GND
red = 5V
yellow = Vcore

But the display works from about 3.5V, yellow and red could both be on VCore.

feipoa wrote on 2025-07-10, 09:31:

This would surprise me. But even if it does nothing in this regard, its nice to have a general purpose QFP-PGA interposer with VRM. The spec sheet for the RT5789AGJ8F shows it can handle 6 A, so wouldn't the Cyrix 5x86, even if more power hungry, still work on your v2 interposer? Do you have a photo of the bottom side of v2?

The PCB is just manufactured in JLCPCB so we have to wait some time for real tests 😀 I have only Am5x86, not the Cyrix.
Yes, the switcher IC is quite current capable, so it should easily handle also Cyrix load. It depends on quality of used inductor. To keep it simple I use standard 2-resistor 0603 voltage divider (with optional capacitor for f-comp.) in feedback so let user to calc. and select output voltage as needed...

Thanks, the confusion for me was that we don't normally have an SRAM write wait-state option on socket 3 boards. As such, I thought you writing of '0WS' was referring to the DRAM. So the SRAM, even at 3-1-1-1, made a decent impact on Quake performance.

RayeR, those look neatly placed. Hopefully it works out.