Can you post a picture with the table bottom right of SpeedSys 4.78? (part of DOSBENCH suite)
(Or 2 pictures, one with turbo on and the other with turbo off)

2) basically correct, but depends on amount of DRAM and SRAM (cache)

Screenshots of BIOS settings and SpeedSys 4.78 with Turbo button ON and Turbo button OFF.

Sorry, I miss the table bottom right in your SpeedSys charts.
It is generated right after memory test.

Dido wrote on 2023-02-07, 08:01:

Screenshots of BIOS settings and SpeedSys 4.78 with Turbo button ON and Turbo button OFF.

Please re-run Speedsys without EMM386 (or a similar memory manager) loaded, and with at least some megabytes available extended memory (i.e. don't let smartdrv or another disk cache use all your RAM). It should display memory timing graphs and read/write/move data rates in that case. This diagram can point to issues with the cache configuration.

Hello!
Sorry, the screenshots I uploaded yesterday are with QEMM and SMARTDRV.
Today I booted the computer with bypassing AUTOEXEC.BAT and CONFIG.SYS.

Here are the new screenshots:

I have the same motherboard, as you can notice the cpu speed doesn't decrease but it greatly decrease the memory speed, thus making the system slower anyway.
You can notice it even from bios screen if you use turbo mode memory count and the beep will be slower.

My experience in the past is that 486 CPU's with a L1 write back cache scheme don't slow down as much compared to CPU's with L1 write thru cache scheme.

I'm not quite sure why that happens but, just something I also observed.

jesolo wrote on 2023-02-08, 11:11:

My experience in the past is that 486 CPU's with a L1 write back cache scheme don't slow down as much compared to CPU's with L1 write thru cache scheme.

Oh, that's easy.
Since deturbo does not affect what's inside the CPU, and most contemporary benchmarks run completely in an area that fits into L1-cache, they are not affected by reducing memory speed.
If benchmarks write to RAM too, you may notice a difference between L1 WB and L1 WT too.

Disruptor wrote on 2023-02-08, 14:15:

Oh, that's easy. Since deturbo does not affect what's inside the CPU, and most contemporary benchmarks run completely in an area […]
Show full quote

jesolo wrote on 2023-02-08, 11:11:

My experience in the past is that 486 CPU's with a L1 write back cache scheme don't slow down as much compared to CPU's with L1 write thru cache scheme.

Oh, that's easy.
Since deturbo does not affect what's inside the CPU, and most contemporary benchmarks run completely in an area that fits into L1-cache, they are not affected by reducing memory speed.
If benchmarks write to RAM too, you may notice a difference between L1 WB and L1 WT too.

I will have to run some benchmarks again. I think a Doom benchmark might be a good measurement in terms of performance between L1 write thru vs L1 write back with turbo on and off.

I did the benchmark with doom and amd 5x86 133, with turbo mode the performance was around a 486 25 mhz

I did some deturbo benchmarks with Asus VL/I-486SV2GX4 last summer. Here are the results.

De-turbo with that SiS chipset works by holding the CPU 4µs or 8µs after every 12µs. So deturbo slows CPU to 2/3 or 1/3 speed (default). So of course it will affect everything done by the CPU, not just memory speed. Why the speedsys result is like that? I don't know, maybe the CPU doesn't properly halt when instructed?

Edit: Here are the utils to change de-turbo from 1/3 to 2/3. Just don't use the L2 disable stuff. They won't work properly. And it turned out there is no real need to disable L2 at all.

Wow what a complete testing you did, congratulations!
I will compare those results with mine, I have the motherboard in a case so I won't be able to switch cpus though...
If I remember right, 5x86 cpus supported both cache settings, I could try to switch the value in the bios.

aitotat wrote on 2023-02-08, 16:04:

De-turbo with that SiS chipset works by holding the CPU 4µs or 8µs after every 12µs. So deturbo slows CPU to 2/3 or 1/3 speed (default).

How do you change deturbo from 1/3 to 2/3? Is it a setting in the BIOS?

florian3 wrote on 2023-02-08, 16:13:

How do you change deturbo from 1/3 to 2/3? Is it a setting in the BIOS?

It could and should have been an option in the BIOS but I don't know why no manufacturer provided that setting. So I made my own utils to change the deturbo mode (L2 enable/disable is broken. Do not use those). These are only for SiS 471 chipset and won't work on any other chipsets.

aitotat wrote on 2023-02-08, 16:04:

Why the speedsys result is like that? I don't know, maybe the CPU doesn't properly halt when instructed?

Perhaps it uses CLI / STI to interrupt interrupt requests.

aitotat wrote on 2023-02-08, 16:22:

It could and should have been an option in the BIOS but I don't know why no manufacturer provided that setting. So I made my own utils to change the deturbo mode (L2 enable/disable is broken. Do not use those). These are only for SiS 471 chipset and won't work on any other chipsets.

Thank you, works on my Chaintech board with SiS 471.

Dido wrote on 2023-02-08, 07:22:

Hello! Sorry, the screenshots I uploaded yesterday are with QEMM and SMARTDRV. Today I booted the computer with bypassing AUTOEX […]
Show full quote

Hello!
Sorry, the screenshots I uploaded yesterday are with QEMM and SMARTDRV.
Today I booted the computer with bypassing AUTOEXEC.BAT and CONFIG.SYS.

Here are the new screenshots:

Hi,
some off-topic comments, as you asked only about turbo on/off.
1)
The figures show, that you use a slow configuration.
on 33MHz normally you always want use cache read 2-1-1-1, but auto-config configures 2-2-2-2. I am sure the faster will work, too.
Some RAM parameters are not set to "fastest" too. With 33MHz normally this always works, also when you have "only" 70ns rated simm modules.
Also DRAM slow refresh On and DRAM write burst can be turned on (settings to make it even more fast 😁)
2)
I think L1 is configured to Write Through.
Did you configure the jumpers for L1 Write back (P24D setting)? Please note that bios setting for L1 WT/WB is not relevant, it does not matter what you set there.
I am not sure about point 2) though, maybe the low values for L1 cache throughput are caused by point 1, too.

CoffeeOne wrote on 2023-02-08, 19:08:

The figures show, that you use a slow configuration. on 33MHz normally you always want use cache read 2-1-1-1, but auto-config c […]
Show full quote

The figures show, that you use a slow configuration.
on 33MHz normally you always want use cache read 2-1-1-1, but auto-config configures 2-2-2-2. I am sure the faster will work, too.
Some RAM parameters are not set to "fastest" too. With 33MHz normally this always works, also when you have "only" 70ns rated simm modules.
Also DRAM slow refresh On and DRAM write burst can be turned on (settings to make it even more fast 😁)

I can test and post results with the settings you suggested, but I don't know how to set the cache in 2-1-1-1 mode.
In BIOS "Cache Burst Read" and "Cache Write Cycle" are with value "2T" and can't be changed.

CoffeeOne wrote on 2023-02-08, 19:08:

Did you configure the jumpers for L1 Write back (P24D setting)?

Yes, jumpers are configured for P24D Write-back from user manual.

Disruptor wrote on 2023-02-08, 17:10:

aitotat wrote on 2023-02-08, 16:04:

Why the speedsys result is like that? I don't know, maybe the CPU doesn't properly halt when instructed?

Perhaps it uses CLI / STI to interrupt interrupt requests.

What does this mean? CPU problem or chipset incompatibility?