Mostly just experimental findings, although at one point I came across a document, either a chipset databook or some BIOS writers guide, which contained a table showing the AUTO settings for double-banked cache vs. single-banked and the doubled-banked mode used faster SRAM timings.

pshipkov wrote on 2021-03-26, 00:01:

@Feipoa Didn't look at this ALI chipset specification, but you are probably right. I remember you talking about single/double ba […]
Show full quote

@Feipoa
Didn't look at this ALI chipset specification, but you are probably right.
I remember you talking about single/double banked caches. Do you have more detailed info somewhere about that ? Findings, etc.

@H3nrik V!
Look here

Cool! Loving it!

Was curious how Intel's RapidCad CPU stacks against standard 386DX-40 and a "high-end" upgrade such as TI's 486SXL2-50.

Used 3 high-quality motherboards to cross-check results, but the final tests were performed on this very fast 386 system based on 486 Symphony Haydn chipset, but in classic 386 implementation - ISA, 30-pin FPM RAM, 132-pin PGA socket.
For the record - the other two boards were based on late OPTi and UMC chipsets.

RapidCad consists of two chips - the real 132 pin processor and a placeholder FPU one.
Interestingly enough the FPU chip was not needed in any of the motherboards.
In fact the systems either hang, or do not boot at all when it is inserted in the FPU socket.

The processor is 33MHz rated, but works really well at up to 45MHz. Gets unstable beyond that.
It is one of the coldest CPUs i have seen. Even under 100% load for more than 2 hours it barely gets warm.

There is not much else to add - things work as advertised.

Here is the suspect:

Speedsys screengrapbs.
On most systems SXL2 CPU garbles speedsys'es screen, so not included here.
The above screenshot is for 386DX, the one below is for RapidCad.
For some reason 386dx shows up as running at 40MHz, but is 45 really.

And the usual set of benchmarks.
The 3 CPUs clocked at 45MHz. Where both the 386DX and SXL2 are accompanied by Cyrix FasMath (black top) FPU.
Goal was to see the relative performance difference between them.
Gray bars indicate peak perf using SXL2 CPU with the same FasMath (black top) FPU clocked at 50MHz.
I could use gray top FasMath and improve FPU perf by 1-4%, but forgot to do that. This will affect only the rendering tests and P3B to a less degree.

RapidCad is much faster than standard 386DX CPUs.
Applications that rely on floating point math further benefit from it, but in general SXL2 + FasMath FPU is the better deal.
3D Studio loves it, while ALU speed affects LightWave3D as much as FPU does.
WinTune2 did not like it at all. Crashes hard at startup every time.

Too bad Intel disabled its L1 cache, otherwise the story could be different.

I remember seeing someone I think CPU-World pair the RapidCAD with a Weitek, which is pretty neat. This way you get Intel's FPU and a specialised math co-processor.

I also wondered if weitek can be combined with rapidcad.
It will be very interesting how different Pro apps will react to it.
Will they prefer 3167 or fall back on 487.

pshipkov wrote on 2019-02-18, 10:27:

[…]
Show full quote

• motherboard: VLSI VL82CPCAT-16QC
  cpu: Intersil CS80C286-25 25/30MHz
  fpu: IIT 2C87-20 22MHz
  ram: 4Mb 60ns parity FPM
  vga: Diamond SpeedSTAR 24 1Mb (Tseng Labs ET4000AX, 24-bit DAC) (primary)
  • Diamond Flower VG-1000 rev 2 512Kb (Cirrus Logic CL-GD510/520)
    Trident TVGA8800CS rev 2.0 512Kb
    Western Digital WD90C00-JK 512Kb
    OTI-037 512Kb
    Ahead V5000-50PC-B 1Mb
    Headland GC208-PC 1Mb
  i/o: UMC 82C862F
  audio: Sound Blaster 1.5 CT1320C with C/MS
  lan: Tiara Lancard/E-PC 16
  psu: 300W AT
  input: BTC 5121, MS Mouse
  os: DOS 4.01 + Windows 3.0
  partitions: 1x504Mb

Hi, I've got a similar board (mine has SIPP) but I'd really like to know the jumper settings/definitions.
Do you happen to know what model motherboard this is?

This is what i know (and don't know) about the jumpers. Hope it helps.

Apparently these boards were available in Australia.
Remember at least 4-5 references related to your region.

pshipkov wrote on 2021-04-01, 03:17:

On most systems SXL2 CPU garbles speedsys'es screen, so not included here.

This is an indication that your CPU is not flushing it's internal cache properly.

Anonymous Coward wrote on 2021-04-02, 00:13:

pshipkov wrote on 2021-04-01, 03:17:

On most systems SXL2 CPU garbles speedsys'es screen, so not included here.

This is an indication that your CPU is not flushing it's internal cache properly.

Could be, or some odd Cyrix-like hardware compatibility issue, but that was usually with CPUID. I suggest he take a look at this thread on register settings, Register settings for various CPUs

I know which speedsys graphics corruption pshipkov is talking about, I've seen it many times during my own tests with 486SXL. The problem only appeared when the registers were not set correctly. It wasn't enough to just enable the barb function. You have to exclude caching in the 640-1024k region. Maybe it's because adapter ROMs sit there. That might explain the corrupted graphics.

Ahh, yeah, I always disable caching in that region. Also, sometimes you need to disable caching of all 1 MB boundaries, sometimes not. This issue rears itself when trying to boot into Windows and you get a hang-up.

Also, always be sure to keep an entire copy of your HDD incase of corruption, esp. when using these SXL's with L1 enabled. It is much easier to copy a few files over rather than setup the OS again. If you run Scandisk, make note of which files are corrupt and copy the fresh ones over. Using SXL w/L1 w/SCSI DMA can be a challenge.

4xtx wrote on 2021-04-01, 12:17:

pshipkov wrote on 2019-02-18, 10:27:

[…]
Show full quote

• motherboard: VLSI VL82CPCAT-16QC
  cpu: Intersil CS80C286-25 25/30MHz
  fpu: IIT 2C87-20 22MHz
  ram: 4Mb 60ns parity FPM
  vga: Diamond SpeedSTAR 24 1Mb (Tseng Labs ET4000AX, 24-bit DAC) (primary)
  • Diamond Flower VG-1000 rev 2 512Kb (Cirrus Logic CL-GD510/520)
    Trident TVGA8800CS rev 2.0 512Kb
    Western Digital WD90C00-JK 512Kb
    OTI-037 512Kb
    Ahead V5000-50PC-B 1Mb
    Headland GC208-PC 1Mb
  i/o: UMC 82C862F
  audio: Sound Blaster 1.5 CT1320C with C/MS
  lan: Tiara Lancard/E-PC 16
  psu: 300W AT
  input: BTC 5121, MS Mouse
  os: DOS 4.01 + Windows 3.0
  partitions: 1x504Mb

Hi, I've got a similar board (mine has SIPP) but I'd really like to know the jumper settings/definitions.
Do you happen to know what model motherboard this is?

Chop off that heinous nicad bomb asap bro. Those boards are too nice to risk

maxtherabbit wrote on 2021-04-02, 02:09:

4xtx wrote on 2021-04-01, 12:17:

pshipkov wrote on 2019-02-18, 10:27:

[…]
Show full quote

• motherboard: VLSI VL82CPCAT-16QC
  cpu: Intersil CS80C286-25 25/30MHz
  fpu: IIT 2C87-20 22MHz
  ram: 4Mb 60ns parity FPM
  vga: Diamond SpeedSTAR 24 1Mb (Tseng Labs ET4000AX, 24-bit DAC) (primary)
  • Diamond Flower VG-1000 rev 2 512Kb (Cirrus Logic CL-GD510/520)
    Trident TVGA8800CS rev 2.0 512Kb
    Western Digital WD90C00-JK 512Kb
    OTI-037 512Kb
    Ahead V5000-50PC-B 1Mb
    Headland GC208-PC 1Mb
  i/o: UMC 82C862F
  audio: Sound Blaster 1.5 CT1320C with C/MS
  lan: Tiara Lancard/E-PC 16
  psu: 300W AT
  input: BTC 5121, MS Mouse
  os: DOS 4.01 + Windows 3.0
  partitions: 1x504Mb

Hi, I've got a similar board (mine has SIPP) but I'd really like to know the jumper settings/definitions.
Do you happen to know what model motherboard this is?

Chop off that heinous nicad bomb asap bro. Those boards are too nice to risk

Yep coming off soon 😀

@Anonymous Coward
Ok, so i am not the only one seeing this. Good to know.

@feipoa
The CPU Register Settings Bible is a well known source of information.

The problem is related to this specific motherboard/cpu, regardless of control registers (tried all the combination and other things).
Cannot remember seeing this on any of the other 386 mobos i tried so far, or if there was a problem, it could be resolved with some combination of CPU registers.

pshipkov wrote on 2021-04-02, 06:09:

The problem is related to this specific motherboard/cpu, regardless of control registers (tried all the combination and other things).
Cannot remember seeing this on any of the other 386 mobos i tried so far, or if there was a problem, it could be resolved with some combination of CPU registers.

If disabling L1/L2 cache entirely (and shadow), does the problem still occur?

I am pretty sure i tried that. At some point everything was turned off with most increased wait states, etc.
The precise answer is that i can get SpeedSys to display with readable text, but it hangs when tries to identify CPU - shows "Cyrix" and locks.
The board is very stable. Everything else i threw at it works fine.
So apparently hitting a specific issue with SXL2/SpeedSys/this_board.

I've run into that a few times on different systems. I think on some I had to disable CPUID and on another I had to disabled System BIOS Cacheable or Shadow or something. I think there was also some speedsys flag which helped the situation in some rare case.

This board has no option in the BIOS for "shadow BIOS" and CPUID.
Tried again couple of SpeedSys flags. Same result - SS pauses right after displaying "Cyrix". Feels like it cannot detect the CPU speed.
It does not bother me. I don't pay much attention to these synthetic benchmarks anyway. 😀

I agree, I don't put much weight in the Speedsys results, especially on a 386 because the results are so varied with and without an FPU. In general, when this error occurs w/Cyrix chips and Speedsys, there is sometimes some cute trick to get it to run, and sometimes there isn't - and most of the time there isn't.

Agreed.

---------------

Last night i moved some of the junk around, to compress the space it occupies.
Saw this Texas Instruments 486DX2-80 CPU and two lingering mechanical hard drives.
Decided to give them a spin in that 486DX2-66 PC in a cardboard case from several posts ago, with the intention to eventually replace the current HDD and CPU in it.

Here is what i found:

Basically TI 486DX-80 is clock-to-clock faster than AMD 486DX-66 (clicked at 80 MHz).
Depends on which test you ask the variation is between 1-to-4%. For example Wolf3D and Doom get 4 and 2 extra frames per second, respectively.
Visible and in the SpeedSys above.
Decided to keep the TI CPU in the case.

I am not very interested in old mechanical HDDs, but since i was at it - gave these 3 a chance to present themselves.
Always thought that Seagate Medalist were ... well ... the medalists, buts that Caviar 2850 is just the better place. They may be a different generation, etc. Didn't check release dates. Not that curious.
Tested with the Promise controller in the PC.
Was able to use EIDE speed 7 (the maximum one the controller provides) for Caviar 2850 and Seagate Medalist.
But Caviar 2340 could go only to speed 5.
Replaced 2340 which was in the system until this point with 2850. Its seek and access times improved to a point where you feel it while working with the PC, it is not just a paper tiger.

---------------

Something weird.
I sharpened this Asus VLI based PC for top DOS graphics performance.
Recently it got some POD100 love as explained here.
At some point i removed the SB16 and LAN card from it to test something.
The system became unstable - not able to pass Quake1 test, sometimes hung in Doom, and other bad stuff.
Started reducing RAM, moving cache chips around, trying different video cards, tweaking BIOS settings.
No and no. I was like what the heck is going on - this one was always very stable. Did i damage it ?
At some point i put back the SB16. Instant stability.
What ?!
Adding more hardware usually means more volatile system - more things to go wrong.
Apparently not in this case. 😀
Whatever.