haha. Not sure how that got past their PR. I'm probably going to need some of that soon. or maybe just forget about it, shave my head, and become forest monk.

That seems to be a veeeery potent chipset 🤣

That's quite funny. Could this chipset come out before the pills?

Viagra was approved by FDA on March 27, 1998.
VIA GRAphics (MVP4) was announced in Aug, 1998.

So either VIA's marketing team was not aware of what's coming down the pipe out there, or the chipset simply delivers that extra boost.
Any proud owners who can confirm ?

These dates really make the situation even funnier! I am kind of surprised that Viatris never launched a trademark suit and shut this down! After all, Dolby sued PC Chips for their HX Pro chipset, which was just their renaming of the ALi Aladdin III.

As far as I am aware, this is not exactly VIA's naming scheme, but rather PC Chips's usual renaming mess. It's just the MVP4 chipset underneath, so it's kind of meh, with an integrated Trident Blade VGA. MVP3 is a much more robust chipset and has an AGP slot, although I think there's one or two weird MVP4 boards with AGP slots. I imagine those would be better.

The best Super 7 chipset with integrated graphics has to be the Aladdin 7 ArtX, which uses dual-channel SDRAM and has a few other tricks up its sleeve. Boards are very rare and kind of clunky, however.

So how Aladdin 7 ArtX is considered best, but mobos based on it are not good ?
Is it a paper tiger, or there are rare assemblies that are/may_be actually good ?

---

ABit AB-FS3 based on SiS Rabbit - 85C310/320/330

On multiple occasions Feipoa spoke highly of his Chaintech 340SC motherboard in combination with IBM BL3 and TI 486SXL2 processors, so i wanted to give that stuff a try.
I actually checked long time ago another board based on the same chipset - Asus 386/33-64k, but it was limited to 64Kb level 2 cache, 8Mb RAM and rudimentary BIOS. Also, at that time AMD 386DX-40 was the best processor i had.
The Asus board was not that bad really, but nothing to brag about.
A better assembly was needed.

Battery leak destroyed several traces in the upper left corner.
Previous owner did some repairs, but also caused additional damage.
Improved on that.

Two clock oscillator sockets. One for FSB another for async Intel 387 FPU. Not applicable for my use case - Cyrix FasMath, ULSI DX/DLC at max frequency.
No lights past 50MHz FSB (100MHz oscillator).

BIOS has few options only.
Used Chaintech 340SC BIOS from Feipoa.
Used STB Nitro video card based on Cirrus Logic GD-5434, 2Mb.

--- IBM BL3 at 90MHz (2x45), ISA bus at 12.5 MHz

System cannot be fully stable at 50MHz FSB.
Stepped down to 45MHz for complete stability.

All BIOS settings on max, except:
BUS CLOCK SELECT = SCLK/8 (best is /3)
EXTERNAL CACHE = DISABLED

Encountered 1-2 sporadic disk write access issues, which hint at potential problem, but inconclusive.
Most of the time finding the right IDE controller solves this. So, i am willing to conclude complete stability.

Despite level 2 cache is disabled, performance is still great.
SpeedSys confirms pretty good system memory access times, which explains it.

ULSI DX/DLC is faster than Cyrix in Quake 1.
FasMath hangs in the LightWave3D test, used the ULSI FPU.
FasMath used for PC Player Benchmark and 3D Studio R3.

Number 2 in PC Player benchmark, that is at 90MHz.
Sharing the first spot with DTK PEM-4036Y in Quake 1. <- !!!
Number 3 in the complex offline graphics tests.
On the slow side in the rest of the tests.

--- TI 486SXL2 at 80MHz (2x40), ISA at 13.33MHz

I am still fresh with the SXL2 adapter stuff.
My only SXL2-66 CPU does not scale past 80MHz.
System is not stable - it is the CPU itself.

SpeedSys hangs, which is often the case with SXL2 chips.

Few numbers for reference:
Wolf3D: 42 fps
Superscape: 33.2 fps
PC Player: 10.5 fps
Doom: 16.4 fps
The rest of the tests don't complete.

--- TI 486SXL2 at 50MHz, ISA at 12.5MHz

All BIOS settings on max, except:
BUS CLOCK SELECT = SCLK/6 (best is /3).
DRAM WRITE CAS WAIT STATE = 1 W/S (best is 0)
DRAM READ CAS WAIT STATE = 3 W/S (best is 2)
EXTERNAL CACHE = DISABLED

System is completely stable.
Cyrix FasMath used for PC Player Benchmark and the complex offline graphics tests.
ULSI DX/DLC used for Quake 1.

SpeedSys hangs, which is often the case with SXL2 chips.

Great Quake 1 score.
Slow at everything else.

--- AMD 386DX at 50MHz, ISA at 12.5MHz

All BIOS settings on max.
System is completely stable.
Nothing more to add really.

Below average performance.

---

benchmark results

---

Great potential in this assembly.
Unfortunately there is this trouble with level 1/2 caches for BL3/SXL2 processors when FPU is present.
If i set C000-DFFF memory range to not cacheable then all is good, but performance drops a lot in this case which defeats the purpose.
There is a dedicated option in the BIOS for that.
Any ideas ?

pshipkov wrote on 2023-03-25, 21:46:

On multiple occasions Feipoa spoke highly of his Chaintech 340SC motherboard in combination with IBM BL3 and TI 486SXL2 processors, so i wanted to give that stuff a try.

I only spoke highly of the Chaintech 340SCD in combination with the IBM BL3, not the SXL2. SXL2 performance isn't great.

pshipkov wrote on 2023-03-25, 21:46:

--- IBM BL3 at 90MHz (2x50), ISA bus at 12.5 MHz

Typo. I think you meant '90 MHz (2x45)'.

pshipkov wrote on 2023-03-25, 21:46:

All BIOS settings on max, except:
BUS CLOCK SELECT = SCLK/8 (best is /3)
EXTERNAL CACHE = DISABLED

From my experience with my motherboard, faster ISA speed doesn't necessarily translate to faster [Doom] benchmark scores. Depending on the FSB, the optimal ISA speed is somewhere in the 11.1 - 15 MHz range. Your IDE issues may or may not be related to ISA speed.

pshipkov wrote on 2023-03-25, 21:46:

--- TI 486SXL2 at 80MHz (2x40), ISA at 13.33MHz I am still fresh with the SXL2 adapter stuff. My only SXL2-66 CPU does not scale […]
Show full quote

--- TI 486SXL2 at 80MHz (2x40), ISA at 13.33MHz
I am still fresh with the SXL2 adapter stuff.
My only SXL2-66 CPU does not scale past 80MHz.
System is not stable - it is the CPU itself.

Were you able to get 80 MHz stable? If so, at what minimal voltage?

pshipkov wrote on 2023-03-25, 21:46:

If i turn-off C000-DFFF are is not cacheable all is good, but performance drops a lot in this case which defeats the purpose.

Typo to the extent that I don't understand what is being conveyed. This looks like a range in RAM to, for example, shadow your SCSI card's BIOS.

Fixed the 2x45 and the sentences at the end about non-cached memory range.

SXL2 at 2x40 is stable at max voltage and with at least 5V Peltier.
Even with 12V Peltier the system is unstable.
Tried other motherboards - same result.
The CPU is not very good.

Isn't C000-DFFF the reserved region of memory for expansion ROMs and if so, why would not caching this region kill performance? When I run an SXL, for example, I usually set the 640-1024K as non-cacheable (via cyrix.exe) to avoid trouble in Windows.

Too bad about your sole SXL2 being a lemon. I guess you should have grabbed one or two from CPUshack. Are you able to at least run 75 MHz at 4 V?

This address space is shadow memory.
C000-C800 is usually VGA BIOS, but the 16Kb segments are chipset dependent, to a point at least.

Don't understand why this was a problem for L1/L2 caches to work simultaneously.
Bothers me - going to try again tomorrow.

feipoa wrote on 2023-03-25, 23:00:

I only spoke highly of the Chaintech 340SCD in combination with the IBM BL3, not the SXL2. SXL2 performance isn't great. […]
Show full quote

pshipkov wrote on 2023-03-25, 21:46:

On multiple occasions Feipoa spoke highly of his Chaintech 340SC motherboard in combination with IBM BL3 and TI 486SXL2 processors, so i wanted to give that stuff a try.

I only spoke highly of the Chaintech 340SCD in combination with the IBM BL3, not the SXL2. SXL2 performance isn't great.

pshipkov wrote on 2023-03-25, 21:46:

--- IBM BL3 at 90MHz (2x50), ISA bus at 12.5 MHz

Typo. I think you meant '90 MHz (2x45)'.

pshipkov wrote on 2023-03-25, 21:46:

All BIOS settings on max, except:
BUS CLOCK SELECT = SCLK/8 (best is /3)
EXTERNAL CACHE = DISABLED

From my experience with my motherboard, faster ISA speed doesn't necessarily translate to faster [Doom] benchmark scores. Depending on the FSB, the optimal ISA speed is somewhere in the 11.1 - 15 MHz range. Your IDE issues may or may not be related to ISA speed.

pshipkov wrote on 2023-03-25, 21:46:

--- TI 486SXL2 at 80MHz (2x40), ISA at 13.33MHz I am still fresh with the SXL2 adapter stuff. My only SXL2-66 CPU does not scale […]
Show full quote

--- TI 486SXL2 at 80MHz (2x40), ISA at 13.33MHz
I am still fresh with the SXL2 adapter stuff.
My only SXL2-66 CPU does not scale past 80MHz.
System is not stable - it is the CPU itself.

Were you able to get 80 MHz stable? If so, at what minimal voltage?

pshipkov wrote on 2023-03-25, 21:46:

If i turn-off C000-DFFF are is not cacheable all is good, but performance drops a lot in this case which defeats the purpose.

Typo to the extent that I don't understand what is being conveyed. This looks like a range in RAM to, for example, shadow your SCSI card's BIOS.

SiS Rabbit chipset needs memory interleave for extra performance, you are using 4x4MB modules for total of 16MB, is this correct?
Then otherwise try 8x 1MB for 8MB or 8x 4MB for 32MB and retest the performance, in other words fill all the memory module slots.

Cheers,

Was not aware about that.
Where did you learn it ?

I suppose this will help SXL2 and 386DX chips.
BL3 can cache 16Mb only. Bigger buffer than that and perf drops massively.
8Mb is not interesting. Omitting it right away.

I have the board back on the test bench.
First thing will be to check one 2 ram banks.

Patriot SL4100 BL3 is busting my balls still, fix one niggle, get 5 others.... I joked somewhere about boiling it in deoxit... that's starting to sound more sensible by the day... about a week ago I gave it a test run at 40Mhz and it acted possessed... the turbo switch was hung up and it decided to start half working. on of on off, randomly... and I got hung up on fixing the turbo switch, searching everywhere for a spare latching switch, swear I must have one, but after getting distracted with more stuff I found while looking for it, I rigged a small toggle for now. Thing is I think if you've got a delicate CPU starting to overheat and you kill all power, the die could cook with no airflow and bake for a few mins, whereas if you slam it into deturbo first, at least it's got airflow for a bit and substantially reduced power, might save it in an edge case. .. Though I am not so sure now, now the turbo is working it only drops performance by half... I think the board is taking the AT/2 clock off the Avsem instead of the 8mhz clock for turbo, which is disappointing (for XT games)

Anyhoo, that was all a fiddle and waste of time, then I wanted to stick a swappable HDD bay on the 2nd IDE, dump results, export import, easy peasy... apart from the secondary IDE ain't working. Tried about a dozen combos of drive and cable now. I started messing with jumper but merely managed to kill the PS/2 ports for half a day until I figured out that I had set stuff back right, just the jumpers were nasty.

I was proceeding under the assumption that I had default jumper settings and pics of board on twitter had default jumper settings and therefore everything should work. So thought it all through from scratch as Sherlock Holmes this evening and had a big DOH!... This system came stock with a multimedia upgrade, 16 bit sound blaster, 2X CDROM, I had mine missing but the CD audio cable loose in the case should have clued me in. Mr Berner on Twitter has his CDROM IDE cable draped over the riser like it got unplugged from a card over there. Secondary IDE is probably disabled in default shipped configuration for this system because they left the soundcard to do IDE for CDROM. Riiight, so now I have suspicions about a different jumper block.... gah, took me long enough. I gotta study the durned thing though, jumpers go every which way in this system, gotta try to see if they're boxed on the silk screen in groups.

One single piece of good news is I got a 16MB 60ns SIMM working in it, wasn't real happy with the janky collection of fours in tired sockets. Red Hill hardware guide in the CPU section 2nd 486 one I think talking about BL3 boards says they (Think it's a cougar pic on site) never take double bank/sided SIMMs, no 8MBs... well yah this one didn't like a damn one, even IBM made/marked ones.

Pentiumspeed, the issue with BL3 is the 16 MB cacheable limit. To get interleaved, need to use 8 MB (8x 1 MB SIMMs). I don't think 2 MB SIMMs were commonly available for the PC, were they? I also had to run my SiS Rabbit with 4x 4 MB SIMMs, but I cross-checked it with 8 x 1 MB and didn't notice any DOOM benchmark drop.

Spent more time today with the ABit Ab-FS3 (SiS Rabbit) motherboard from few posts above.

Single vs double banked RAM - similar to what Feipoa said above, there is no performance difference between the two. Checked graphics, offline compute, local storage i/o. DOS and WIN.

Level 2 cache instabilities:
Cycled quite a bit on that.
Initially i thought it is related to FPUs, but it is the assembly itself. It cannot handle level 2 cache at 50MHz FSB. Even with most conservative timings.
It is not the cache chips. They are curated set for 60/66MHz + tightest timings. I use them for 486 overclocking activities.

Apparently Chaintech's assembly based on this chipset is much more successful.

---

@BitWrangler

The common series of unfortunate events.
Did you look for IBM manual for the mobo ?
They were diligent and a lot of their stuff is still preserved today.
But also, you are taking this through the long route.
Hot-swappable HDDs and what not.
I would discard all that stuff and make the system minimalistic performance hotrod. : )

pshipkov wrote on 2023-03-27, 06:25:

It cannot handle level 2 cache at 50MHz FSB. Even with most conservative timings.
It is not the cache chips. They are curated set for 60/66MHz + tightest timings. I use them for 486 overclocking activities.

Does everything work just fine at 45 Mhz or less FSB? If so, it might be best to start testing with lower clocked crystals oscillators, then up the frequency once a slow base frequency is working well.

That was quantified carefully.
Level 2 cache works well at 40 and 45 mhz, but 45mhz + level cache is slower than 50mhz without level 2 cache.
Apparently ram access is pretty good.
Also, the level 1 cache of bl3 and sxl2 help a lot to mitigate the issue.
This is evident by the shared top Quake 1 score.

It is slower because you have to add some wait states to make the system stable or slower because the board or BIOS is doing something wonky?

Both 45 and 50 MHz run with all BIOS settings on max, but without level 2 cache.
The extra 5MHz is enough to outrun the 40MHz + level 2 cache.
50MHz is unstable no matter what.

This is curious. I don't think I grabbed cachechk/speedsys images for the case of 2x50 on the BL3 w/SiS Rabbit, but I'm pretty sure cachechk showed a step at 256K for L2. If your system is still out and available, could you provide a Speedsys and Cachechk screenshot for 2x50 when L2 is enabled? In the screenshot you've shown above, you listed L2 as being disabled.

EDIT: What was your DOOM result at 2x50 with BL3? My notes indicate my SiS Rabbit with GD5434 at 2x50 BL3 gets 26.15 fps in DOOM. If your L2 is not working, I suspect your result to be less than this.