Having only ever used VLB 486 motherboards, I figured it would be fun to mess around with some PCI variants. I bought these two boards from a Ukrainian eBay seller and after some minor repairs both are in tip-top shape. One is the infamous Lucky Star 486e rev C1, and the other is an SMT 4321-22B "Terminator".
But the "Terminator" with it's ALI M1487/M1489 chipset seems to be far less common. I've read speculation that the M1487/M1489 might be a good performer too, but haven't seen any detailed comparisons yet. In fact there doesn't seem to be much information about this chipset at all.
For the purpose of a fair comparison, both boards have IBM 5x86C-100 CPU's installed, and for the moment neither have been overclocked (33MHz x3). Both have 256K of L2 cache, an identical stick of 8Mx32 60NS EDO DRAM, and the testing was performed using the same Matrox Millennium PCI video board.
I've performed some basic benchmarks on the pair to compare their performance, and at first glance the ALi chipset seems to have a pretty significant edge against the SIS 496/497. Results are below...
Last edited by FesterBlatz on 2017-03-09, 16:08. Edited 9 times in total.
It's obvious by looking at the PC Player benchmark results that the ALi M1487/M1489 is outperforming the SIS 496/497 by a significant margin!
Looking closer at the Speed Test results reveals the RAM performance of the ALi seems to be quite a bit better than the SIS, even with the reducded DRAM Read timings I had to use to make the system stable.
At this point, at least at stock 33MHz bus speeds, the ALi M1487/M1489 seems to be the faster chipset.
But is it really as simple as the ALi being "better" chipset than the SIS, or am I overlooking something? Does anyone familiar with either of these boards have any comments or suggestions on any particular settings I may need to adjust on either board that may account for such a dramatic difference?
Interesting results. It kind of surprised me, since the SiS496/497 is considered one of the fastest 486 PCI chipsets. I suspect the memory performance differences may be related to the fact that you are using EDO memory, which not always means better performance in 486.
I certainty can, but I believe my only true FPM sticks are 4Mx36 and I don't think the LS-486 BIOS has an option to disable parity checking like the SMT board does. Do you know if the SIS 496/497 chipset will automatically use the parity bits when available, or does it ignore them? I'd like to be sure it's an even comparison...
Also, I was under the impression the SIS 496/497 Stepping B4 and newer (I believe mine to be B5) also supports EDO RAM, although nothing in the POST summary screen confirms it like the ALi board does. Maybe ALi managed to get "functional" EDO support where SIS didn't?
Either way, I'll rerun the tests tonight with my FPM sticks and post my results.
Do you know if the SIS 496/497 chipset will automatically use the parity bits when available, or does it ignore them? I'd like to be sure it's an even comparison...
I've taken a look at the SiS496/497 datasheet posted by NJRoadfan here:
It looks as if the chipset supports enabling/disabling parity checks by setting a bit flag on a register, but if there is no option in the BIOS for that then maybe the motherboard automatically enables it when parity RAM is detected, so it's difficult to know.
FesterBlatz wrote:
Also, I was under the impression the SIS 496/497 Stepping B4 and newer (I believe mine to be B5) also supports EDO RAM, although nothing in the POST summary screen confirms it like the ALi board does. Maybe ALi managed to get "functional" EDO support where SIS didn't?
Yeah, that may probably be the case, I've read a few cases here on Vogons where people had trouble running EDO properly (if at all) on some 486 PCI boards that claim to support it. EDO is definitely supported by the SiS (both the datasheet and your own tests are proof of that), the question is how good is that support.
That could be useful to compare your results with the UMC 8881F/8886BF, which is another fast 486 PCI chipset. Those tests use FPM and a Matrox G200, but for the same 5x86 CPU it should be a fair comparison with your systems.
and I *think* that I used FPM only (though as there is no way of knowing if each sticks are EDO or FPM, I'm not really sure ...)
One thing to note : you can install a driver on the SiS chipset for MS-DOS that seems to speed up HDD access quite a bit, this may (or may not) help with performance
Edit : I checked the ram sticks and 2 of them out of 3 were EDO. Then I put sticks fromm 1993 and 1994 in the machine assuming that they are FPM memory but I'm getting the same results as before
It looks as if the chipset supports enabling/disabling parity checks by setting a bit flag on a register, but if there is no option in the BIOS for that then maybe the motherboard automatically enables it when parity RAM is detected, so it's difficult to know.
Another option is I could always remove the 9th chip from those sticks to convert them to 4Mx32...
I've glanced over it before but I just read it in it's entirety, very interesting! Quite a bit of information to absorb...
The LS-486e manual I have doesn't explain what the individual CPU selection jumpers are, only where to put them for a particular processor. Could anyone share a more detailed explanation of the jumpers (PCI bus divider, CPU multiplier, etc.)?
I wonder if the SMT Terminator could be made to run the FSB at 66MHz. The PLL chip is labeled ARK929C but it looks to me like a "rebadged" chip to hide it's true part number. I can't find any datasheets for it.
The LS-486e manual I have doesn't explain what the individual CPU selection jumpers are, only where to put them for a particular processor. Could anyone share a more detailed explanation of the jumpers (PCI bus divider, CPU multiplier, etc.)?
I wonder if the SMT Terminator could be made to run the FSB at 66MHz. The PLL chip is labeled ARK929C but it looks to me like a "rebadged" chip to hide it's true part number. I can't find any datasheets for it.
If I remember correctly.
When looking at your image above. The 3 jumpers to the right of the clock generator controls the FSB. The one above to the left of the clock generator controls the PCI divider.
New PC: i9 12900K @5GHz all cores @1.2v. MSI PRO Z690-A. 32GB DDR4 3600 CL14. 3070Ti.
Old PC: Dual Xeon X5690@4.6GHz, EVGA SR-2, 48GB DDR3R@2000MHz, Intel X25-M. GTX 980ti.
Older PC: K6-3+ 400@600MHz, PC-Chips M577, 256MB SDRAM, AWE64, Voodoo Banshee.
The 3 jumpers to the right of the clock generator controls the FSB.
The one above to the left of the clock generator controls the PCI divider.
Thanks! I had the FSB ones figured out but wasn't sure about the PCI divider. Do you know which jumpers are used to enumerate the various CPU multiplier pins?
The 3 jumpers to the right of the clock generator controls the FSB.
The one above to the left of the clock generator controls the PCI divider.
Thanks! I had the FSB ones figured out but wasn't sure about the PCI divider. Do you know which jumpers are used to enumerate the various CPU multiplier pins?
I would start with tinkering with the two at the top left of the CPU, the ones that are not closed at the moment.
Otheriwise look for differences between 2x and 3x for AMD / Intel WB DX4 CPUs in the manual, the multiplier setting for those CPUs should be the same as 4x and 3x for the AMD DX5. The Cyrix 586 could perhaps use another register to set the multiplier, I have not used one in 20 years. 😁
New PC: i9 12900K @5GHz all cores @1.2v. MSI PRO Z690-A. 32GB DDR4 3600 CL14. 3070Ti.
Old PC: Dual Xeon X5690@4.6GHz, EVGA SR-2, 48GB DDR3R@2000MHz, Intel X25-M. GTX 980ti.
Older PC: K6-3+ 400@600MHz, PC-Chips M577, 256MB SDRAM, AWE64, Voodoo Banshee.
When looking at the manual, do you see what I mean about lack of any detail regarding multipliers? When I get some time tonight I'll look closer and see if I can figure out which jumpers are for the multipliers.
EDIT: I also have reason to suspect this is the clock generator for the SMT Terminator. Looking at this datasheet it also supports up to 66MHz FSB and would be an easy mod to enable. I could probably epoxy an angled pin header right to the top of the clock generator to add third jumper.
Considering how well this board performs with just a 33MHz FSB, I'm curious to see how one of my IBM 5x86c's would do @ 66x2 assuming the motherboard and CPU would be up to the task. Perhaps at least one of my three 5x86's would handle it.
The only problem is 66MHz is pushing 15ns SRAM to it's absolute limit...
Your results are very interesting. I would not have thought the difference so large. Would you be able to run DOOM timedemo 3 and Quake timedemo 1 using a Trio64V2/DX and an Am5x86-160 to compare with my results?
From my past tests, the ALi board has faster memory read times than SiS 496 or UMC 8881. Faster than SiS 471 as well. For memory read times using cachechk, I believe UMC was 48 MB/s, SiS was 51 MB/s, and ALi at 54 MB/s. The issue for me was that the RAM write time is capped at 54 MB/s, while SiS and UMC get 83 MB/s. Ideally, I would like to run a more thorough comparison and update this thread, Performance comparison of 486 motherboards with VLB-only, PCI-only, and PCI+VLB
I recall my Ali board wasn’t able to use the fastest cache timings when a 40 MHz FSB was used. This really limits its usefullness with an Am5x86-160 or Cyrix 5x86-120. I had 512K of cache installed, which can sometimes influence what fastest cache timings are stable.
I do not believe that enabling parity will benefit the benchmarks, but if the option is not showing up in the BIOS, you should use modbin to unhide the parity setting, then save the new BIOS image and reflash it your motherboard’s EEPROM.
I have never witnessed first hand or second hand any improved benchmark results when using EDO on a PCI 486 motherboard. On the contrary, using EDO RAM sometimes requires reducing your cache or memory timings in the BIOS. From memory, the only SiS 496 chipset which people claim to fully support EDO RAM has a PR, PU, or PV revision to it. I really only remember it starts with a ‘P’ and is two characters.
You’ll want to set your cache to its slowest setting, PCI divisor to 1/2, and RAM read to 2ws or more. Perhaps RAM write to 1 ws. And stick with 256K cache. You could also source a 12 ns TAG, but I’m not convinced this is necessary at this stage.
Plan your life wisely, you'll be dead before you know it.
Hey feipoa, I was hoping you'd chime in! 😀 Thanks for the detailed reply!
I'll have to take a look at modbin, it would certainly be useful to have some control over parity since most of my DRAM modules are 36-bit. Parity should theoretically decrease performance since it's more overhead for the chipset to calculate the parity bits, so for the purpose of testing I was hoping to ensure parity isn't being used on either board.
Do the LS-486e boards normally come with an EEPROM? I sort of assumed a low-buck motherboard like these tiny 486 PCI boards would have come with cheaper OTP PROMs. I'd rather not tear the sticker off to find out (I'm a stickler for details), but I suppose I could drop the chip into to one of my device programmers and see if they can identify the chip using the device ID bits. Either way, I have plenty of spare EPROMs at my disposal.
My limited research found that SIS 496/497 stepping B4 and newer is supposed to support EDO RAM, and since my chipset pair is PR/NU, I believe I have stepping B5 or even B6... but there isn't much information available. Either way, based on the stuff I've read here, it seems the SIS 496/497 EDO support is useless anyway.
I didn't have as much time last night as I had hoped, but I did mange to run some benchmarks on the ALi chipset with regular FPM DRAM installed. Sure enough the memory benchmarks dropped by a pretty noticeable amount, so it does seem like the ALi chipset will indeed leverage EDO RAM when available and unlike SIS, provides a tangible improvement in performance. I didn't have time to save the results, but I hope to get back to it tomorrow since I'll have the house to myself for most of the day.
I'd be happy run some time demos. I have a two AMD 5x86-133's I can attempt to run at 160MHz, but my only PCI graphics board my Matrox Millennium so I don't know how useful the results would be for a meaningful comparison. If anyone in the states has a spare S3 Treo64 they'd like to send I'd gratefully accept, otherwise I can hunt one down on eBay. I've been planning to find an S3 board anyway, although I was thinking more along the lines of the ViRGE variety.
I'm not necessarily gunning for a 66MHz FSB, I was just surprised to see how much better this ALi based motherboard performed compared to the SIS board. If it turns out the ALi really does outperform an SIS at 33MHz by such a significant amount, it would be interesting to see if stable operation at 66MHz is possible. Based on the 33MHz results there appears to be some real potential here!
As for the CPU voltage needed to run a 5x86c at 2x66MHz, this board supports 4 different voltages. Since I'd likely never need to use the 4V option, I could always replace the relevant divider resistor to change the 4V selection to 3.65V. Otherwise, yep I could sneak a trimpot in there somewhere too. I need to take a closer look at the VRM circuit but I think it uses an adjustable voltage reference like the TL431 and a PNP pass transistor to carry the current.
Besides needing to add a 3rd clock speed jumper (S2) to the clock generator, I think the most difficult part of achieving 66MHz might be figuring out which jumpers are the PCI divider (assuming it has one) and CPU multiplier. This board isn't as popular as the LS-486 there's barely any information available.
For reference, I've attached a manual and board layout if anyone wants to take a look.
That is fascinating. I'm looking forward to your test results of EDO vs. FPM on this ALI board. I did not bother to check mine with EDO.
Most PCI 486 boards did not come with OTP; the majority came with UV EPROM chips. The ones with OTP were usually PC CHIPS. Of course, some came with EEPROMs as well. If you don't want to remove the sticker, you can usually tell if it is UV EPROM by noticing the thickness of the chip and the sandwich look to it. For OTP vs. EEPROM, sometimes you can tell from the markings on the bottom.
I was under the impression that the ViRGE and Trio64V2+ results were about the same, however, I have not seen a comparison of such performed.
All my PCI 486 boards have a VRM with a Vadj. resistor and are very easy to mod. with a trimmer. I'd use a trimmer rather than a fixed resistor if you're testing for an IBM 5x86-133. One of my cyrix 5x86-120 chips needed 3.85 V to run at 133 MHz. Others will run well at 3.6 V.
If the ALI board has a 1/2 PCI divisor, it might be near the PLL clock generator circuit. That is the case with SiS 486 boards anyway.
Plan your life wisely, you'll be dead before you know it.
Most PCI 486 boards did not come with OTP; the majority came with UV EPROM chips. The ones with OTP were usually PC CHIPS. Of course, some came with EEPROMs as well. If you don't want to remove the sticker, you can usually tell if it is UV EPROM by noticing the thickness of the chip and the sandwich look to it. For OTP vs. EEPROM, sometimes you can tell from the markings on the bottom.
I've got a (dead) Jetway J-446A SIS 496/497 motherboard with an OTP PROM for the BIOS, that's why I figured it may be common practice. That board won't complete it's post and stops on a memory testing code (C0 or C6) so I think it needs the chipset replaced. Hopefully I eventually stumble on a cheap/free SIS 496/497 that was destroyed by a leaking battery so I can rob it for the chipset....
Anyway, the LS486e BIOS PROM definitely isn't UV erasable since it's not a ceramic package. I'll stick it in the device programmer and see if the chip will IDs itself. If it turns out to be an EEPROM that's great, but if it's only OTP like the Jetway board I'll an either use an EPROMs or whip up a carrier board to plug into the socket and use a 32-LLC 28F010 EEPROM I have a small stockpile of. Hopefully it's an EEPROM as you suggest.
Modbin was pretty easy to figure out so I've already got a binary with the "MEMORY PARITY CHECK" toggle un-hidden as well as "CPU Burst Write"...whatever that does.
EDIT: I forgot to ask, I noticed there's a section in Modbin devoted to Cyrix configuration registers. I don't suppose it would be possible to adjust those to enable some of the performance enhancing features that normally need to be enabled from a DOS utility?
Last edited by FesterBlatz on 2017-03-10, 20:58. Edited 3 times in total.
I was under the impression that the ViRGE and Trio64V2+ results were about the same, however, I have not seen a comparison of such performed.
They are!
In VGA the Virge 325 is less than 2% faster, but it's almost 5% faster in SVGA. At least with my cards, averaged over multiple DOS games.
The Virge GX is the fastest S3 card I tested in DOS (about another 2% faster than the Virge 325).
When I put them in a P3-933 though, you could see just how much faster the GX really is (~15% above the Trio64V2+). At 486 speeds, they would probably perform 1% or less different.
I plan to do some timedemo benchmarks with my Matrox Millennium tomorrow anyway, but I'll run through them again when the S3 board arrives.
Speaking of timedemos, whenever I use the Quake timedemo that's bundled in "Dosbench" on any of my 486 motherboard/CPU combos it crashes right after the player gets the yellow key. Has anyone else experienced this?
