This simple 😀

Bang on with 1k

Could you post a bigger picture of the 1MiB cache module?
I recently wanted to acquire a 919 but did not because of the PCI bus divider.

bertrammatrix wrote on 2024-11-24, 18:32:

Because of how the 3.3 - 4 volt selection jumper JP4 is implemented in circuit, all one has to do to modify the voltage to anything between those two values is to vary the resistence between the pins of JP4! So basically you can just plug your 5k potentiometer in here, no soldering to board required. It can't get much better then that right? But it does! Turns out that the value you need to have right around 3.6 volts is 1kohm, so a resistor of that value is all you need. Obviously the resultant voltage may differ slightly from board to board, however it should be in the ballpark as this value worked for both my v3.4 and v1.5 board

Nice! So this runs 1K in parallel with whatever SMD resistor is on the motherboard? I ran a similar configuration with parallel resistors on a Lucky Star motherboard because I didn't want to remove the onboard resistor. One issue is that the range of voltages you can select wasn't as broad as replacing the resistor itself. I think I ultimately replaced the resistor years later. However, if you only want to run at 3.6 V, then your solution has much beauty in its simplicity. This is something that just about anyone can do.

amadeus777999 wrote on 2024-12-13, 09:27:

Could you post a bigger picture of the 1MiB cache module?
I recently wanted to acquire a 919 but did not because of the PCI bus divider.

What aspect of the PCI bus divisor do you not like? Is it the automatic 2/3 multiplier when FSB is run at 40/50 MHz? Or is it the 1/2 multiplier when the FSB is run at 60 MHz? You can get around this by connecting a switch to the FSB jumper to boot at 33 MHz, then flip the switch at the DOS prompt. It's not very elegant, but effective enough. Another option is to use VLB for graphics. I run my M919 at 3x60 MHz, and if memory serves me, the VL bus is still run at 60 MHz. I'd have to double check to be sure though.

amadeus777999 wrote on 2024-12-13, 09:27:

Could you post a bigger picture of the 1MiB cache module?
I recently wanted to acquire a 919 but did not because of the PCI bus divider.

It is the original 256k module. I was trying to get someone from here to whip me up one of the 1mb badboys but they ended up ghosting me. I am a sucker for cache though and tracked one down on ebay, it is enroute to me now, or whenever are striking postal service gets back to work should I say.

The seller actually has one more of these listed with a m919 for a pretty decent price right now, maybe it's just what you're looking for

After much performance comparisons- an m919 will beat almost anything else even with only 256k of cache, so really it's not that big a deal

feipoa wrote on 2024-12-13, 10:10:

Nice! So this runs 1K in parallel with whatever SMD resistor is on the motherboard? I ran a similar configuration with parallel […]
Show full quote

bertrammatrix wrote on 2024-11-24, 18:32:

Because of how the 3.3 - 4 volt selection jumper JP4 is implemented in circuit, all one has to do to modify the voltage to anything between those two values is to vary the resistence between the pins of JP4! So basically you can just plug your 5k potentiometer in here, no soldering to board required. It can't get much better then that right? But it does! Turns out that the value you need to have right around 3.6 volts is 1kohm, so a resistor of that value is all you need. Obviously the resultant voltage may differ slightly from board to board, however it should be in the ballpark as this value worked for both my v3.4 and v1.5 board

Nice! So this runs 1K in parallel with whatever SMD resistor is on the motherboard? I ran a similar configuration with parallel resistors on a Lucky Star motherboard because I didn't want to remove the onboard resistor. One issue is that the range of voltages you can select wasn't as broad as replacing the resistor itself. I think I ultimately replaced the resistor years later. However, if you only want to run at 3.6 V, then your solution has much beauty in its simplicity. This is something that just about anyone can do.

amadeus777999 wrote on 2024-12-13, 09:27:

Could you post a bigger picture of the 1MiB cache module?
I recently wanted to acquire a 919 but did not because of the PCI bus divider.

What aspect of the PCI bus divisor do you not like? Is it the automatic 2/3 multiplier when FSB is run at 40/50 MHz? Or is it the 1/2 multiplier when the FSB is run at 60 MHz? You can get around this by connecting a switch to the FSB jumper to boot at 33 MHz, then flip the switch at the DOS prompt. It's not very elegant, but effective enough. Another option is to use VLB for graphics. I run my M919 at 3x60 MHz, and if memory serves me, the VL bus is still run at 60 MHz. I'd have to double check to be sure though.

Ah Feipoa, finally we meet so to say, I wasn't sure if you were still active on here. Your posts have been very helpful for torturing my hardware, thanks for all the dedication to the 5x86. Early this year there was a reject wafer with developed cores and some engineering notes for sale on the bay, I secretly hope you were the one who got it 😁

Yes. Tweaking the resistance one can get basically anything between 3.3v - 4v. Obviously it isn't possible to get the voltage any higher then 4 volts this way, so people wanting to fry an egg on an amd 5x86 at 180/200 that needs something in that range still have to do it the complicated way.

I have a v1.x revision of the board and the voltage produced with this mod is identical.
I also have an m918 (I was trying to figure out which is a bigger money pit...) and though I haven't tried this on it I think it may work as well as the voltage regulator circuit /jumpers look near identical. I'll give it a try and post it for the record next time I fire it up.

haha, not me. My purchases of vintage hardware have dropped by 95% or so. If the listing has been sold, could you provide the eBay URL?

I also have an M918, but I found the M919 to be the better board, at least once the 1024K modules came about, and that strange 64 MB performance hole figured out.

feipoa wrote on 2024-12-14, 06:41:

haha, not me. My purchases of vintage hardware have dropped by 95% or so. If the listing has been sold, could you provide the eBay URL?

I also have an M918, but I found the M919 to be the better board, at least once the 1024K modules came about, and that strange 64 MB performance hole figured out.

I think it's long gone/ sold/cant find any sign of it.

An interesting thing I found with the m918 was that at 60mhz fsb I could run 512k of quality cache at 3-1-1-1 whereas the 919 v1.5 could only handle 3-2-2-2 with the same chips, however, cachecheck would show identical performance, so the 918 is a bit of a slouch. Where it had an edge was being able to run the pci at 60mhz, had some pretty impressive dosbench scores, however I did have an unexpected board failure after long term use like this (if it's even related) so I haven't been running my second 918 that way. Also, whilst trying to (unsuccessfully) coax 5x86c's to run at 66mhz fsb I made it further on the 918 then anything else, I could run (but eventually crash) quake at 133mhz whereas I could barely make the 919 complete post. Clearly the Ali chipset is not all that bad (even though I found it to be a little finicky to set up so it was 100% windows stable)

Indeed, the dreaded but relatively unknown cacheable memory bug. My v3.4 b/f doesn't have it but the v1.5 sure does (around 53mb iirc). I swapped bioses and the v1.5 behaved flawlessly (installed windows 98) with the bug free 3.4 bios. There is a stock bios chip from a v3.3 inexpensively available on Ebay, I was considering buying it - do you have any idea if it still has the bug? Unfortunately it's not in a board thus the seller has no way of checking the ID string

bertrammatrix wrote on 2024-12-14, 05:17:

It is the original 256k module. I was trying to get someone from here to whip me up one of the 1mb badboys but they ended up gh […]
Show full quote

amadeus777999 wrote on 2024-12-13, 09:27:

Could you post a bigger picture of the 1MiB cache module?
I recently wanted to acquire a 919 but did not because of the PCI bus divider.

It is the original 256k module. I was trying to get someone from here to whip me up one of the 1mb badboys but they ended up ghosting me. I am a sucker for cache though and tracked one down on ebay, it is enroute to me now, or whenever are striking postal service gets back to work should I say.

The seller actually has one more of these listed with a m919 for a pretty decent price right now, maybe it's just what you're looking for

After much performance comparisons- an m919 will beat almost anything else even with only 256k of cache, so really it's not that big a deal

Looking forward to a pic once you have it.
What kind of score do you get with the 919 in Doom shareware - "-timedemo demo3" with HUD and one green border?
As for the cache - even 128KiB gets you a really good boost but I had 1MiB on a GA-486AM/S(50/150 & all timings tightest) and it yielded the best Doom "timedemo3 score" I had ever seen... it was only some hairs over a 1000 realtics.

amadeus777999 wrote on 2024-12-14, 17:34:

Looking forward to a pic once you have it. What kind of score do you get with the 919 in Doom shareware - "-timedemo demo3" with […]
Show full quote

bertrammatrix wrote on 2024-12-14, 05:17:

It is the original 256k module. I was trying to get someone from here to whip me up one of the 1mb badboys but they ended up gh […]
Show full quote

amadeus777999 wrote on 2024-12-13, 09:27:

Could you post a bigger picture of the 1MiB cache module?
I recently wanted to acquire a 919 but did not because of the PCI bus divider.

It is the original 256k module. I was trying to get someone from here to whip me up one of the 1mb badboys but they ended up ghosting me. I am a sucker for cache though and tracked one down on ebay, it is enroute to me now, or whenever are striking postal service gets back to work should I say.

The seller actually has one more of these listed with a m919 for a pretty decent price right now, maybe it's just what you're looking for

After much performance comparisons- an m919 will beat almost anything else even with only 256k of cache, so really it's not that big a deal

Looking forward to a pic once you have it.
What kind of score do you get with the 919 in Doom shareware - "-timedemo demo3" with HUD and one green border?
As for the cache - even 128KiB gets you a really good boost but I had 1MiB on a GA-486AM/S(50/150 & all timings tightest) and it yielded the best Doom "timedemo3 score" I had ever seen... it was only some hairs over a 1000 realtics.

Sometimes I wonder how conclusive some of the benchmarks are. I don't typically run any of the benchmarks from the "dosbench" suite with anything other then default settings for simplicity sake. Sure, one can tweak benchmark settings for a better benchmark result, but I'm more interested in real world/ overall gains as I like pain (clearly) and I run win 98se on these systems. My opinion is that speedsys and (especially) cachecheck numbers are much more indicative of overall system speed then some of the game benchmarks which seem to be heavily influenced by the particular vga and its bios. As an example I've gotten scores of over 90 in 3dbench on a 918 with a S3VirGe, but, a 919 with a cirrus 5464 that scores only on the 70s is noticeably faster under demanding windows games like c&c tiberian sun (not that most would play this on a 486)

amadeus777999 wrote on 2024-12-14, 17:34:

Looking forward to a pic once you have it. What kind of score do you get with the 919 in Doom shareware - "-timedemo demo3" with […]
Show full quote

bertrammatrix wrote on 2024-12-14, 05:17:

It is the original 256k module. I was trying to get someone from here to whip me up one of the 1mb badboys but they ended up gh […]
Show full quote

amadeus777999 wrote on 2024-12-13, 09:27:

Could you post a bigger picture of the 1MiB cache module?
I recently wanted to acquire a 919 but did not because of the PCI bus divider.

It is the original 256k module. I was trying to get someone from here to whip me up one of the 1mb badboys but they ended up ghosting me. I am a sucker for cache though and tracked one down on ebay, it is enroute to me now, or whenever are striking postal service gets back to work should I say.

The seller actually has one more of these listed with a m919 for a pretty decent price right now, maybe it's just what you're looking for

After much performance comparisons- an m919 will beat almost anything else even with only 256k of cache, so really it's not that big a deal

Looking forward to a pic once you have it.
What kind of score do you get with the 919 in Doom shareware - "-timedemo demo3" with HUD and one green border?
As for the cache - even 128KiB gets you a really good boost but I had 1MiB on a GA-486AM/S(50/150 & all timings tightest) and it yielded the best Doom "timedemo3 score" I had ever seen... it was only some hairs over a 1000 realtics.

Here's pictures from the auction. It is one of Pancakepuppys (from here) modules that are floating around, so there's likely better pictures in the 1mb cache threads. It's probably doubled in price by now 😁 but it is what it is. Can't wait to benchmark with it, sounds like those 12ns chips are good at fast timings

Benchmarks results:

2x60mhz, 0ws write/1ws read, 256kb 3-2-2-2 in WT mode, 64 mb of 50ns EDO, with the more important cyrix cpu features enabled with peter mosses utility

Cachecheck

Doom max detail

PC player

Quake (note this is on the low side and I've definitely gotten higher scores with a virge)

There are quite a few variances which can affect operation at 66 MHz FSB, aside from the chipset, like motherboard differences in the "3V3" voltage output, BIOS timings, differences in hardware implementation of onboard memory buffers, PCB trace routing, and how the MB manufacturer handled parasitics, etc.

My notes on the M918 are poor, but this is what I was able to find: "Tested Am5x86 at 100 MHz. Using EDO DRAM, Cachechk reports that L2 read is slower than DRAM read. Testing at 180 MHz: system unable to boot to IDE HDD or floppy."

I suspect that I abandoned all aspirations of using this board thereafter.

Concerning M919 BIOSes, wouldn't you just grab them from the retroweb rather than buying a EPROM chip online?
https://theretroweb.com/motherboards/s/pcchip … ver-3.3b-f#bios
https://theretroweb.com/motherboards/s/pcchip … 9-ver-3.4b#bios
https://theretroweb.com/motherboards/s/pcchip … -m919-v1.x#bios

Strange that you report your M919 v1.5 has the '64mb performance bug' with its native BIOS, but not with the v3.4's BIOS. And that your M919 v3.4 doesn't have the bug with its native BIOS. Were you using DOS Quake as the bug indicator? With 1024K L2 installed, run your v3.4 and v1.5's with 32 MB (L2 WB mode). Then run again with 64 MB (L2 WB mode). What Quake benchmark numbers you see using an Am5x86 and again with a Cx5x86?

feipoa wrote on 2024-12-15, 02:53:

There are quite a few variances which can affect operation at 66 MHz FSB, aside from the chipset, like motherboard differences i […]
Show full quote

There are quite a few variances which can affect operation at 66 MHz FSB, aside from the chipset, like motherboard differences in the "3V3" voltage output, BIOS timings, differences in hardware implementation of onboard memory buffers, PCB trace routing, and how the MB manufacturer handled parasitics, etc.

My notes on the M918 are poor, but this is what I was able to find: "Tested Am5x86 at 100 MHz. Using EDO DRAM, Cachechk reports that L2 read is slower than DRAM read. Testing at 180 MHz: system unable to boot to IDE HDD or floppy."

I suspect that I abandoned all aspirations of using this board thereafter.

Concerning M919 BIOSes, wouldn't you just grab them from the retroweb rather than buying a EPROM chip online?
https://theretroweb.com/motherboards/s/pcchip … ver-3.3b-f#bios
https://theretroweb.com/motherboards/s/pcchip … 9-ver-3.4b#bios
https://theretroweb.com/motherboards/s/pcchip … -m919-v1.x#bios

Strange that you report your M919 v1.5 has the '64mb performance bug' with its native BIOS, but not with the v3.4's BIOS. And that your M919 v3.4 doesn't have the bug with its native BIOS. Were you using DOS Quake as the bug indicator? With 1024K L2 installed, run your v3.4 and v1.5's with 32 MB (L2 WB mode). Then run again with 64 MB (L2 WB mode). What Quake benchmark numbers you see using an Am5x86 and again with a Cx5x86?

Ugh, the older I get the less I like flashing bioses, I'd just rather pop another one in.

I will have to wait for the 1024k module to show in order to try the 3.4 as I can only run it in WT with 64mb and the 256k and cache everything. There is a slight chance this board has a non original bios- it is a little battered, well used. Are you implying that this version of the board should definitely suffer from the bug?

On the v1.5 I can confirm that I was getting more fps in quake with 32mb then with 64 with its original bios (this board was NOS in a factory box with manual, genuinely pristine so definitely original). I wasn't using quake as a primary indicator though- in my experience with this 1.5 and 2 other lesser random boards with the umc chipset - I've always had cachecheck reliably report something like "looks like mb xx isn't being cached" when it hits that point, along with increased access times after that.

Your observations with the 918 are accurate. Dram is indeed faster then cache at high FSB / and or it gets auto turned off because of this UNLESS you can get it running at 3-1-1-1. At this point it is SLIGHTLY faster yet again then the fastest possible dram speed. Also the onboard IDE didnt like anything but CD roms at 60mhz+, I was using a promise ide card to circumvent this. I also had to turn off "pci to dram" buffer and a few other settings to get rid of hangs on boot and other errata.

My two 3.4b/f boards suffer from the 64 mb bug.

I think L2:WT mode should also demonstrate the bug with 256K and 64 MB.

I found Quake the most telling with the 64 mb bug.

If DRAM is faster on the M918, wouldn't it be meaningless to use L2 cache on that board? Were you able to get L2:3-1-1-1 stable to make cache worth having?

How did you get around the floppy drive not working at 60 MHz? sometimes you can use LS-120 floppy drives (connects to IDE) to work around that issue.

bertrammatrix wrote on 2024-12-14, 20:40:

Here's pictures from the auction. It is one of Pancakepuppys (from here) modules that are floating around, so there's likely be […]
Show full quote

amadeus777999 wrote on 2024-12-14, 17:34:

Looking forward to a pic once you have it. What kind of score do you get with the 919 in Doom shareware - "-timedemo demo3" with […]
Show full quote

bertrammatrix wrote on 2024-12-14, 05:17:

It is the original 256k module. I was trying to get someone from here to whip me up one of the 1mb badboys but they ended up ghosting me. I am a sucker for cache though and tracked one down on ebay, it is enroute to me now, or whenever are striking postal service gets back to work should I say.

The seller actually has one more of these listed with a m919 for a pretty decent price right now, maybe it's just what you're looking for

After much performance comparisons- an m919 will beat almost anything else even with only 256k of cache, so really it's not that big a deal

Looking forward to a pic once you have it.
What kind of score do you get with the 919 in Doom shareware - "-timedemo demo3" with HUD and one green border?
As for the cache - even 128KiB gets you a really good boost but I had 1MiB on a GA-486AM/S(50/150 & all timings tightest) and it yielded the best Doom "timedemo3 score" I had ever seen... it was only some hairs over a 1000 realtics.

Here's pictures from the auction. It is one of Pancakepuppys (from here) modules that are floating around, so there's likely better pictures in the 1mb cache threads. It's probably doubled in price by now 😁 but it is what it is. Can't wait to benchmark with it, sounds like those 12ns chips are good at fast timings

The attachment Screenshot_20241214-113625_eBay.jpg is no longer available

The attachment Screenshot_20241214-113642_eBay.jpg is no longer available

Very nice - thx.
As for the bench scores - Doom is the only demanding software I use so I'm solely interesting in its benchmark results. It's my gold standard for 486s and earlier Pentiums.
Very fast gfx cards yield unnatural peaks in Doom performance which one can circumvent somewhat by timedemo-ing Doom2's demo3("The Abandoned Mines") which features a much more steady baseline of geometry than E1M7("Computer Station"). Any demo which is happening in a level where there are tight empty corridors gets unnatural boosts in frames with a capable gfx card.

feipoa wrote on 2024-12-15, 04:56:

My two 3.4b/f boards suffer from the 64 mb bug. […]
Show full quote

My two 3.4b/f boards suffer from the 64 mb bug.

I think L2:WT mode should also demonstrate the bug with 256K and 64 MB.

I found Quake the most telling with the 64 mb bug.

If DRAM is faster on the M918, wouldn't it be meaningless to use L2 cache on that board? Were you able to get L2:3-1-1-1 stable to make cache worth having?

How did you get around the floppy drive not working at 60 MHz? sometimes you can use LS-120 floppy drives (connects to IDE) to work around that issue.

That's interesting. I'm pretty sure the 3.4 did exhibit the bug with the 1.5 bios and 256k/ wt /64mb but it does NOT with it's own bios. I may retest this now that you question it, my testing in this configuration was limited as I was attempting to "fix" the 1.5 and just wanted to confirm it's a bios issue. I will also post a picture of the 3.4 bios ID string, perhaps someone can confirm if it is the boards original bios or not.

Yes! I was able to get 3-1-1-1 windows stable on the m918, with 512k, with edo. I found the ultimate test (even more then os system install without problems) was to be the opening of Tiberian Sun that would freeze if there was any cache data reliability problem. The floppy issue you are describing is directly related to this (or was in my case) and would happen anytime the board cache settings were too fast/there was an issue. It would progress further if cache was removed/disabled/once I found a combo of ICs that the board liked. I had NO luck using any of my cheap Ebay "fake" sram on this at 60mhz, I could only get it working with genuine vintage sram, so I'd definitely classify the board as picky.

feipoa wrote on 2024-12-15, 04:56:

My two 3.4b/f boards suffer from the 64 mb bug. […]
Show full quote

My two 3.4b/f boards suffer from the 64 mb bug.

I think L2:WT mode should also demonstrate the bug with 256K and 64 MB.

I found Quake the most telling with the 64 mb bug.

If DRAM is faster on the M918, wouldn't it be meaningless to use L2 cache on that board? Were you able to get L2:3-1-1-1 stable to make cache worth having?

How did you get around the floppy drive not working at 60 MHz? sometimes you can use LS-120 floppy drives (connects to IDE) to work around that issue.

Here-

v1.5 with cyrix at 2 x 60, 512k, 32 vs 64mb edo, only basic lssr and linear burst enabled

Bios Id

32mb cachecheck

32mb quake

64 mb cachecheck

64 mb quake - a lot worse without question

Note my v1.5 board seems to have a bios dated o3/06/96 which is newer then the one available at the link you posted (02/16/96), must be late production board

The bios on my v3.4 is dated 05/06/96