turbooo wrote on 2022-01-25, 06:40:

I bought one of your 256k sticks at the beginning of the month (should have checked this thread first, whoops) and it’s working great so far. A 1MB stick does sound tempting though…

What kind of performance increase do you see in Windows with the bump in cache size?

I haven't done much performance comparison between cache sizes in Windows, just DOS so far (and some stability testing in Windows.) Any suggestions of things you want me to test? I'm usually using Win9x on much faster systems so I'm coming up blank on what to run on '95 on a 486.

pancakepuppy wrote on 2022-01-25, 06:46:

I haven't done much performance comparison between cache sizes in Windows, just DOS so far (and some stability testing in Windows.) Any suggestions of things you want me to test? I'm usually using Win9x on much faster systems so I'm coming up blank on what to run on '95 on a 486.

Hmm good question, I guess there isn’t such a thing as Phil’s Windows benchmark pack for easy data collection. I recall reading somewhere on the forum that the cache boost would be more beneficial in Windows than in DOS so I was curious if that actually panned out.

Maybe doing some winzip compression.
Another vogons thread suggest Winstone and Winbench 96 availiable here :
https://archive.org/details/PCMagBenchmark
Also i've got this old CPUmark 99 version 1.0 that made 725 points on my R5 3600 @ 4.3ghz.
Should work on W95
http://crazytiti.free.fr/divers/CPUmark.exe

There is a CPU-Z version for Win9x and it has included benchmark.

Thanks for the suggestions! I'll see what kind of results I can get out of these 😀

256KB

1MB

CPU-Z doesn't seem to care

256KB

1MB

CPUmark 99 increases 10%

256KB

1MB

Overall Winstone 96 up about 5.5%

https://github.com/pancakepuppy/M919_3.3V_1MB_cachestick
Github for the 1MB 3.3V design. If you want to have your own boards made, change the silkscreen art on it, maybe have it built in a fun color, go nuts! I have 19 more PCBs and will be working on soldering them up. Those who have messaged me, I will get in touch with you when I have one ready for you 😀

Looked at your schematics, it's really good that A18 and A19 lines are present in the slot. Makes things much easier.

Is there any reason for choosing 3.3V SRAM chips other than general availability? I mean UM8881F is a 5V chip and probably expecting 5V signal levels. Should be fine this way at regular speeds like 33MHz and 40MHz, but I'm talking about 60MHz and 66MHz.

Another thing, 3.3V is generated by a single LDO that feeds the CPU and the board is really short on capacitors, ceramic ones especially. I don't think voltage stability is great on this rail as it is, and we're putting additional load. Could be a good idea to place a 3.3V LDO right on the cache stick PCB. Something like LM1117 or TLV1117 in SOT-223 as it doesn't take much space.

ph4nt0m wrote on 2022-01-27, 14:05:

Looked at your schematics, it's really good that A18 and A19 lines are present in the slot. Makes things much easier.

Is there any reason for choosing 3.3V SRAM chips other than general availability? I mean UM8881F is a 5V chip and probably expecting 5V signal levels. Should be fine this way at regular speeds like 33MHz and 40MHz, but I'm talking about 60MHz and 66MHz.

Another thing, 3.3V is generated by a single LDO that feeds the CPU and the board is really short on capacitors, ceramic ones especially. I don't think voltage stability is great on this rail as it is, and we're putting additional load. Could be a good idea to place a 3.3V LDO right on the cache stick PCB. Something like LM1117 or TLV1117 in SOT-223 as it doesn't take much space.

I went with parts I could get easily and didn't much care for having it work with 5V CPUs. Figured if the power situation was good enough for the original 3.3V cache sticks, should be fine for mine with more capacitance close to the SRAM power pins and more modern chips using a pinout optimized for power integrity. Operating above the spec 486 bus speeds wasn't a design criteria for me but if you want to make a revised PCB and leverage on-board regulation, go for it I say! I was aiming more at a "production optimized" design with as few parts as possible.

Sure thing it will do the job without much overclocking. Just checked, those IDT71V124SA12Y chips you put in are rated at 130mA each. That's 1.2A additional current on that LDO regulator. It doesn't seem real tough, could be an issue.

ph4nt0m wrote on 2022-01-27, 17:11:

Sure thing it will do the job without much overclocking. Just checked, those IDT71V124SA12Y chips you put in are rated at 130mA each. That's 1.2A additional current on that LDO regulator. It doesn't seem real tough, could be an issue.

130mA operating at maximum frequency, sure. May be an issue in your cases of high bus speeds but for normal speeds, I'm not concerned with the chipset trying to access them nearly that fast/often. Would be more worried if the 3.3V pass transistors got more than slightly warm or if I had stability problems.

Edit: If you're looking for things to optimize, the address/data routing is less than optimal. Some rearranging of the data and address lines could probably make it a bit cleaner. I look forward to seeing your version! Have fun 😀

Hello, any news about making the cache sticks ?

crazytiti wrote on 2022-03-22, 23:08:

Hello, any news about making the cache sticks ?

Been working on getting them sent out to interested parties, I'll message you before too long 😀

I ran some simple analysis on the M919 v3.4 and the 1024K module. These modules save the board. It looks like there is some performance bug around the 64 MB mark, which was limiting us to 32 MB of RAM. It looks like we now can jump up to 128 MB of RAM or 256 MB of RAM, not suffer from the "64MB bug", and still have all L2 cached. Table and analysis here: Re: 3 (+3 more) retro battle stations

EDIT: This 64 MB bug is found on the 10/16/1996 and 09/15/1998 BIOSes, but not the 5/6/1996 BIOS.

Hey,
I've got the 1Mb stick from Pancake puppy.
Very nice stick, well assembled.
I had to switch the memory from "normal" to "EDO" to get it booting, otherwise it would hang after the table.
It's working fine.
Putting 14 fps in quake bench with AMD 486 dx4 120
and 15.2 fps with intel dx4 100 @120 !
All memory timings at the lowest.
I'm wondering if i have one of the fastest 486 ?

I'm interested in at least 3 of the 1M sticks for my M919 boards... any up for sale, let me know surely. Thx. My eyes are too crappy to build these like in the old days:( so looking to but pre-built.

I've noticed my short distance starting to worsen in the last 2 years. I find reading glasses can help when working on electronics. Makes you feel old though.

pancakepuppy wrote on 2022-03-22, 23:29:

crazytiti wrote on 2022-03-22, 23:08:

Hello, any news about making the cache sticks ?

Been working on getting them sent out to interested parties, I'll message you before too long 😀

Know anyone that can design a 72 pin simm? I'm looking to make some 50ns fpm sticks and just cant find any reference kicad files whatsoever 🤣.

Is it still possible to obtain a 1mb cache board? I just barely repaired my m919 3.4B/F board.
Thanks for working to build this awesome cache module!

pancakepuppy wrote on 2022-03-22, 23:29:

crazytiti wrote on 2022-03-22, 23:08:

Hello, any news about making the cache sticks ?

Been working on getting them sent out to interested parties, I'll message you before too long 😀

Hello Pancakepuppy,
do you have any left in stock?

rtstaley wrote on 2022-11-19, 21:18:

pancakepuppy wrote on 2022-03-22, 23:29:

crazytiti wrote on 2022-03-22, 23:08:

Hello, any news about making the cache sticks ?

Been working on getting them sent out to interested parties, I'll message you before too long 😀

Hello Pancakepuppy,
do you have any left in stock?

Nope, sorry. All out for now.