I found this, but it's not an exact match. My cpu is the P54CQS and mobo has the VX chipset

https://www.ebay.ca/itm/284225471250?ul_noapp=true

Does that matter?

Chipsets are not the issue in most cases. At the time there were VX, FX and HX mainboards.
But there´s always a small risk of incompatibilities when you buy a cache stick today, so be prepared to have a second or even third try.
But chances are high to have a hit in your case when the cache stick has 256K of SRAM (2 chips x 128K) and when it comes with it´s own TAG-RAM.
So the HP stick "should" work (although HP are in the habit of making proprietary solutions sometimes).
It also depends on the mainboard. Some models accept and correctly detect nearly everything (like the ASUS P/I-P55TP4XE), some are picky.
The Shuttle 557 should be quite versatile, it´s also providing many core voltages for different CPUs. The latest BIOS wouldn´t hurt though.

Looks like you might need one without additional TAG - this is a rough translation from a Chinese FAQ page on the old Spacewalker site.

Seems to refer to it as "Haoxin Quick Access Memory Module with only two 128KB chips"

https://web.archive.org/web/19991008101013/ht … _c/CACHEMOD.HTM

Ouch! These are hard to find.
Another proprietary abominaion.
They should have mentioned this in the manual.

Whoa! I'm simultaneously impressed with your level of research in being able to zero in on the correct item I need and also dismayed at how the heck I'm supposed to find it.

FYI, I checked my board and it has a um61256ak chip right next to the jumpers, which appears to be a tag module?

Also what's with the cache below the slot on the motherboard? I guess I'm misinterpreting that because it looks like "onboard" cache but I thought that's what the coast stick is for?

Also the top picture says make sure it only has "2 chips" and the bottom one says "3 chips" but visually I count the same number of chips on both, unless there's a big square one on the other side on the bottom instance.

Here's another candidate that kinda looks like it might be similar:

https://www.ebay.com/itm/L2-cache-module-256K … cUAAOSwAexgJALV

Gahhhrrrlic wrote on 2021-04-10, 20:46:

Whoa! I'm simultaneously impressed with your level of research in being able to zero in on the correct item I need and also dis […]
Show full quote

Whoa! I'm simultaneously impressed with your level of research in being able to zero in on the correct item I need and also dismayed at how the heck I'm supposed to find it.

FYI, I checked my board and it has a um61256ak chip right next to the jumpers, which appears to be a tag module?

Also what's with the cache below the slot on the motherboard? I guess I'm misinterpreting that because it looks like "onboard" cache but I thought that's what the coast stick is for?

Also the top picture says make sure it only has "2 chips" and the bottom one says "3 chips" but visually I count the same number of chips on both, unless there's a big square one on the other side on the bottom instance.

Here's another candidate that kinda looks like it might be similar:

https://www.ebay.com/itm/L2-cache-module-256K … cUAAOSwAexgJALV

Think the chip count they mean is cache + tag, so either 2 + 0 or 2 + 1 (which seems to be only for the HOT-541)

Here's a better pic of your board with the proper COASt installed

http://www.amoretro.de/2011/03/shuttle-spacew … otherboard.html

You can also see the same type of COASt design (if not spec) on the Wikipedia page

https://en.wikipedia.org/wiki/Cache_on_a_stick

Indeed there is no tag chip.. I see what you mean. All the ones I've seen for sale so far have one or look completely different. Just my luck these manufacturers had to be "unique".

Even if you don´t find a fitting cache module, you still have 256K onboard cache. According to the manual the maximum RAM for the HOT-557 is 128MB.
256K (onboard) cache can cache 32MB of RAM in write back mode and 64MB in write through mode.
Depending on actual RAM size on your board the additional 256K might not make a big difference anyways (or even no difference if you are using 32MB or 64MB of RAM).

The cache stick in the AmoRetro picture seems to be a regular 256K stick with soldering pads for it´s own TAG-RAM but it´s just not populated. Maybe taking a regular stick and removing the TAG chip would do the trick but I´m not sure if it´s worth the effort.
A total of 512K cache (the onboard TAG-RAM has 32Kx8 and is sufficient for 512K) could cache 64MB in write back and 128 MB in write through mode.
However, the original "Shuttle" cache stick will be really hard to find.

Wait... You're saying I have 256kb L2 already???

Benchmarks show negligible improvement from 256 to 512 so what's the point of the slot then?

512KB are making sense, if you have more than 32MB of RAM installed.
The 256KB onboard cache provides a cachable area of 32MB, since the cache policy is "write back" (according to the manual).
So if you have just installed 32MB or less for a DOS / WIN 95/98 machine, you can be happy with the existing 256KB of cache. You wouldn´t even notice the additional cache.
And, as long as you are operating with 256K you schould avoid populating more than 32MB RAM. If you do, performance can deteriorate, because some operating systems are using the upper (uncached) range of RAM first while the lower range is cached and faster, but is not being used.
Btw. this issue also occurs, when you have 512KB of cache and max out the RAM to 128MB. The cachable area will be 64MB only, performance will suffer.
On the other way to install more than 512KB of cache would be pointless here, because Intel´s VX chipset can not cache more than 64MB by design!

As it happens I have 64 mb exactly installed. Maybe I should remove a dimm.

BTW, I am just about to remove 1 of the memory modules to test the whole "256 can only address 32mb of ram" theory. Is there a Win95 utility anybody knows of which would prove as to whether or not reducing memory to 32 mb forces the remaining 32 to be used in write-back mode? Essentially is there a test I can do to show the speed increase?

Testing completed. Results are not what I thought they would be:
430VX, 133MHz, 256kb onboard L2, ATI Mach64 GT 2B 2MB

32MB RAM:

DOS ONLY:
CACHECHK: L1 cache = 183.7 mb/s, L2 cache = 117.6 mb/s, memory = 77.0 mb/s
SPEEDSYS: L1 cache = 253.90 mb/s, L2 cache = 106.84 mb/s, memory = 75.23 mb/s, bandwidth = 222.07mb/s
3dbench 2.0 = 106.1 fps

WIN95:
QUAKE DEMO = 26.8 FPS
QUAKE 2 DEMO = 11.7 FPS

64MB RAM:

DOS ONLY:
CACHECHK: L1 cache = 183.7 mb/s, L2 cache = 117.6 mb/s, memory = 76.3 mb/s (slightly slower memory)
SPEEDSYS: L1 cache = 253.88 mb/s, L2 cache = 106.84 mb/s, memory = 74.99 mb/s, bandwidth = 221.83mb/s (slightly slower memory)
3dbench 2.0 = 106.1 fps (no change)

WIN95:
QUAKE DEMO = 26.9 FPS (negligibly faster)
QUAKE 2 DEMO = 13 FPS (significantly faster)

So basically everything is almost within measurement error except for quake 2 which was actually faster with 64MB and I'm going to assume this is because after w95 takes its cut, q2 actually runs out of ram to cache stuff and therefore there's a bit of lag with 32MB while it tries to catch up during the timedemo. Just a theory. In any case, maybe I didn't do the right tests but it looks like adding ram only makes the system faster, not slower despite 256kb of L2 not being able to address the entire RAM. Maybe it's addressing the first 32MB of it after all. I just don't see a performance detriment to having 64 and only see a benefit with more demanding games.

Question: Is there a way to toggle write-back vs write-through to see if it's even on the correct setting now? Maybe it was never set to write-back and that's why my testing didn't show improvement with the smaller ram?

First, a test with onboard cache disabled would be useful to see / proove which / if tests profit from cached RAM.

If you run applications or tests that consume a lot of /all RAM, note that uncached RAM is still faster than using the harddrive´s swapfile so additional RAM above the cached area IS good for performance in many cases. BUT if you have let´s say 256MB of RAM that is never used completely by the OS and applications, then it´s possible only uncached RAM is used while the cached area remains unused. Performance suffers compared to a situation with less RAM.

The final limit for the cached area of 64 MB is fixed by the VX chipset´s limitations. BUT if you have 128MB installed and applications really need it, the system is faster than with 64MB, since uncached ram is still faster than swap.
I think we can already see here that a bigger cache of 512KB will only make sense under certain conditions. The rule of thumb back in the days was: "The first 256KB of cache make most of the difference."

Further the actual cachable area for a given cache size depends on the chipset (FX, HX, VX), cache policy and the organisation of the TAG-RAM. It´s possible that in your case 256KB cache DO cache 64MB of RAM completely, while you would need 512KB for RAM sizes of 64 - 128 MB only - which would be useless due to the VX chipset limitations.

A lot of parameters and situations have to be considered to decide the optimal size of cache and RAM.

Yeah that's why I figured that Q2 was ram limited but also included Q1 because there's no way in hell that would use the rest of the first 32mb. Windows uses some but I had nothing running at the time. Just a fresh boot and no bloatware in the tray. It's highly likely that windows and quake operated within that first 32mb so whatever effect the additional stick of ram had, it appears not to have hindered the system. Again, that may just mean that nothing changed with between the 2 cases, when it actually should have, but that's why I'm asking if there's a way to get feedback on how the ram is being used. If there was a way I could see whether write-back is enabled for example or if I could see how much memory is being addressed by the cache, that would help. No doubt disabling the cache would show a hit to performance but that doesn't tell the whole story if there's no clear correlation to the RAM space.

I´m not aware of any tool reporting L2 cache policy right now. Maybe someone has a suggestion.
Documentation about all the details concerning these old chipsets, setups, BIOSs, policies, bus-widths etc. have become quite rare today. I used to store an in-depth analyis about the whole issue somewhere, but can´t find it.

If you can´t find significant performance differences between 64MB and 32MB I would *assume* that the cachable area in your setup is 64MB.
Additional cache would be absolutely obsolete in this case due to the VX limitations.

If the mainboard had the HX chipset for example this would be a completely different matter.

PC Hoarder Patrol wrote on 2021-04-10, 08:41:

Looks like you might need one without additional TAG - this is a rough translation from a Chinese FAQ page on the old Spacewalke […]
Show full quote

Looks like you might need one without additional TAG - this is a rough translation from a Chinese FAQ page on the old Spacewalker site.

Seems to refer to it as "Haoxin Quick Access Memory Module with only two 128KB chips"

https://web.archive.org/web/19991008101013/ht … _c/CACHEMOD.HTM

Shuttle CACHEMOD.jpg

Thanks Hoarder! That FAQ sheet is really helpful! I'll be on the lookout for one of these.
Hmm, wonder if you could just de-solder the TAG chip? Probably not...