Sure if you can.

If you've got it, flaunt it.

Some 386DX25 boards do have cache, some don't - and they even lack sockets for SRAM.
Both variants are period correct.

Both configurations with and without cache are period correct but if you had the coprocessor you'd likely have the cache.
There is at least one chipset with integrated cache though: I used to have a Jaguar V 386 board spoting 8kB of cache in the MACRONIX MX83C305FC/MX83C306FC chipset.

Get a system with cache else you are just having a 286/16 speed-wise system that can run 386 instructions...
I suppose if you are aiming at the lower-end of 386 speeds or a nice "286-like" setup then keep it but if you want 386 performance then cache is crucial.

keropi wrote:

Get a system with cache else you are just having a 286/16 speed-wise system that can run 386 instructions...
I suppose if you are aiming at the lower-end of 386 speeds or a nice "286-like" setup then keep it but if you want 386 performance then cache is crucial.

That's exactly what I am using my 386SX for.. Deturboed it becomes nearly a perfect 286/12, in normal operation it is around 286/20 and can execute 386 code which REALLY simplifies life for me.

My 386-25 doesn't have cache. It's nice and faster than the 286 boards I have. It has a weird proprietary slot, which I guess might be for a cache module. But I never seen one on eBay or anywhere else.

So while a cache would be really nice and would speed up some computations, it doesn't make that much difference when dealing with many games and applications.

If you can drop in 60ns RAM, you might be able to get the wait states low enough that it's possibly even faster than a cached 386-25.

Anonymous Coward wrote:

If you can drop in 60ns RAM, you might be able to get the wait states low enough that it's possibly even faster than a cached 386-25.

I'd like to see some numbers confirming this, have you actually measured it?

keropi wrote:

Get a system with cache else you are just having a 286/16 speed-wise system that can run 386 instructions...l.

No it's not wrt a 386DX25 compared to a 286/16. 386SX of the same speed maybe.

Caluser2000 wrote:

keropi wrote:

Get a system with cache else you are just having a 286/16 speed-wise system that can run 386 instructions...l.

No it's not wrt a 386DX25 compared to a 286/16. 386SX of the same speed maybe.

you will be surprised if you bench a cacheless 386DX/25 system 🤣
keep in mind that a 386DX/25 is a pretty old 386DX system and a 286/16 is a newer one so the chipset can also make a difference

Best reason to go for cache: if you want the performance, you have it. If not disable it whenever you want. No cache and you don't have that option.

keropi wrote:

Caluser2000 wrote:

keropi wrote:

Get a system with cache else you are just having a 286/16 speed-wise system that can run 386 instructions...l.

No it's not wrt a 386DX25 compared to a 286/16. 386SX of the same speed maybe.

you will be surprised if you bench a cacheless 386DX/25 system 🤣
keep in mind that a 386DX/25 is a pretty old 386DX system and a 286/16 is a newer one so the chipset can also make a difference

Cacheless SX25 and DX25 perform basically identically in all my testing, at least in DOS. They both perform about the same speed as a 286/20.

^ yes that's what I found out as well - maybe it had +1 or 1.5fps more from a 286/16 in wolfenstein3D benchmark with the same VGA card (both terrible scores anyways 🤣 )
I need to mention I only tested under DOS as well

keropi wrote:

^ yes that's what I found out as well - maybe it had +1 or 1.5fps more from a 286/16 in wolfenstein3D benchmark with the same VGA card (both terrible scores anyways 🤣 )
I need to mention I only tested under DOS as well

Well, Wolf3D is a 1992 game so 286 class hardware struggling with it is hardly surprising, is it 😁

I use my 386SX to play Sierra SGI/AGI, early Lucasfilm, Might & Magic, Ultima, Wing Commander and SSI AD&D games. It's essentially my MT-32 PC. (It has keropi's wonderful Music Quest clone card in it.. Thanks a ton again mate! 😀 )

Heheh
Sure wolf3d is a 1992 game too new for 286 but in this case it shows how older cacheless 386dx boards really perform.
Anyways, it is better to have cache and disable it rather than no option at all as it's been said earlier in the thread.

Should a period correct 386DX25 have an external cache?

Period-correctness is overrated. Period. 😉

appiah4 wrote:

Cacheless SX25 and DX25 perform basically identically in all my testing, at least in DOS. They both perform about the same speed as a 286/20.

Doesn't make much sense to me. 386SX is 16 bit externally with much lower RAM speed due to this. This is where missing cache is especially noticeable. 386DX is 32 bit externally and can access RAM (and other things) much faster. So unless you tested 16-bit stuff only that also does not need much memory bandwidth, there's something wrong with the tests you did.

A 386SX performs like a 286 at the same clock speed, with the 286 even being faster sometimes with 16-bit software. A 386DX with external cache performs like a 486 at about 2/3 of the clock speed.

I noted "at least in DOS", so yeah I was referring mostly to 16-bit code. Are there any 32-bit games or applications that you would run on a 386SX/DX anyway?

My experience and testing is that clock for clock the 286 is actually faster than the 386 for 16-bit, by the way - they are not equally fast at equal speeds.. I don't have a single clue why.