JohnBourno wrote on 2022-03-22, 12:58:

What is interesting, is that without a FPU Duke Nukem 3d sometimes gets slowed down significantly. It drops from 20 frames per second to 7 fps in some scenes. Turns out that Duke3D uses floating point operations for rendering slopes. So without an FPU every time you have some slopes in your viewports the CPU has to do all the extra work and the game gets almost unplayable. Just by adding a IIS FPU the fps only drop to 15 instead of 7. So quite playable.

90MHz NexGen Nx586 drops down to 10fps on first screen with sloped roof for that very reason:
https://youtu.be/41O2bNG2qKA?t=234

Im sure today someone could come up with a faster integer/lookup table method of calculating those slopes.

Next time I find/boot my U5S CPU, I should try it out on DN3D with one of the 387 emulators see if there's a speed boost.

So... here is an update.
My 387DX FPU collection has increased by a lot.
All overclocked to the max (up to 55MHz). Enjoy:
https://www.youtube.com/watch?v=Zvisa_-uEqI

On the 4x4 instruction.. you load the 4x4 transformation matrix into alternate register sets, then leave them resident while you load the fairly small 4x1 vector. Typically the transformation matrix is reused many times, so this saves a huge amount of communication overhead. For the time, a very big win. You also presumably (I haven't tried it) get a pretty good chunk of time that the FPU is busy and you might be able to do something useful with the integer unit. Easier than trying to work in a couple int instructions in between normal FPU instructions.

Often you can get away with a 3x3 matrix for 3d graphics, which would be a bit faster, but they didn't support that.

A downside.. Operating systems don't know about it, so they don't save the extra register sets. It didn't matter much at the time.

Deunan wrote on 2021-04-27, 08:44:

Personally I see one big problem with the 4x4 operation - it needs all the data to be fed to FPU and then the results be read back. AFAIR you need to start with empty stack, use IIT-specific stack extension instructions, and some of the input arguments are over-written to store the result. This, coupled with the rather slow CPU-NPU comm channel, limits the usefulness of such instructions. Weitek worked around that by having their NPU register space memory-mapped, at the cost of even lower compatibility with typical x87 code.

Long story short: It took MMX to finaly have some direct access to FPU register space, and even that was flawed due to cost of switching between MMX and x87 modes. SSE finally made FPU on x86 family somewhat saner by today's standards.

I’m going to be building a 386 er ‘486’ system soon. And using the texas instruments sxl2 66 on the new interposer.

And was wondering what the best fpu to pair with it would be.

This will be the slowest system that I have ever built. So it will be kinda fun new territory for me.

I’ll have to post the exact mobo model later as I can’t seem to remember it, but I think it is a shuttle hot with 16 isa and 8 bit isa. I seem to recall it having a AMD-40 soldered on and a witek socket as well as a cpu upgrade socket

For up to 40-45MHz base frequency that will be a combination of Weitek Abacus 3167 + Cyrix FasMath (gray top) or ULSI DLC.
Where ULSI is faster in Quake 1 only and Fasmath at everything else.
At 50mhz base you will most likely be going with Cyrix Fasmath (black top) or ULSI DLC. No idea is 3167 can manage that speed.

The sxl2-66 cpu might be operating at as much as 80 mhz

Looks like specific units can hit 90.
At that frequency they can start encroaching on BL3@100.

Sphere478 wrote on 2022-12-15, 04:08:

I’m going to be building a 386 er ‘486’ system soon. And using the texas instruments sxl2 66 on the new interposer. […]
Show full quote

I’m going to be building a 386 er ‘486’ system soon. And using the texas instruments sxl2 66 on the new interposer.

And was wondering what the best fpu to pair with it would be.

This will be the slowest system that I have ever built. So it will be kinda fun new territory for me.

I’ll have to post the exact mobo model later as I can’t seem to remember it, but I think it is a shuttle hot with 16 isa and 8 bit isa. I seem to recall it having a AMD-40 soldered on and a witek socket as well as a cpu upgrade socket

For x2 CPU I'd use the x2 FPU - like ULSI 66 or IIT 50.

Dont know if the (very rare) 2x models can tick reliably at 80, 90 or 100 mhz.
Also, IIT 387 FPUs are actually very slow.

sphere, a run of the mill black-top Cyrix FasMath at 40 Mhz is your best bet for an SXL2 at 80 Mhz. There is only benefit in the ULSI DX2-66 if you are running your SXL2 at 66 MHz. I doubt the ULSI DX2 can do 80 MHz.

Here is the intended motherboard

Nice forex.

So the ulsi dx2 is faster but can’t do 80mhz

So a 40mhz black top is the fastest option for a overclocked sxl2

but a stock speed sxl2 the fastest would be ulsi dx2?

What would be the fastest overall combo?

Ulsi and sxl2 at 75 mhz?

I’m going for maximum upgrade here.

I think it best to forget about the ULSI DX2-66. This FPU takes rare to a new level. Even if the ULSI DX2-66 ran at 75 MHz, which is doubtful, an SXL2 system at 80 MHz coupled with a Cyrix FasMath at 40 Mhz will be faster.

And it will for sure work in that mobo? It says witek?

Got one of these correct one?

It can use a Weitek or a regular 387 FPU, but the Weitek only works for very select applications.

Sphere478 wrote on 2022-12-17, 18:03:

Got one of these correct one?

Check again my last video.
Assuming testing at the same clock, the fastest FPU is the Cyrix -KN.
No question about that. This one is a bit rare though.

Now benches are nice but in practice a 387 FPU are a bit pointless these days.
Games that need one, just run to slowly with a system of this class.
CAD programs are just too old and pointless.

So in the end a 5% difference in FPU performance, you will not really notice.
And the Cyrix you got is also a good one.

The "KN" is the older design with the grey top, right? Don't those have compatibility issues with Cyrix's own SLC/DLC CPUs?
How about a CX83D87-40HP? Those are also very rare. I assume the only difference over the standard black top is just the packaging...but has anyone every benched one?