There are some 286 boards that have exceptional performance, while others are lackluster. I found all 286 boards based on the Headland HT12 chipset to be very fast (for a 286 that is). Opposite to that i have an M219 board which is horribly slow. It is slower at 20MHz than my Headland HT12 board at 16Mhz.

Intresenting, my NMOS 287-10Mhz got 751 KWhetstones on NSSI FPU Bench 😀
So the 287XL isnt really much faster if software isnt optimized to use it 😳

Try also Checkit and Comptest as NSSI sometimes reports strange results.

There is a chance that it is actually a real score. The 287XL is very accurate IEEE -wise and a good overclocker, very compatible. But speed demon ? No.
That is the territory of 387 math coprocessors retrofitted to the 287 socket, especially the Cyrix 82S87 chips manufactured after 1991.

matze79:
what is your CPU frequency? FPU might run faster than what it says on the chip. Also might be interesting that NSSI reports 563K Whetstones when 287 isn't even installed - using internal emulator 😲 While Checkit reports more normal 84K Whet and Comptest also low 64.2K Whet.

Jolaes76:
all later 287 FPU's from Intel, IIT, Cyrix used 387 core and would run much faster than original Intel 287.

---

I've been doing some FPU testing on this board and would like to get more performance out of it. How exactly is FPU frequency derived? I've read there is a pin 39 that either divides clock by 2 if high or by 3 if low (grounded).

Do the usual sysinfo programs like NSSI, Comptest correctly report the FPU frequency?

My testing with Comptest show that they run (crystal clock is 48MHz):

AMD 287 at 9.32MHz
Intel 287XL+ at 12.60MHz
IIT 2x87 at 10.81MHz
Cyrix 287XL+ at 14.57MHz

This is way slower like my other 286 board that is overall worse than this one, but FPU is really fast (crystal clock is 40MHz).

AMD 287 at 15.21MHz
Intel 287XL+ at 21.14MHz
IIT 2x87 at 18.39MHz
Cyrix 287XL+ at 24.38MHz

Later I'll attach some photos of FPU testing: FBENCH, CABT, ACAD

ACT 286-24MHz + Cyrix 287XL+

HT18/c 286-20MHz + Cyrix 287XL+

This HT18/c chipset is actually for 386SX and has an option for Fast I/O. With this enabled ISA bus is super fast - look at BUS and VIDEO speed of Dr.Hard. Too bad it doesn't have a socketed oscillator so I could try with 50MHz one. Also the FPU is amazingly fast. With heavy FPU usage almost 100% faster than the ACT motherboard.

I also tested with Autocad 10. Even though it requires a FPU I guess it doesn't use it all that heavy. Results are in favour with a faster CPU when doing HIDE operation. Regen is more FPU intensive.

ACT 286-24MHz + Cyrix 287XL+

Regen: 7.66 sec
Hide: 7m 18sec

HT18/c 286-20MHz + Cyrix 287XL+

Regen: 5.86 sec
Hide: 8m 00sec

I really can't tell why the HT18/c motherboard has all that faster FPU. Sysinfo programs report very high fpu frequency but why is this so, I don't know. I measured the Pin39 and on both it's grounded. So internally crystal clock should be divided by 3.

>With 50MHz crystal it freezes in DOS. I'll try to get a real Harris 25MHz one day 😉

More voltage and add cooling.

awgamer wrote:

>With 50MHz crystal it freezes in DOS. I'll try to get a real Harris 25MHz one day 😉

More voltage and add cooling.

I did get a 25MHz Harris but it still freezes in DOS. I guess the motherboard can't keep up.

kixs wrote:

I did get a 25MHz Harris but it still freezes in DOS. I guess the motherboard can't keep up.

probably just a problem in RAM-chips?

1/80ns=12,5mhz
1/70ns=14mhz
1/60ns=16,6mhz

I had to find a selective memory (siemens 60ns) to make it work at 25 MHz

I'm already using 60ns. I also have other 60ns ones. Maybe it's worth a try.

Checked installed SIMMs. They are two different pairs of 60ns, three chips. Tried only with one pair (2x1MB) resulted in much stable system but still crashed in a minute. Tried with the other pair and this one is much more stable but still crashed after running benchmarks - Wolf3D. Now I put a 12cm 2500RPM fan close to the motherboard. Will see how it goes 😀

It passed all the benchmarks. So far so good. Will also try with other 60ns SIMMs, to get it stable without a fan and working with 4MB.

Best case results at 25MHz + Cyrix 287XL+:

3DBench 1.0: 10.2
Checkit3:
Dry: 6254
Whet: 716.5
Video BIOS: 9158
Video Direct: 78186
HDD: 2433.5
PMIPS: 4.0
Sysinfo: 19.8
NSSI :
CPU: 6180
FPU: 922
Dr.HARD:
CPU: 5522
FPU: 662
RAM MAX: 9.79
BUS: 5.12
VGA: 5.2
CCT: 1003
3DSpace: 14%
Wolf3D: 16.5
F1bench1: 102%
ACAD10
Regen: 7.39
Hide: 6m 59s 47
CABT: 7.31
Speed600:
CPU: 39.25
FPU: 25.11
Video: 5313.73

kixs wrote:

awgamer wrote:

>With 50MHz crystal it freezes in DOS. I'll try to get a real Harris 25MHz one day 😉

More voltage and add cooling.

I did get a 25MHz Harris but it still freezes in DOS. I guess the motherboard can't keep up.

You'll have to hack it yourself but more voltage.

Congratulation! 😀 your motherboard is super fast 😀

kixs wrote:

Checkit3:
HDD: 2433.5

It is CompactFlash ?

Yours is still faster 😉

Yes, I use CF cards exclusively.

Did more tests with memory. 80ns won't even boot. Some 70ns too. Now I found some IBM 70ns that work as good as better 60ns. Most 60ns won't work reliably.

48MHz isn't a problem while 50MHz gets errors within a few minutes. Tested 48MHz for around 15min with Wolf3D demo and memory gets hotter than the CPU itself - too hot to touch. I guess I'll have to mount some thin heat sinks. For now I'll put 48MHz crystal back.

Just didn't realised that memory would get so hot 😲

Any ideas what heat sinks to use for regular 30-pin three chip SIMMs? There is maybe around 4mm clearance between SIMMs.

You could maybe get 1mm thick copper sheet and use some thermal glue to attach them to the chips? With the right tools you could maybe fold it above the module itself to furhter increae the contact area.

I think comparing this 25MHz 286 to a 25MHz 386 is slightly unfair in that you are running with 0ws memory. To the best of my knowledge the fastest 0ws 286 that was sold new was 16MHz. I think for a better comparison you should turn off the L2 cache on the 386 and try to run with 0ws memory on that system as well.

I guess you missed the thread... as I don't make any comparison to the 386. Only between different 286 boards.

Sorry, I got this 286 thread confused with the other one.

please, run the ATPERF from archive BENCH286.RAR on your 286?