@feipoa:

here is the landmark results on this mobo:

Most 386/40 boards I have, on the landmark speedtest v.6 score about 67-69MHz AT speed. I see that you're running v.2 so maybe that makes it different. If it isn't too hard for you to find v.6, it would be interesting to see the results there.

You're AMD 386-40 performs just as expected for an AMD 386. My motherboard with an AMD 386-40 scored a 61 in Landmark (version 2). As from the other thread,

Cyrix 486DLC L1 disabled ---> L1 enabled

Landmark
ALU, 86 ---> 129 (50% improvement)
FPU, 174 --->185 (6% improvement)

Speedsys
9.64 ---> 10.52 (9% improvement)

3Dbench
15.6 ---> 17.5 (12% improvement)

AMD 386DX-40 ---> Cyrix 486DLC-40 w/1KB L1 enabled

Landmark
ALU, 61 ---> 129 (111% improvement)
FPU, 140 ---> 185 (32% improvement)

Speedsys
6.84 ---> 10.52 (54% improvement)

3Dbench
13.5 ---> 17.5 (30% improvement)

The FPU was not changed to a Cyrix 487DLC-40 ?
This would be important in this combination. Also refer to the 387 FPU guide for details.

I'm still using a Cx83D87-40GP. What benefit will a Cx487DLC-40GP give? I thought this was just a renaming/marketing gimmick. Could you link the 387 FPU guide that you are refering to? Thanks.

Ok I uploaded here some files:
ftp://78.46.141.148/docs/cyrix/cyrix.txt
this one has some details for the special combination of Cyrix CPU/FPU.
I'd like to note that the 486DLC is in fact a 486 class CPU and was available as late upgrade option for 386 system. It does not reach a standard 486 performance, is often detected as 486SX + FPU and as far as I remember has some compatibility issues.

might be also interesting:
ftp://78.46.141.148/docs/FPU/coproc.txt

I'am currently filling a bit this ftp, so maybe you find other interesting stuff.

elianda, would you like to store these text files and other odds-and-ends on VogonsDrivers?

We could set up a non-driver-files section or something for ya. 😀

@elianda
Thank you for sharing those documents, however I failed to see the requirement for a Cx487DLC-40 FPU when using a Cyrix 486DLC ALU. That article mentions that a Cx-87D87-40 is the FPU of choice for a 486DLC FPU. I have bolded the section of interest. It was suggested to me that the Cyrix 487DLC was a rename of the Cx-87D87 (Cyrix 387) for marketing reasons. If you have other information, please let me know.

With the 486SLC/DLC, one buys a 387 compatible coprocessor to add floating-point capabilities. It is recommended to get a Cyrix coprocessor for this purpose, since these are the fastest 387 compatible coprocessors available. Also, Cyrix sells kits consisting of a 486SLC/DLC and a coprocessor that have a favourable value for money ratio. The floating-point performance of a Cyrix 486DLC + Cyrix 83D87 combination is about 50% of that of an Intel 486DX running at the same frequency.

Here is another neat little blip from that article,

The Cyrix 486SLC/DLC have a RISC-like execution unit with a flexible five stage pipeline, just as the 80486SX has. Unlike the Intel 80486, which has an 8 kB, 4-way associative cache on chip, the Cx486SLC/DLC only have an 1 kB, 2-way associative cache (the cache on the Cyrix chips can also be configured to be of the direct mapped type). The 486DLC provides up to 80% more integer performance than a 386DX at the same clock frequency, with the average performance gain being about 35%. With the 1 kB on-chip cache enabled, the 486DLC provides about 75% of the integer performance of a 486SX at the same clock frequency. With the cache disabled, the 486DLC provides about 65% of the integer performance of a 486SX. The lower performance of the Cyrix 486DLC as compared to the Intel 80486SX is mostly due to the slow 386DX bus interface the 486DLC uses, which is up to 2 times slower than the 486 bus interface. Some additional performance penalty is imposed by the smaller cache on the 486SCL/DLC, which provides significantly lower hit rates than the 8 kB cache of the 80486SX.

The 80% more integer performance compared to the 386DX-40 seems to be well in line with the benchmarks I ran (shown above), which had an average integer performance benefit of 65%. Those are only synthetic benchmarks, so perhaps in terms of real life applications, the average benefit is only 35%. I'd like to see these results from someone with a Ti486SXL, which has 8 KB of L1 cache. Unfortunately, my system didn't show any benefit when using a Ti486SXL w/8 KB of cache compared to the 1 KB of 486DLC, which may be a limitation of my motherboard.

However, Vogons member Anonymous Coward has a Cyrix/Ti 486SXL setup. Attached was his Speedsys result, which is double that of my Ti 486DLC. I wonder which FPU he had installed? None?

Well this is interesting, look at what score I get in Speedsys when I remove the NPU. My 486DLC Speedsys score jumped from a 10.53 to a 20.59. I suspect that Ti486SXL Speedsys result is without an NPU as well (as is stateed on the second-from-top line).

I tried and found the V2 of Landmark speedtest, and indeed the results are different when it measures my 386, compared to the newer version. On V2 it has a landmark speed of 62.42MHz, where in V6 it's 69.45MHz.

Also I removed the IIT 4C87DLC-40 FPU I have on my 386. On speedsys the cpu score is 9.34, and when I re-installed the FPU, without changing anything else, it dropped down to 6.33. But in both cases, the 3dbench scores 16.1 fps.

Why not test using a Doom timedemo?

I guess because you want to test just the cpu speed. I believe the timedemo result will be affected by the graphics card.

@elianda
Thank you for sharing those documents, however I failed to see the requirement for a Cx487DLC-40 FPU when using a Cyrix 486DLC ALU. That article mentions that a Cx-87D87-40 is the FPU of choice for a 486DLC FPU.

Well, I didn't meant that it is a requirement, but replacing a classic 386 with a Cx486DLC is like targeting 486 system speed range, thus also a faster FPU has to be considered. And the Cyrix CPU/FPU combination is probably the fastest combination you can plug into a 386 mainboard.
Still be aware that this raises incompatibilities with some software.
This is mostly due to the fact that it looks like a 486SX + 487. So depending on the software you will have it detected as 486SX only with no FPU support or 486DX which fails sometimes since (AFAIK) the Cyrix combo does not implement all i486DX features or behaves differently at some situations.

Still if you have such a system, I would like to know if NT4 could be installed f.e.

Good question; I have never tried installing NT 4.0 on a Cyrix 486DLC, but would be up for trying it sometime. It seems the min. sys. req. for NT4 is 12 MB of RAM and a 486-25.

Markk wrote:

I guess because you want to test just the cpu speed. I believe the timedemo result will be affected by the graphics card.

As long as you can push ~1.5MB/s to your video card your results should be accurate up to 25fps.

Maybe. Anyway, I'm used to testing my pre-pentium stuff using 3dbench, speedsys and the landmark speed test. Main reason for this is that I can have all of them on a single floppy disk. And also I have several 286 and a couple of 8088 boards, where speedsys won't run, but the other two will do the job.

just an update, the machine is still under construction 🤣 , I need to get a power switch that fits and replace the dead 5.25" fdd , other than that I really enjoy using it!

today I made the final sound-related change: installed a SB1.5 with CMS chips and a SW-60XG for midi... 😀

That is the most attractive 386 mini-tower I have ever seen. I see you know how to appreciate matching white colours!

^ thanks, this particular tower is yellow though, the photo makes it look better 😀
I happen to have 2 of these baby-at towers, the way better one houses my 2nd p200mmx machine:

both are restored , they were rusty and dirty... here are some pics of the process ... http://s125.photobucket.com/albums/p55/restqp/at-restore/ 😊

I find that a little bit of acetone and a gentle abrasive will remove the non-matching lettering from CD/DVD-ROM drives.