jesolo wrote:

Thank you for posting all the results skv400. Would be interesting to see how the TI 486SXL 40 MHz CPU stacks up against the TI […]
Show full quote

Thank you for posting all the results skv400.
Would be interesting to see how the TI 486SXL 40 MHz CPU stacks up against the TI 486DLC 40MHz & the Cyrix 486DRx2 66MHz on these benchmark tests.
I must be honest, I didn't even know there was an ULSI Math-co DX2, but then again, I also didn't investigate this much further.

The first PC I bought for myself was a Cyrix 486DLC 40MHz with the Cyrix FastMath 40MHz FPU (I still have it).
For its time, it was a very cheap alternative and was able to stand up to (almost) the speed of an Intel 486DX 33MHz.

But I agree, the speed increase of the Cyrix 486DRx2 66MHz over the Cyrix/TI 486DLC 40MHz simply doesn't justify the price premium.
If Cyrix had brought out the Cyrix 486DRx2 66MHz much earlier, it might have been worthwhile buying one and perhaps paying that price premium, but by the time that CPU hit the market, your (true) 486 CPU's were already affordable to most users.
I can still remember that, when I bought my Cyrix 486DLC 40 MHz back in late 1993, an equivalent Intel 486DX 33MHz rig cost only slightly more. But, back then, every cent counted and I was very intrigued by the Cyrix DLC CPU's and therefore decided to go for that one.

thanks to your reply.

i interest cyrix/amd's upgrade cpus. 486slc.. 486dlc.. 486drx.. 5x86.. am5x86(486).. Cyrix MII, K6-2,3,etc..

your 1993's memory is give to me vitality. i will put in 386dx and shot movie on tomorrow.

and find my pci cl-gd5434 cards and separate ram chip and put in VLB/ISA 5434s.

i hear cirrus logic 5434 has 32-bit memory bandwidth with 1MB Video ram, and 64-bit memory bandwidth with 2MB Video ram.

if chip ram put in and increase memory bus to 64-bit, maybe doom and 3dbench performance.

i will shoot this benchmark tomorrow.

I suspect that like the ET4000W32 cards, the CL GD5434 only benefits from the second meg of memory in Windows acceleration. It would make for an interesting test.

I am also not totally sure that the 487DLX-50 is slower than the ULSI Math Co DX-2. At least I remember in my synthetic benchmarks that the IIT came out ahead at the same clock speed. They may have made some tweaks over the 4c87-40.

Anonymous Coward wrote:

I suspect that like the ET4000W32 cards, the CL GD5434 only benefits from the second meg of memory in Windows acceleration. It would make for an interesting test.

I am also not totally sure that the 487DLX-50 is slower than the ULSI Math Co DX-2. At least I remember in my synthetic benchmarks that the IIT came out ahead at the same clock speed. They may have made some tweaks over the 4c87-40.

iit 40mhz is faster than ulsi 40mhz?

How tuning used? I have iit cyrix ulsi 387sx fpus. But iit is worst..

My ulsi math-co dx 40mhz is broken but can running speedsys.

And over 10points. But put in iit 4c87dlc 40mhz. Can't reach 10 pionts. 9.xx point.

I need your fpu and cpu tuning know-how.

Speedsys really isn't a good benchmark for FPU's. 386s without a FPU have the highest score. I'd suspect that ULSI doesn't even work in Speedsys.

kixs wrote:

Speedsys really isn't a good benchmark for FPU's. 386s without a FPU have the highest score. I'd suspect that ULSI doesn't even work in Speedsys.

Can you tell me more 387 fpu benchmark applications?

Quake,FBench,FLops,cabt i know 4 applications.

Whetstones in Checkit and Comptest.

I also use AutoCAD 10 & 11 to perform regen and hide on complex objects. It can take a few minutes. So differences between fpus are more obvious.

kixs wrote:

Whetstones in Checkit and Comptest.

I also use AutoCAD 10 & 11 to perform regen and hide on complex objects. It can take a few minutes. So differences between fpus are more obvious.

Checkit and comptest... i'll try it. Thanks.

And can you tell me autocad 10&11 fpu test method?

I mostly use version 10. It has sample drawings. The largerst one is "stpauls.dwg". You load it and input command: "hide" and use your stopwatch to time the execution. It should take around 3-4 minutes on a fast 386 with FPU. Acad doesn't work without it, although one can use fpu emulator 😉

kixs wrote:

I mostly use version 10. It has sample drawings. The largerst one is "stpauls.dwg". You load it and input command: "hide" and use your stopwatch to time the execution. It should take around 3-4 minutes on a fast 386 with FPU. Acad doesn't work without it, although one can use fpu emulator 😉

Thank you. I'll try this method

Fpu emulator is discrete execute file?

Not too sure if this counts as it is not a pure 386... though I see there are some upgrade CPU results here already so hopefully it won't offend anyone posting this here.

The specs; -

Motherboard: Alaris Cougar
Chipset: Opti 82C499
Cache: 256k (20ns) / 512k (15ns)
Memory; 8MB (70ns) Proprietary Compaq / IBM
CPU: IBM 486DLC3 a.k.a BL3
FPU: No

BL2 / BL3 (256k cache)

BL2 / BL3 (512k cache)

The first set were with 256k (20ns) cache, the second... with 512k (15ns).

It's the best performing "386" I have tested. With only TI's SXL2-50 coming close, when in DLC2 (BL2) mode... not DLC3 (BL3).

skv400 wrote:

Can you tell me more 387 fpu benchmark applications?

there is a pack of benchs. some have fpu data, like Diag
BenchDOS has ones

Tertz wrote:

skv400 wrote:

Can you tell me more 387 fpu benchmark applications?

there is a pack of benchs. some have fpu data, like Diag
BenchDOS has ones

Thanks. I finished download this and executed batch file.

I do not think I have contributed to this thread yet, better late than never I guess. 😀

SPRING CIRCLE COMPUTER, INC. MDX340C/FOREX386 with an AMD 386DX33 and 64KB cache.

https://th99.bl4ckb0x.de/m/S-T/30013.htm

There are almost no speed related settings in the very bare BIOS setup so I ran Speedsys with defaults loaded, the memory throughput still seems decent for a 33 MHz 386.

When I get same spare time I'll go through all my 386 boards (around 30) and put them to the Speedsys and Phil's tests - should be alot of fun 😁

I see I've never answered this question:

skv400 wrote:

kixs wrote:

I mostly use version 10. It has sample drawings. The largerst one is "stpauls.dwg". You load it and input command: "hide" and use your stopwatch to time the execution. It should take around 3-4 minutes on a fast 386 with FPU. Acad doesn't work without it, although one can use fpu emulator 😉

Thank you. I'll try this method

Fpu emulator is discrete execute file?

Yes, there are several FPU emulators as an ordinary exe file. Usually they load tables in extended/expanded memory.

The best is/was Q387. It's shareware and newer versions (V4.xx) are quite difficult to find. Simtel has 3.66. Franke 387 is also good.

It seems Franke.387 was released as a full version by the authors:
http://icfs.de/english/franke387.html

This thread has all the Q387 versions - Google finds anything 😀 :
http://www.vcfed.org/forum/showthread.php?508 … -in-this-thread

SKV400: Very interesting comparison. Thank you for posting these results. We don't get nearly enough 486DLC-40 and DRx2-66 users, especially those with VLB graphics!

The system I have setup is an AMI Mark V Baby Screamer based on the VLSI 330/331/332 chipset. It has 256K and 32 MB RAM. It has a Cyrix/TI 486SXL-40 with 8 KB cache as opposed to 1 KB. The graphics is CL-GD5434. FPU is Cyrix FasMath.

For DOOM timedemo 3, I get 14.1 fps. For your system with the same graphics card, you get DLC-40 = 11.3 fps, and DRx2 = 12.6 fps. With VLB graphics, you got DLC-40 = 12.85 fps, or DRx2 = 16.4 fps. Why is the performance increase for the DRx2, when using VLB, much greater than when the DLC uses the VLB graphics?

What is your CPU score from Nortin Sysinfo? Mine is 65.6. For some reason, the number is not showing on the graph from your screenshot.

For Quake, I get 2.7 fps. I did not see your ISA scores for Quake in your bar chart, but found it pretty interesting that the DRx2-66 w/33 MHz FPU only scored 2.6 fps. But once you clock doubled your FPU, you get quite a boost percent-wise, to 3.0 fps.

For Landmark v2, I get 130/185 (CPU/FPU). Yours weighs in at 198/215, which is a pretty nice [synthetic] boost.

For 3Dbench, I got 23.2. I did not see the ISA results on your bar chart. Your VLB results are impressive in this regard at 35 fps.

For cachechk, my L2 cache is at 33.5 MB/s, which is slightly slower than your 34.7 fps. For RAM read speed, I got 16.1 MB/s, while yours is only 9.4 MB/s. Is that correct? The worst speed I've seen was around 14 MB/s. Could you report the cachechk RAM write results? e.g. cachechk -w -d -t4

The best memory read times tend to come from SiS Rabbit-based boards, at 23.6 MB/s.

Did you try your DRx2 at 2x40 MHz?

Based on your results, it seems like the DRx2 might be worth it if your board has VLB. In my mind, though, if a board has a 486 CPU socket, then using one of these PGA-132 CPU upgrades looses some of its prestige. Does your board have a 486 and 386 CPU socket, or just a 386 CPU socket?

I also have a DRx2-66 and would some day like to compare it against the 486SXL-40 on my Mark V Baby Screamer. I would also like to attempt 486SXL2-66 operation with lots of cooling. An SXL2 at 2x25 MHz just isn't that exciting for me.

386_junkie: to me, anything which fits into a PGA-132 is 386 enough. The IBM BL3 is the end of the road.

AMD 386DX-40 @ 50MHz (20ns)
RAM = 8MB (60ns)
Cache = 256KB (20ns)
Chipset = Forex 421a
Graphics = Western Digital 90C31

Attached below are all the tests I did with three different graphics cards: -

Cirrus Logic 5434
Tseng 4000/W32
Western Digital 90C31

Wow, non-cached main memory writes and moves are really bad, especially for 50MHz. Did you have to kill the DRAM performance with a lot of wait states?

I agree, the RAM performance is bad. Here is a regular 386DX-40:

elianda wrote:

I agree, the RAM performance is bad. Here is a regular 386DX-40:

What chipset does this board have?

Anonymous Coward wrote:

Wow, non-cached main memory writes and moves are really bad, especially for 50MHz. Did you have to kill the DRAM performance with a lot of wait states?

The DRAM throughput is typical for this board: FOREX 386 motherboard thread

When you start to put FSB's beyond 40MHz = 25ns to 50MHz = 20ns... I would say this is where DRAM starts to become the primary bottleneck anyway. Since the fastest 30 pin DRAM around is 60ns... with the CPU latency shortening to 20ns along with cache, it could be that the DRAM latency is the main drag on the system.

The system is still pretty fast for a DX40... nearly 10fps for DOOM, 18.8 on 3Dbench using only an ISA VGA card... it's still pretty fast even with DRAM becoming a bottleneck.