VOGONS


3/486 OPTI495SLC VL-BUS benchmarks

Topic actions

First post, by kixs

User metadata
Rank l33t
Rank
l33t

Hi,

this is a hybrid 386/486 motherboard. I'll put it through some benchmarks with different CPUs and VGA cards. Later I'll do the same tests with my other hybrid boards (in separate threads). I'll update the thread as the time permits to make the tests.

Here is a photo of the board:
tqNSSfpm.jpg

And the jumper settings:
https://stason.org/TULARC/pc/motherboards/U/U … -BUS-3-486.html

Basic settings:
4x4MB 60ns SIMM
256KB cache (fastest still stable timings)

What I have prepared for tests:

FVSp2wcl.jpg

CPUs
Intel 386DX-33
Am386DX-40
Intel RapidCAD at 33Mhz (maybe at 40Mhz)
Cyrix 486DLC-33
Ti486DLC-40
Cyrix 486DRx2-66
Ti486SXL2-66 3V (maybe at 80Mhz)
Ti486SXL 5V CPUs doesn't work with this mobo - gets very unstable
Maybe a true 486DX - TurboChip 5x86-133.

FPUs
Intel 387-33
Cyrix 487-40
IIT 487-40
ULSI 387-40
IIT X2 25/50 (runs ok at 33/66, maybe I'll try 40/80)
ULSI X2 33/66 (maybe 40/80)

VGA cards
ISA Diamond Speedstar64 2MB DRAM
VLB Diamond Stealth64 2MB DRAM

Maybe I'll try other VGA cards although I think these two should be the best.

Benchmarks more or less the usual
3DBench 1.0c
PCPbench
Doom
Quake
SpeedSys
Comptest
Winbench 3.11
Wintach 1.0
...

I won't do every combination as this would take too long 😅

and for now... a teaser 😊

yVu2Cc4m.jpg

Stay tuned 😉

Edit:
As requested I dumped the BIOS of this board with NSSI. It's dated 08/08/93.

Filename
495SLC.ZIP
File size
44.88 KiB
Downloads
128 downloads
File comment
Bios for Opti495slc dated 08/08/93
File license
Fair use/fair dealing exception
Last edited by kixs on 2019-03-30, 01:00. Edited 2 times in total.

Requests are also possible... /msg kixs

Reply 3 of 35, by keropi

User metadata
Rank l33t++
Rank
l33t++

can't wait to see the results!

edit:
btw, this is the mobo you have 😀 https://stason.org/TULARC/pc/motherboards/U/U … -BUS-3-486.html

🎵 🎧 PCMIDI MPU , OrpheusII , Action Rewind , Megacard and 🎶GoldLib soundcard website

Reply 4 of 35, by kixs

User metadata
Rank l33t
Rank
l33t

Me too 😀 Was really excited as this is the first time I tried RapidCAD, Ti486SXL2-66 and Cyrix 486DRx2. Till now I didn't even know if they work and they do 😁

The link for this mobo is already below the photo 😉

Requests are also possible... /msg kixs

Reply 5 of 35, by feipoa

User metadata
Rank l33t++
Rank
l33t++

Very nice looking hardware choices (cannot comment on the MB though).

Curious that your 486SXL 5V isn't stable. I suspect it is related to the register parameters you are using. I suggest using the cyrix.exe utility written by Paul Gortmaker as it awards the greatest flexibility.

Option			Action
------ ------
-a[-] Enable [disable] A20M input.
-b[-] Enable [disable] BARB input.
-c[d] Two way assoc. [direct-mapped] cache.
-d Raw dump of control registers and exit.
-e[-] Enable [disable] 1k cache via CR0.
-f[-] Enable [disable] FLUSH input.
-h Print this help screen and exit.
-i<n> Inhibit non-cachable region <n>, where n={1,2,3,4}.
-k[-] Enable [disable] KEN input.
-m[-] Enable [disable] cache of 1st 64k of each Mb.
-o Override check for existence of Cyrix DLC chip.
-p Disable all power saving modes.
-q Query DLC configuration status and exit.
-r[-] Disable [enable] cache of 640k -> 1M ROM shadow area.
-s[-] Enable [disable] SUSP i/o.
-v Print version number and exit.
-x<hex>,<size> Don't cache <size>kb from segment <hex>.

I have found that stability issues with the SXL are related to incorrect settings. I suggest taking a look at this thread, Register settings for various CPUs

It is likely that the Evergreen SXL2 module takes care of some of these issues. For starters, try cyrix.exe -e -f- -b -m- -xA000,128 -xC000,256

You mentioned that the IIT x2 25/50 runs ok at 33/66. This wasn't the case with mine. What tests are you running to determine it is ok at 66 MHz? I have found that it gives the appearance of working, until you stress it. Try running Quake two or three times in series then run CABT. I couldn't even get mine going well at 55 MHz.

Where did you source the ULSI FPU 33/66? I think user Anonymous Coward ran his at 80 MHz. Not sure what the outcome was.

What I found very interesting about the clock doubled FPU's is that they aren't all that much faster than a Cyrix FasMath at 40 MHz. For example, SXL2-50 with IIT-50 scores 188 in Landmark FPU, while SXL-40 with Fasmath-40 scores 179.

Further interesting is to compare running an SXL2-50 w/FasMath at 25 MHz compared to an SXL-40 with Fasmath at 40 MHz. The Landmark FPU results are 169 vs. 179, respectively. Makes me wonder about the benefit of these clock doubled FPUs.

Are you able to run your SXL2 at 80 MHz w/FasMath and compare the Landmark ALU/FPU results with SXL2-66 w/ULSI x2? Unfortunately, I do not own the ULSI x2 to run this. My theory is that the overall benefit (ave. of FPU and ALU) of SXL2 at 80 MHz is more desirable than running the SXL2-66 w/ULSI x2. My ideal system is to run an SXL2 at 80 MHz. My Evergreen SXL2-66 runs at 80 MHz, so I suspect yours will as well. I'd like to get the PGA132-to-PGA168 module built so pretty much everyone can have access to this premium configuration.

What are you going to do about a heatsink on the Evergreen?

Are you able to run a PGA-168 version of the TI486SXL2 on this motherboard (on VRM interposer)? I'd be very interested to see if the PGA version of the SXL2-66 can run at 80 MHz or if we are, for some reason, limited to the QFP-144 for such speeds.

Would be nice to see an IBM Blue Lightening BL3 included in this comparison if you have one. Also, could you just run a few PGA168 CPUs so we can add perspective to things? maybe an i486DX-25, i486DX-33, i486DX50, and i486DX2-66? Its pretty clear that the Am5x86-133 will completely clobber any PGA132 CPU.

Plan your life wisely, you'll be dead before you know it.

Reply 6 of 35, by Anonymous Coward

User metadata
Rank l33t
Rank
l33t

My tests with ULSI Math Co DX2 was far from conclusive. I just ran a synthetic benchmark to compare the clock doubled FPUs to the Fasmath on a 40MHz bus. I was too afraid to stress them, because they're very rare and I didn't want to risk frying them. I agree with the conclusion that there isn't much benefit to the clock double parts over the Fasmath 40MHz. That led me to suspect that perhaps the clock doubling was not enabled by default and required some utility, but so far one hasn't been found. It would be nice to contact some of the people who designed these chips to find out.

"Will the highways on the internets become more few?" -Gee Dubya
V'Ger XT|Upgraded AT|Ultimate 386|Super VL/EISA 486|SMP VL/EISA Pentium

Reply 8 of 35, by kixs

User metadata
Rank l33t
Rank
l33t

Yesterday I did half the benchmarks with Cyrix 486DRx2-66. Will add the results when I finish them. I'd like to test it at 80MHz, but don't want to break it...

I might reconsider doing that many configurations. Maybe just same basic config with 386DX-40 or 486DLC-40 as these are pretty common. And then do more extensive tests with rare cpus/fpus.

I tested both X2 fpus some months ago and if I remember correctly both work at 40MHz bus (I don't really like 33MHz bus). You are correct, there isn't much benefit in real apps (like AutoCAD). FPU benchmarks are pretty high tho. I also have a Waitek 3167-33 but the motherboard and apps must support it.

I'll check cache parameters with SXL 5V. Now I just use the BIOS enable internal cache.

I might check other hybrid 386/486 boards first to see which is the fastest. Memory subsystem is pretty low on OPTI495xxx chipsets. Not sure what the 386 VLB Panda uses.

Requests are also possible... /msg kixs

Reply 9 of 35, by feipoa

User metadata
Rank l33t++
Rank
l33t++
kixs wrote:

Yesterday I did half the benchmarks with Cyrix 486DRx2-66. Will add the results when I finish them. I'd like to test it at 80MHz, but don't want to break it...

There is little fear of breaking it because it simply won't work at 80 MHz.

I might reconsider doing that many configurations. Maybe just same basic config with 386DX-40 or 486DLC-40 as these are pretty common. And then do more extensive tests with rare cpus/fpus.

I believe the DRx2-66, SXL2-66, and perhaps RapidCAD were introduced into the market to compete against low-end [real] 486's. I personally think adding at least one real 486 to the list will add much needed perspective on the potential of these 386 upgrades. If just picking one, then i486DX-33.

I tested both X2 fpus some months ago and if I remember correctly both work at 40MHz bus (I don't really like 33MHz bus). You are correct, there isn't much benefit in real apps (like AutoCAD). FPU benchmarks are pretty high tho.

Tested the x2 FPUs with what software? I don't see how a 50 MHz FPU is going to work reliably at 80 MHz. Overclocking potential for 386-era chips was rather limited.

I also have a Waitek 3167-33 but the motherboard and apps must support it.

The only thing the Weitek 3167 is good for is to consume volume in your CPU storage bin. I think it would be really neat to have some type of translation software (or hardware?) which allowed the Weitek to run any x86 floating point code.

I'll check cache parameters with SXL 5V. Now I just use the BIOS enable internal cache.

I am almost positive your issues are related to incorrect CPU settings. What hard drive controller card are you using?

I might check other hybrid 386/486 boards first to see which is the fastest. Memory subsystem is pretty low on OPTI495xxx chipsets. Not sure what the 386 VLB Panda uses.

Good idea.

Plan your life wisely, you'll be dead before you know it.

Reply 10 of 35, by kixs

User metadata
Rank l33t
Rank
l33t

Small update... found some 12ns cache chips and swapped them. Now it works fine with fastest 2-1-1-1 cache timings and 256KB cache. Before it would run reliably only with 128KB cache. With 256KB cache it needed 3-1-1-1 and any performance gains with larger cache were negated. I tested different 15ns chips but usually Quake would crash. Average boost is around 5%.

Benchmarks are still on my TO-DO list... damn time flies 😠

Edit:
Typo...

Last edited by kixs on 2019-03-30, 00:55. Edited 1 time in total.

Requests are also possible... /msg kixs

Reply 11 of 35, by feipoa

User metadata
Rank l33t++
Rank
l33t++
kixs wrote:

IIT X2 25/50 (runs ok at 33/66, maybe I'll try 40/80)

You sure? Did you try running Quake and CABT? I found the IIT 25/50 chip couldn't handle even 55 MHz reliably. First impressions were that it worked, but it ultimately failed.

Plan your life wisely, you'll be dead before you know it.

Reply 13 of 35, by Deunan

User metadata
Rank Oldbie
Rank
Oldbie
kixs wrote:

Small update... found some 12ns cache chips and swapped them. Now it works fine with fastest 2-1-1-1 cache timings and 256KB cache. Before it would run reliably only with 128KB cache. With 256KB cache it needed 3-1-1-1 and any performance gains with larger cache were negated. I tested different 15ns chips but usually Quake would crash. Average boost is around 5%.

Benchmarks are still on my DO-TO list... damn time flies 😠

Argh, I have the exact same problem with this mobo. 15ns chips will not work with 2-1-1-1 setting at 40MHz, the system is not stable and will crash Doom during the loop test and the DOS-extender version of NASM just hangs on execution. Did you use modern SRAM chips or got those from ebay?

As a side note, this chipset has worse RAM read timings with 0WS settings in BIOS than my other DX mobo that requires 1WS. It seems the cache WS are actually added to RAM WS - that other mobo has that effect too but it lets me use 0WS for cache so the end result is way better. All in all it's nice since it can take both 386 and 486 CPUs but 386 performance suffers because of that, and 486 I guess a bit as well but the L1 cache is able to mask that (mostly).

Reply 14 of 35, by kixs

User metadata
Rank l33t
Rank
l33t

I found these chips locally - just 9 pieces. They were made in 1999.

I agree the memory subsystem is pretty bad in this one. But also other combo 3/486 boards aren't any better, unfortunately 🙁

Requests are also possible... /msg kixs

Reply 15 of 35, by feipoa

User metadata
Rank l33t++
Rank
l33t++

I have a similar board based on the OPTI 495SX. It is horrifically fussy and many CPUs didn't work on it. This is a board that you want to give to someone who enjoys testing their patience.

Plan your life wisely, you'll be dead before you know it.

Reply 16 of 35, by kixs

User metadata
Rank l33t
Rank
l33t

I actually like this board. Maybe because it was my first 386/486 combo 😉

Only Ti486SXL 5V isn't compatible with it. SXL2-66 works fine.

Requests are also possible... /msg kixs

Reply 17 of 35, by canthearu

User metadata
Rank Oldbie
Rank
Oldbie

I have one of these as well. (Opti-495SLC)

I can agree, with a 486DX2-66 overdrive chip, performance lags by around 10%-15% at times over the better 486 board I have.

Plus being limited to 30 pin chips, makes it harder to fit in a decent amount of memory in it (4meg 30pin SIMMs are kind of rare)

Reply 19 of 35, by Deunan

User metadata
Rank Oldbie
Rank
Oldbie

I figured this might help someone and serve as a reference point:

1) AMI BIOS dumped with NSSI will crash on checksum error and must be corrected before burning onto a chip. The procedure is simple, clear (as in replace with zeros) bytes in 0xe13c to 0xe14f range. Maybe turning off shadow RAM before dumping would help?
2) This BIOS detects U5S(X) CPUs as TI POTOMAC - which is incorrect but doesn't seem to hurt anything. The previous version (dated 11/11/92) will detect a 486SX.
3) The 495SLC chipset (and perhaps all 495 series) doesn't work with more than 128kB of cache at 2-x read WS setting unless you have installed SRAM chips faster than 15ns

I did some benchmarks with U5S CPU at 40MHz, 8MB RAM set to 0/0 WS and VLB Cirrus CL-GD5424 (1MB), the results are in the following order:
NSSI (CPU, FPU); SI 8.0; Speedsys 4.78; TOPBENCH; Speedsys 4.78 VESA mem in kB/s; 3DBench (1.0, 1.0c); PC Player Benchmark (320, 640); Landmark 6.00; Doom (low, high); FractBench

256kB cache 3-2-2-2, 1 WS: 26750, 3392 ; 111.8 ; 21.12 ; 190 ; 8883 ; 38.4, 37.7 ; 7.3, 2.8 ; 188 MHz AT, N/A, 9187.29 chr/ms ; 2134/1331, 2134/3862 ; 0m:09s.44
256kB cache 3-1-1-1, 1 WS: 26390, 3392 ; 111.8 ; 21.12 ; 190 ; 8916 ; 38.4, 38.2 ; 7.4, 2.9 ; 245 MHz AT, N/A, 9102.22 chr/ms ; 2134/1286, 2134/3793 ; 0m:09s.40
256kB cache 3-1-1-1, 0 WS: 27075, 3398 ; 111.8 ; 21.90 ; 201 ; 8944 ; 40.0, 39.9 ; 7.4, 2.9 ; 246 MHz AT, N/A, 9102.22 chr/ms ; 2134/1268, 2134/3756 ; 0m:09s.45
128kB cache 2-2-2-2, 1 WS: 26325, 3392 ; 111.8 ; 21.04 ; 190 ; 8910 ; 38.4, 37.5 ; 6.9, 2.8 ; 246 MHz AT, N/A, 9187.29 chr/ms ; 2134/1485, 2134/4083 ; 0m:09s.39
128kB cache 2-2-2-2, 0 WS: 27100, 3396 ; 111.8 ; 21.82 ; 201 ; 8928 ; 40.0, 39.0 ; 6.9, 2.8 ; 246 MHz AT, N/A, 9187.29 chr/ms ; 2134/1466, 2134/4042 ; 0m:09s.45

Few things to note:
NSSI result can "float" up and down, so it's not very accurate. Obviously the FPU score is from emulation code. TOPBENCH result is very VGA-speed dependent.
I was unable to even boot the system with 2-1-1-1 or 3-1-1-1 settings on 128kB of cache, while it was no problem at all with 256kB. It's the opposite for a 386DX where I'm forced to use 128kB to be able to use 2-x latency and not 3-x which greatly reduces the system performance. 15ns SRAM chips will crash Doom (eventually) with 2-x settings on 386.
0 WS for cache write is not long term stable at 40MHz, at least not for a 486. I will have to redo 386DX benchmarks someday to test that.

The short version seems to be this:
If you plan to use this mobo with a 386, either get 12ns or better SRAM chips or stick to 128kB of cache with 2-x settings. 386 doesn't have burst mode so it doesn't matter if you set 2-1-1-1 or 2-2-2-2. The safer option is however to set cache write WS to 1.
With a 486 go for 256kB of cache at 3-1-1-1,1 - might just work OK with 15ns chips as it did for me and that seems to give better gaming perfomance than 2-2-2-2,1. Even at 3-2-2-2,1 the additional amount of cache seems to be more important than latency. The difference is however not that great either way.