The ECS Local Bus - Overview and Benchmarks on a 486DX-50 system \ VOGONS

The ECS Local Bus - Overview and Benchmarks on a 486DX-50 system

Topic actions

Post a reply

First post, by TheMobRules

Posted on 2023-06-26, 04:42

TheMobRules Offline

Rank Oldbie

Rank: Oldbie
Posts: 1447
Joined: 2016-01-19, 23:29

The pre-VLB/PCI 486 era is of great interest to me, in particular the attempts of several manufacturers to overcome ISA bus limitations by introducing their own proprietary local bus implementations, none of which seems to have been successful as shortly after VLB became the de-facto standard until PCI took over for good.

My goal with this thread is to focus in the local bus implemented by ECS in a couple of their motherboards, benchmark its graphics performance when using an ECS Local Bus (ELB) video card vs. an ISA card with the same chipset from that era and determine how much of an advantage it would have in that short period before VLB became commonplace. I will be conducting the tests using an Intel 486DX-50, which was the "ultimate" Intel x86 CPU at the time, at least until the DX2-66 was introduced (Aug-10 1992 according to CPU World).

*** Motherboard ***
I found this ECS SL-486E board on eBay for a very reasonable price on eBay when looking for EISA stuff:

Initially, I thought is was VLB and it was listed as such (for obvious reasons, just look at the bottom slot), but after some quick research I learned that it is actually the proprietary ECS local bus. So, I hesitated to buy it since the seller listed it as "working and tested with a VLB card", knowing that inserting a VLB card would either just not work (best case scenario) or produce some magic smoke, maybe blowing some components out. Also, even if it did work, it would be almost impossible to find an ELB graphics card, without which it would be really uninteresting.

However, while browsing other items listed by the seller I found what was listed as a "VLB graphics card" with the Tseng ET4000AX chipset and suddenly it clicked! A quick search on TH99 revealed that this was one of the two known ELB graphics cards (the other one having an S3 911 chipset), so it was most likely pulled from the same system as the board and it is what the seller probably used to test the board thinking it was VLB. So I pulled the plug and bought both!

*** Graphics Card ***
The label on the BIOS chip seems to identify the card as a VI-811/833. It looks just like a standard VGA card with the ET4000AX chip, 1MB of 70ns DRAM, AT&T DAC, clock generator and a few PALs and logic chips:

When I got the board and graphics card, I did some quick tests, modified the Dallas DS1387 chip to use a coin cell battery and everything seemed to work just fine. But suddenly I started to get artifacts on the screen, which made me very sad due to the relative rarity of these components. Luckily, the great VOGONS community came to the rescue and the amazing troubleshooting skills of mkarcher allowed me to identify and fix the broken inner trace on the motherboard that was causing the artifacts. See this thread for more details and an almost complete pinout of the ECS local bus!

By digging into the graphics card BIOS I found that it identifies as a Genoa MultimediaVGA 7900 series. There is some info online about the ISA version of the same card, but nothing about the ELB variant. However, the drivers I found detect the card and work perfectly, both on Windows 3.1 and the VESA 1.2 TSR (which is very useful since the BIOS is VGA only, no support for VESA modes built-in and UNIVBE does not work properly). The card can do 16M color, which Genoa proudly touts in the setup program, to the point of including a DOS app to demo that capability. I am attaching both the BIOS and drivers to this thread in case it's useful to someone else:

Filename: VI811-833_REV1.2_BIOS.zip
File size: 15.61 KiB
Downloads: 40 downloads
File comment: VI-811/833 REV 1.2 BIOS
File license: Fair use/fair dealing exception

Filename: GENOA-MM.ZIP
File size: 1.9 MiB
Downloads: 40 downloads
File comment: Genoa MultimediaVGA 7900 Drivers
File license: Fair use/fair dealing exception

*** Test setup ***
This is the setup I will be using for the tests:

CPU: Intel 486DX-50
Motherboard: ECS SL-486E
Cache: 256KB 20ns SRAM
RAM: 8MB 70ns FPM
Graphics card: VI-811/833 ET4000AX 1MB (ECS local bus)
Controller: BusTek BT-742A EISA SCSI Host Adapter
HDD: IBM 449MB SCSI HDD
FDD: 3.5'' 1.44MB floppy drive

Considering the bus is running at 50MHz, the following are the tightest timings I was able to achieve in the BIOS for the system to be stable, otherwise it wouldn't get past the POST (I also had to set JP3 on the graphics card to run at 50MHz, which probably introduces a wait state):

NOTE 1: switching to 15ns cache chips didn't allow me to tighten the timings any further so I left the 20ns ones that came with the board. Similarly, for the RAM I left the original 8MB 70ns as I couldn't use better timings even with 60ns SIMMs.

NOTE 2: regarding the RAM, the board seems to do some kind of interleaving as filling a single bank I get a Memory Bandwidth of around 85 MB/s in SPEEDSYS, while both banks filled get me to 114 MB/s. There doesn't seem to be any other noticeable performance increase though.

The following post will include results of running various tests from Phil's DOS Benchmark Pack, and comparing them to an equivalent ISA card with the same ET4000 chipset.

Reply 1 of 7, by TheMobRules

Posted on 2023-06-26, 04:42

TheMobRules Offline

Rank Oldbie

Rank: Oldbie
Posts: 1447
Joined: 2016-01-19, 23:29

*** System Speed Test Ver 4.78 ***

Results in SPEEDSYS are as expected for a DX-50, with the board providing good main memory performance. More interestingly, VESA memory speed is rated at 9694KB/s, which puts the ECS local bus card at around the same level of your average cheap Cirrus Logic VLB card from 93-94. Not bad for 1992!

*** CACHECHK v4 ***

1 CACHECHK v4 2/7/96  Copyright (c) 1995 by Ray Van Tassle. (-h for help)
2 CMOS reports: conv_mem= 640K, ext_mem= 7,424K, Total RAM= 8,064K
3 Clocked at 486 50.0 MHz
4 Reading from memory.  
5 MegaByte#:      --------- Memory Access Block sizes (KB)-----
6       1    2    4    8   16   32   64  128  256  512 1024 2048 4096 <-- KB
7 0:   21   21   21   21   25   25   25   25   25   34   --   --   --    æs/KB
8 2:   21   21   21   21   25   25   25   25   25   34   34   34   34    æs/KB
9 3  4  5  6  7  8  <--- same as above.
10
11 Extra tests----
12 Wrt  21   21   21   21   21   21   21   21   21   21   21   21   21<-Write mem
13 This machine seems to have one cache!? [read]
14    !! cache is 256KB --  44.1 MB/s  23.8 ns/byte  (137%)          4.5 clks
15    Main memory speed --  32.0 MB/s  32.8 ns/byte  (100%) [read]   6.3 clks
16    Effective RAM access time (read ) is 131ns (a RAM bank is 4 bytes wide). 
17    Effective RAM access time (write) is  81ns (a RAM bank is 4 bytes wide). 
18    Clocked at 486 50.0 MHz.  Cache ENABLED.
19 Options: -t0

CACHECHK seems to struggle telling the difference between L1 and L2 cache speeds, which is strange considering SPEEDSYS has no problem with it. Other than that, results look good.

*** Video speed benchmarks ***

For benchmarking the video speed, I wanted to quantify how much of an advantage the ECS local bus provides with respect to ISA. My initial choice was this generic Tseng ET4000AX card with 1MB of 80ns RAM, it has the exact same chipset as the ELB card so it would be a good candidate to tell what is the impact of the bus speed in video performance:

However, I was unable to run this card at 0WS on this board (runs fine on other boards though) which slightly cripples the video performance. So I added to the comparison another card which can run at 0WS on the ECS board and has the fast WD90C31 chipset: the Diamond SpeedStar 24X. This should be a good example of top ISA video performance on this board (without overclocking the bus, which causes problems with my EISA SCSI controller):

I used a combination of Phil's Benchmark Pack and VIDSPEED v4.0, focusing mostly on VGA 320x200 @ 256 color since that would be the target of most games for this system, and higher resolution VESA modes are too slow to detect significant differences. These are the results:

*** Final thoughts ***
Overall, I had a lot of fun tinkering with this board and graphics card. I very much enjoy the combination of these quirky and uncommon things like the custom local bus or EISA. This one would definitely have been a top performer around mid-1992 or so, just before the DX2-66 and VLB became a thing! I would like to estimate how much a system with these parts would have cost back then, just for giggles.

Oh, and if someone has the S3 911 variant of the ECS local bus VGA (VI-911) and wants to part with it, let me know! It would be very interesting to compare it with the Tseng.

Last edited by TheMobRules on 2023-07-01, 18:06. Edited 1 time in total.

Reply 2 of 7, by mwdmeyer

Posted on 2023-06-26, 04:59

mwdmeyer Offline

Rank Oldbie

Rank: Oldbie
Posts: 620
Joined: 2012-03-06, 11:35
Location: Sydney, Australia

Oh perfect timing, I just picked up a very nice DX50 that has one of these slots, so very interested to see. I have a ISA ET4000 in it currently. I don't have any ECS cards though!

Vogons Wiki - http://vogonswiki.com

Reply 3 of 7, by Anonymous Coward

Posted on 2023-06-26, 05:36

Anonymous Coward Offline

Rank l33t

Rank: l33t
Posts: 4833
Joined: 2008-03-20, 05:37
Location: Shandong, China

Very nice. I wonder if there are any other EISA boards out there with proprietary local bus slots.

"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 4 of 7, by rasz_pl

Posted on 2023-06-26, 06:42

rasz_pl Offline

Rank l33t

Rank: l33t
Posts: 3523
Joined: 2017-06-04, 00:57

I hope you include BUSCLK overclocking for the ISA card.

Open Source AT&T Globalyst/NCR/FIC 486-GAC-2 proprietary Cache Module reproduction

Reply 5 of 7, by TheMobRules

Posted on 2023-07-01, 18:09

TheMobRules Offline

Rank Oldbie

Rank: Oldbie
Posts: 1447
Joined: 2016-01-19, 23:29

Updated the thread with benchmarks and comparison to ISA graphics cards!

rasz_pl wrote on 2023-06-26, 06:42:
I hope you include BUSCLK overclocking for the ISA card.

Unfortunately overclocking the bus makes my EISA controller freak out, so I had to stick to 8.33 MHz. Later on I may try with a regular ISA controller if I get it to work.

Reply 6 of 7, by mkarcher

Posted on 2023-07-01, 19:30

mkarcher Offline

Rank l33t

Rank: l33t
Posts: 2643
Joined: 2019-01-19, 16:29
Location: Germany

The ISA performance on that board is mediocre. The theoretical maximum for oWS 16 bit writes on ISA is 8MB/s, the practical maximum is around 7.2 to 7.4MB/s due to memory refresh cycles. Both the Diamond Speedstar and the ET4000 card in 0WS mode should easily hit that target, so you can be sure the bottleneck is the main board, not the video card. A major performance killer for ISA is a main board that waits for local bus cards to claim a cycle before it starts forwarding that cycle to ISA. Luckily, many 486 chipsets have the option to choose whether they wait for one FSB clock or two FSB clocks. If you actually have a local bus card installed at 50MHz FSB, waiting two clocks is the recommended setting, but if you use an ISA graphics card, waiting only one clock speeds up graphics performance. The BIOS setup usually calls this option something like "ELBA sample point" or "LDEV sample point". I don't see a setting like that on the chipset settings you posted, but you might have a setting like that on the right half of the chipset setup.

I'm used to see around 5.1 to 5.3MB/s 0WS ET4000 performance on 486 mainboards with LDEV sample point set to the fast configuration, which is still considerably slower than the 7MB/s that should be possible, but is the result of missing one ISA clock per transfer, such that the 16-bit writes happen every 3 clocks instead of back-to-back at every two clocks. Furthermore, I noticed you have "I/O recovery time" set to 11BCLK. This value is quite high, and might hamper performance in ModeX based games and 16-color games that use a lot of port I/O as well as acessing video memory using memory cycles.

Reply 7 of 7, by CoffeeOne

Posted on 2023-07-02, 10:23

CoffeeOne Offline

Rank Oldbie

Rank: Oldbie
Posts: 1155
Joined: 2019-12-25, 16:12
Location: Austria

Anonymous Coward wrote on 2023-06-26, 05:36:

Very nice. I wonder if there are any other EISA boards out there with proprietary local bus slots.

There are.
Re: My cool 486 EISA tower

Go to top of page Go to top of page

Back to General Old Hardware

Main menu

Common searches