The Trident is really bad isn't it. I think I would like an ET4000 but they're quite expensive at the moment so I can wait. I've got an 8900C with the MUSIC chip - what's is significance?

I found all the Tridents to be either noisy or have verticle lines when running DOOM, or PCPBench. Perhaps there is a work-around for that? The Tridents with the MUSIC chip did not have this problem - in fact, they had very clear displays.

I'd expect better results with 486SXL-40 cpu (is this with 8KB L1 cache?).

Yes, L1 is enabled. Some boards have L1 on/off option in the BIOS, while others I used software. I confirmed L1 is enabled based on the Speedsys L1 move speed and by using Landmark. Landmark always shows a big ALU speed jump when L1 is enabled, and a drop when it is disabled. The best Doom scores was 13.7 fps (UMC-based motherboard with ET4000/w32i). I was also very disappointed in the Doom results. Is there some trickery I need to do to the graphics?

It very much depends on the BIOS setup. Wait-states, ISA bus clock, Video/system cacheable... the most you get by fastest cache timings and ISA bus dividers 1/3 or 1/4. Some graphic cards have problems with 1/3.

With optimized BIOS settings you should be getting around 24 fps in 3DBench. Don't know for DOOM - but check phil's spreadsheet.

Just checked some of my old notes. With 486DLC-40 on Opti495slc 128KB cache and VLB S3 864 video card I get this results:

3DBench... 26.6
PCPBench VGA... 5.8
DOOM... 5905

With ISA video card (CL-5429 2MB) the 3DBench was 23.2, PCPBench VGA 5.7, DOOM 6966.

Interesting. My best DOOM timedemo3 score was 5462, which is better than your results, however your 3DBench is much better. I do not understand this. For PCPbench, I am using VESA Modus 100 (640x400 8bpp LFB), which is what my program defaults to.

What are your Landmark v2 results for ALU and FPU? I get 130 for ALU and 185 for FPU.

EDIT: When I run PCPBench /VGAmode (320x200x8bpp) I get a score of 6.7 on my UMC/ET4000/W32i system.

I don't have the system prepared. So can't try anything at the moment. I guess the SXL shows its powers 😁 and UMC chipset is really fast.

I wish my motherboard would support SXL, but it doesn't 🙁

I ran 3dbench again and received a score of 24.3 on the UMC system. I am not sure why it jumped from 19.6 to 24.3, however when I was trying to run 3dbench just now, it would hang up right when the program started, and a mouse pointer whould show on the screen. I disabled my mouse DOS drivers and 3dbench runs flawlessly now. Perhaps having the mouse drivers loaded in the autoexec was lowering my 3dbench score?

With sound enabled, I get 6950 in DOOM. Sound card is an ESS Audiodrive (ES1868F)

For benchmark testing you should make sure ISA bus is running at the same speed on all platforms. I would go with 8MHz as that is standard.

opti495SX , AMIBIOS 40MHz, 16MB DRAM, 256kb L2, 10MHz ISA, ET4000W32i, No sound

386DX (L2WT ON L1 — ) 10681 Realtics
386DX (L2WB ON L1 — ) 9473 Realtics
486SXL (L2WB ON L1 OFF ) 8241 Realtics
486SXL (L2WB ON L1 ON ) 5801 Realtics
486SXL (L2WB ON L1 ON ) 6132 Realtics (8MHz ISA)
486SXL (L2WB ON L1 ON ) 5770 Realtics (MRBIOS)
486SXL (L2WB ON L1 ON ) 6097 Realtics (MRBIOS, 8MHz ISA)
486SXL (L2WB ON L1 ON ) 5891 (CL GD5434, 8MHz ISA)
486SXL (L2 WB ON L1 ON) 5610 (CL GD5434)
486SXL (L2 WB ON L1 ON) 5539 (MRBIOS, CL GD5434)

These are my results for DOOM only. I could not run with sound because my system is not flushing the cache properly despite my board supporting it in hardware (yes I tried the software too). When a DMA happens the system comes crashing down. Works just fine on my PEAK/DM board. Weird, eh? I had another 495SX board in the past that worked even better, so it seems it isn't necessarily a problem with the chipset.

The results were at 8 MHz. At 10 MHz on the UMC board, I get 5223 (no sound), which is 14.3 fps. Actual game play feels a lot faster though.

This is probably also worth mentioning. Having EMM386 running will also slow things down a bit. All of my tests were done with nothing loaded at boot.

Have a look at the last 3 results I added to my post above. I switched to a Diamond Speedstar64 with 2MB. Unlike your tests, mine is faster than the ET4000W32i. It might come down to the manufacturer of the graphics card (the video BIOS) or even the VGA/motherboard combination.

Man, it's incredible how much complex was Doom compared to Wolf3d looking at these framerates.

Attached are the ISA graphics card results from a VLSI-based 386 board. Click image for larger view. I've included the 0ws and non-0ws results because I noticed that cards cannot seem to handle 0ws in Windows 3.11. I am not sure if this is due to having too many ISA cards on the bus (SCSI, I/O, ethernet, sound, and graphics), or if there is one particular ISA card (perhaps SCSI?) which does not do well if the graphics card is at 0ws. Any ideas?

While the GD5434 card had the fastest results, albiet only marginally above the ET4000, it did not product as clear of a VGA image in DOOM and 3dbench compared to the Mach64 or the ET4000AX / ET4w32i cards.

Sorry to resurrect an old thread, but I recently acquired an ISA Micro-Labs ET4000/w32i 2MB and am trying to test it in my Am386DX/40 which happens to also have 32MB of RAM.

From your earlier comments feipoa, it sounded like you were able to get a linear frame buffer (LFB) on this card. However I am only seeing VBE 1.32 on the card, and none of the versions of UNIVBE I have seem to be helping.

Can you share what your configuration was? I'd like to get the LFB working so I can program against some of the high-color modes without any bank switches.

Which versions of univbe did you try? Try out 5.0, 5.1, and 5.3 for starters.

There are Trident 8900D-R and others that have 32bit Memory and offer similar performance to et4000, except for windows accelration.

feipoa wrote:

Which versions of univbe did you try? Try out 5.0, 5.1, and 5.3 for starters.

I have tried 5.3a and 6.53. I can try a few older versions, but I'd like to confirm that you were seeing a linear frame buffer with your card. I have found conflicting information:

http://web.mit.edu/~mkgray/afs/bar/afs/sipb/s … oc/README.tseng

9.5 Restrictions ... On ET4000W32i, things are worse: the linear address base is hardwired on the card, and there is no reliabl […]
Show full quote

9.5 Restrictions
...
On ET4000W32i, things are worse: the linear address base is hardwired on the
card, and there is no reliable way to read it back from the card. You need to
know the address in some way, and specify it. The current code does an intelli-
gent guess at it, but this is no guarantee.

On ISA cards, things are much more simple: ISA only uses 24 address lines, and
hence the linear memory MUST lie within the 16 MB boundary. Together with the
4MB granularity of the linear memory base address on ET4000 cards, this means
that you cannot have more than 12 MB of system memory in the machine if you
want to use linear memory.
...
9.6 Some boards simply cannot work in linear mode

Yes, and in that case, you're out of luck.

There can be at least two reasons for this.

The first is the most common: the board manufacturer has left out the necessary
connections and hardware to be able to use linear addressing. This means that
no coding effort on this planet can help you with your problem: it is physi-
cally impossible to use linear addressing.

matze79 wrote:

There are Trident 8900D-R and others that have 32bit Memory and offer similar performance to et4000, except for windows accelration.

One nice thing about the ET4000/w32i cards is they come with 45ns interleaved memory, which should theoretically give you faster linear transfers if you have 2MB of memory installed.

I was going to run a few blitting benchmarks to see how this performs against the other ISA cards I have, including a Trident 8900D and Cirrus Logic GD5422-80QC.