Here you go 😀

Motherboard is Opti495SLC, 256KB cache, 8MB RAM, Ti486DLC-40MHz + Cyrix 487 FPU.

VLB S3 864 2MB EDO

3DBench: 26.6
PCPBench: 5.8
DOOM: 5905

SysSpeed
CPU: 20.97
Vesa: 11891KB/s
Mem: 23.74MB/s

Winbench 3.11 (Windows 3.11):
640x 8bit: 28233K
640x 15bit: 20943K

ISA ATi Mach32 2MB EDO

3DBench: 23.4
PCPBench: 5.6
DOOM: 6641

SysSpeed
CPU: 20.97
Vesa: 5622KB/s
Mem: 23.74MB/s

Winbench 3.11 (Windows 3.11):
640x 8bit: 12194K
640x 15bit: 6923K

Interesting. Thank you. Do you have a faster ISA graphics card to compare, such as something based on a Cirrus Logic GD5434, Tseng ET4000/w32i, or Match64? Could you run the ISA bus at 10 MHz?

Tests were done with best settings - ISA/4, cache 2-1-1-1,...

I have CL-5434 ISA, ET4000 ISA, ET4000/w32i VLB. Mach64 ISA is on the way. No luck with finding ET4000/w32i ISA...

S3 864 is just a bit faster then ET4/w32i. On ISA these are all higher end graphic cards and shouldn't be any real difference between them - ISA is the limit. Maybe in Windows with cards that have VRAM and memory interleaving.

When I ran DOOM (Ti486SXL) on a Mach64, I got 10.8 fps. When I ran it on a ET4000/W32i, I got 11.4 fps. This difference alone is enough to close the gap between your VLB and ISA results.

Are you willing to run the benchmark comparison with your GD5434? Some people have claimed this is the fastests ISA DOS card.

I'll redo benchmarks when I get more ISA video cards - namely ET4000/w32i 2MB and some VRAM ones. Hopefully soon 😉

Which 386 chipset is on your motherboard? Which motherboard?

Interestingly, on other 386 motherboards, the GD5434 had slightly better results compared to the ET4000/w32i. This is why I'm interested in your GD5434 vs. VLB results.

In my 1st post there is a link to the motherboard description.

When I tested this hybrid 386/486 board I had only a few higher end ISA cards. Now I'm missing just a few. If everything goes well then I'll get ATi Mach64 today. Don't know if its EDO or VRAM based yet.

Overall anything VLB will be faster then ISA. Of course the margin won't be huge as 386 is also too slow. I might even do "all-in" benchmarks with this board from 386DX/486DLC to 486DX/2 80MHz - I don't have voltage regulator adapter so anything faster, like AM5x86-133 isn't possible.

If I get some free time, I might do preliminary testing with CL-5434 VS Mach64 (if I really get it today).

---
Edit: typo

feipoa wrote:

When I ran DOOM (Ti486SXL) on a Mach64, I got 10.8 fps. When I ran it on a ET4000/W32i, I got 11.4 fps. This difference alone is enough to close the gap between your VLB and ISA results.

Don't say that! You'll get sunaiac all over you!

Neither the S3 864 nor Mach64 GX support EDO DRAM. These cards use FPM DRAM.

Not sure if the 495SLC is a good testing platform. VLB support is flawed, which is why the 495SX replaced it. However, my testing showed 495SX wasn't much better. What I'd really like to see are results from hybrid VLB boards based on chipsets from other manufacturers.

What do you mean with: flawed VLB support? In what cases?

The difference between the two chips is the VESA bus implementation. The 495SLC asserts the VESA RDY_RTN signal in the same cycl […]
Show full quote

The difference between the two chips is the VESA bus implementation.
The 495SLC asserts the VESA RDY_RTN signal in the same cycle as the
VESA target asserts LRDY signal while the 495SX asserts the signal in the
next cycle. This means that each VESA bus cycle is one clock slower
on the the SX than on the SLC. So in the worst case an SX write cycle is
50% slower and a read cycle 33% slower than on the SX.

I think the reason for this is that the SX has a buffered VESA bus, while
the VESA bus on the SLC is unbuffered. This means that the SX can drive
a higher load than an SLC, this is why SX motherboards have normally 3
VESA slots and SLC only 2.

So SX based boards can drive a higher load while SLC boards are faster.

I found the differences the hard way as our VESA card would work in an SLC
but not on SX based boards. After hooking up a logic analyser I found the
subtle difference.

I also have an first generation ET4000/W32 card which has such a high load
that it would not work together with our VESA card in the SLC. Our VESA board
alone or the ET4000/W32 alone worked but not together. The SX board had no
trouble in driving both VESA boards.

So if you want to play safe get an SX, if you want the the fastest VESA
an SLC.

Jean-Paul

I went back and had a look on google groups. It seems SLC is not really buggy...it's actually faster, but less compatible.

Good 😁

I'll give the board another shoot when I gather all the different parts and do some systematic tests/benchmarks (386/486, ISA/VLB).

I usually don't use VLB I/O controllers as there isn't much difference from ISA ones.

What I'd really like to know is how the 495XLC is different than the SLC and SX. I have only ever seen it on those super compact 386 boards that lack VLB slots.

Datasheet for the 495XLC is available. Unfortunatley, the SX and SLC versions are not.

http://www.datasheetarchive.com/dl/Datasheets … HI000109887.pdf

XLC also seems to support local bus.

kixs wrote:

Good 😁

I'll give the board another shoot when I gather all the different parts and do some systematic tests/benchmarks (386/486, ISA/VLB).

I usually don't use VLB I/O controllers as there isn't much difference from ISA ones.

This is not True, a decent Controller will allow you enabling DMA Modes. And faster Disk Speed.
I don't know if this is supported too on 486DLC/386 VLB Systems.
Real 486 will benefit from it.

matze79 wrote:

This is not True, a decent Controller will allow you enabling DMA Modes. And faster Disk Speed. I don't know if this is supporte […]
Show full quote

kixs wrote:

Good 😁

I'll give the board another shoot when I gather all the different parts and do some systematic tests/benchmarks (386/486, ISA/VLB).

I usually don't use VLB I/O controllers as there isn't much difference from ISA ones.

This is not True, a decent Controller will allow you enabling DMA Modes. And faster Disk Speed.
I don't know if this is supported too on 486DLC/386 VLB Systems.
Real 486 will benefit from it.

In addition that, many VL multi I/O IDE controllers have their own BIOS (no 540MB limit, but 8GB limit mostly) and have two channels (4 x ATAPI devices). ISA controllers with these two properties are definitely rare.

I'm sure it's not true. But I've tested with different VLB 486 motherboards and different VLB I/O controllers with CF2IDE adapter and I get around 2500kB/s transfer rate... that I also get with an ISA controller 😕

kixs wrote:

I'm sure it's not true. But I've tested with different VLB 486 motherboards and different VLB I/O controllers with CF2IDE adapter and I get around 2500kB/s transfer rate... that I also get with an ISA controller 😕

Yup, the theoretical maximum of an ISA slot is about 16 MB/s (8 MHz, 16-bit).
So if your disk isn't anywhere near that fast, you wouldn't notice much of a difference.
The theoretical maximum of a VLB slot at 33 MHz would be about 133 MB/s, so you could see a big difference if your disk is fast enough.
I think the biggest differences would be with caching VLB controllers. Even with slow disks, they can perform very fast transfers from cache to the host.

I use the same CF card for all my retro rigs - 8GB Sandisk Ultra II. On Pentium boards it gets more then 10MB/s. On 286 I get around 2MB and 386 ISA and 486 VLB around 2.5MB. Not sure how it fares with 486 PCI onboard IDE, will try it sometime.

kixs wrote:

I'm sure it's not true. But I've tested with different VLB 486 motherboards and different VLB I/O controllers with CF2IDE adapter and I get around 2500kB/s transfer rate... that I also get with an ISA controller 😕

If your VLB controller doesn't have a BIOS, unless you load a driver at startup it is stuck in PIO mode, greatly limiting its speed. Look for you controller's drivers and see if it makes a difference.