dionb wrote on 2024-04-11, 11:20:

One of my (many...) long term project ideas is to do some benchmarks of a couple of cached controllers (and the fake cache Hornet VL-230 😜 ) vs uncached, and to do that with a number of different storage devices (very old&slow HDD, CF card and something truly fast) and on some different systems (386DX-16 to Pentium 133 with 32b VLB, and a 486SLC2 with 16b VLB). My suspicion is that with period correct (i.e. old & slow) storage these things will turn out to actually be pretty useful and our 'paper tiger' opinion is vs much faster storage which would not have been available when they were new. But that would take a lot of time I don't have, so for now it's just an idea.

I'll get there at some point. It's just that the cached cards are harder to come by. And that I've gotten side tracked for the past come months with about 6 nice found PCs to refurb, rewriting my speedsys report parser, and that my kids are not old enough to do the benchmarking. Yet. I'll get back to the project as soon as I finish submitting photos of all my cards on theretorweb.

pshipkov wrote on 2024-04-11, 07:20:

Not sure this is a factor that actually affects the SiS 496/497 onboard local storage controller performance. Most SiS 496/497 b […]
Show full quote

Not sure this is a factor that actually affects the SiS 496/497 onboard local storage controller performance.
Most SiS 496/497 based boards vary from intermediate to excellent IDE performance.
It is much better than UMC UM8881/UM8886 based ones in general.
It takes a good VLB IDE or late pci udma-66 or higher card to outdo it.

Alaris Cougar II is hopelessly slow motherboard, but the onboard IDE is pretty good, yet the best it can do as part of this mobo is meet SiS 496/497 at the lower end.
Wonder why Michal used it for comparison …

I've made these bencharks years ago, but roughly, these were my results in Speedsys Buffered read:

420EX: 5.5 MB/s
UMC8886: 3.3 MB/s
SIS496/7: 13 MB/s
Finali: 13 MB/s

What real life workload would cause a system sharing VLB IDE with PCI to perform poorly? Games usually read data in bursts and during loads won't really push a PCI graphics card much. Maybe something with networking?

dionb wrote on 2024-04-11, 11:20:

pshipkov wrote on 2024-04-06, 02:56:

The cached transfer speed of BT-410A is epic.
Non-cached speed is the deep opposite.
Another caching paper tiger, but i admit these complex assemblies are quite interesting.

One of my (many...) long term project ideas is to do some benchmarks of a couple of cached controllers (and the fake cache Hornet VL-230 😜 ) vs uncached, and to do that with a number of different storage devices (very old&slow HDD, CF card and something truly fast) and on some different systems (386DX-16 to Pentium 133 with 32b VLB, and a 486SLC2 with 16b VLB). My suspicion is that with period correct (i.e. old & slow) storage these things will turn out to actually be pretty useful and our 'paper tiger' opinion is vs much faster storage which would not have been available when they were new. But that would take a lot of time I don't have, so for now it's just an idea.

I think this was the target of such controllers. If you for some reason have or want to use an old hard drive, then they will make it faster. But if you use a late IDE drive with a 486 then such controller would become a bottleneck.

Here is a page with programming info for these three chips that are really the same chip: https://www.ardent-tool.com/Reply/Appian_ADI2.html

1. Appian Technology, Inc. (ADI/2)
2. Cirrus Logic, Inc. (CL-PD7220)
3. Adaptec (AIC-25VL01)

In a similar case of the same chip sold with multiple brand names, I'm looking at this: http://vogonsdrivers.com/getfile.php?fileid=1683

1. DTC 2278VL ... PIO Mode 2 Max ... uses the DTC 805 chip
2. DTC 2278S ... PIO Mode 4 ... uses the DTC 803 chip
3. DTC 2278E ... PIO Mode 4 ... uses the DTC 803 chip

DTC likes to put stickers on their chips so you can't see what they are, but I got a chance to pull the stickers off some boards.

DTC 2278VL has a chip labeled DTC 805, but the DTC 2278E and DTC 2278S don't have a DTC 803 chip, they have Atronics 2015PL chips. So I'm starting to think that there is no "DTC 803" chip. They may have grabbed something off the shelf and put a sticker on it.

GL1zdA wrote on 2024-04-11, 14:40:

420EX: 5.5 MB/s UMC8886: 3.3 MB/s SIS496/7: 13 MB/s Finali: 13 MB/s […]
Show full quote

420EX: 5.5 MB/s
UMC8886: 3.3 MB/s
SIS496/7: 13 MB/s
Finali: 13 MB/s

What real life workload would cause a system sharing VLB IDE with PCI to perform poorly? Games usually read data in bursts and during loads won't really push a PCI graphics card much. Maybe something with networking?

Before I started, I ran through a bunch of boards to make sure I had something that was in the right place performance wise using a lion 3+ jumpered to the fastest timings after the first few tests.

I used the same CF for all the tests.

INT13h:
AMI = the AMI BIOS on the board
AWARD = the AWARD BIOS on the board
DTCY2K = https://www.vogonsdrivers.com/getfile.php?fileid=2095
EZ Drv = https://www.vogonsdrivers.com/getfile.php?fileid=1900
Vendor = https://www.vogonsdrivers.com/getfile.php?fileid=1996
XUB = Xtide Universal Bios

RW is the ReadWrite Cycles as jumpered on the card. https://www.vogonsdrivers.com/getfile.php?fileid=1996

1Board          |Int13h|RW|CPU  |Score| Seek|BufRd|Verfy|LinRd|Notes
2PC CHIPS M919  |AMI   |36|AM133|50.05| 0.77| 1257| 9391| 1257|Card Jumpered for RW=36 Cycles
3PC CHIPS M919  |XUB   |36|AM133|50.05| 0.72| 1298| 9589| 1297|Card Jumpered for RW=36 Cycles
4PC CHIPS M919  |DTCY2K|36|AM133|51.55| 0.53| 1350| 9959| 1358|Card Jumpered for RW=36 Cycles.
5PC CHIPS M919  |AMI   |5 |AM133|50.05| 0.52| 3291| 9404| 3293|Turns out that RW=5 Cycles is fast, 36 Cycles was slow
6PC CHIPS M919  |XUB   |5 |AM133|50.05| 0.47| 3611| 9601| 3592|XUB address was shadowed on this board
7PC CHIPS M919  |DTCY2K|5 |AM133|51.55| 0.28| 4002| 9987| 4079|This ROM causes a higher CPU score in SpeedSys sometimes.
8Shuttle Hot419 |AMI   |5 |POD83|61.33| 0.55| 3008| 9355| 2969|POD83 was slower, so is raw clock speed important?
9Shuttle Hot419 |AMI   |5 |AM133|48.12| 0.47| 3746| 9479| 3677|
10Shuttle Hot419 |DTCY2K|5 |AM133|48.12| 0.27| 4542| 9997| 4521|
11Shuttle Hot419 |XUB   |5 |AM133|48.12| 3.32|  508| 5004|  494|Forgot to shadow XUB
12Lion Nice Green|AMI   |5 |AM133|38.19| 0.39| 3335| 9867| 3304|
13Lion Nice Green|DTCY2K|5 |AM133|38.19| 0.28| 4905|10030| 4868|
14Lion Nice Green|XUB   |5 |AM133|38.19| 2.93|  563| 5223|  548|Forgot to shadow XUB
15FIC VIP        |AWARD |5 |AM133|44.52| 0.35| 3235| 9966| 3219|Board has a Beta Y2K ROM is funky
16FIC VIP        |DTCY2K|5 |AM133| Fail| Fail| Fail| Fail| Fail|Boots very slowly, unstable
17FIC VIP        |XUB   |5 |AM133| Fail| Fail| Fail| Fail| Fail|Boots very slowly, unstable
18WinTech MV008  |MRBIOS|5 |AM133|48.29| 0.41| 3336| 9743| 3309|
19WinTech MV008  |DTCY2K|5 |AM133|48.29| 0.27| 4955|10008| 4934|DTC ROM w/ Shadowing
20WinTech MV008  |DTCY2K|5 |AM133|48.29| 0.33| 4927| 9783| 4879|DTC ROM w/o Shadowing, only a little slower
21WinTech MV008  |XUB   |5 |AM133|48.29| 4.14|  416| 4368|  404|XUB w/o shadowing is way slow
22WinTech MV008  |XUB   |5 |AM133|48.29| 0.44| 3957| 9622| 3914|But it was respectable with shadowing
23Lion Nice Green|DTCY2K|5 |AM133|38.19| 0.28| 4905|10030| 4868|
24Lion Nice Green|Vendor|5 |AM133|38.19| 0.33| 5884| 9843| 5781|Driver reports "Mode 2"
25WinTech MV008  |DTCY2K|5 |AM133|48.29| 0.27| 4955|10008| 4934|
26WinTech MV008  |Vendor|5 |AM133|48.29|  0.3| 6014| 9816| 5950|Driver reports "Mode 2"
27WinTech MV008  |EZ Drv|5 |AM133|48.29| 0.26| 6861|10021| 6792|32bit access enabled in the EZ drive config
28WinTech MV008  |EZ Drv|5 |AM133|48.29| 0.27| 6186|10020| 6139|32bit disabled

dionb wrote on 2024-04-11, 11:20:

One of my (many...) long term project ideas is to do some benchmarks of a couple of cached controllers (and the fake cache Hornet VL-230 😜 ) vs uncached, and to do that with a number of different storage devices (very old&slow HDD, CF card and something truly fast) and on some different systems (386DX-16 to Pentium 133 with 32b VLB, and a 486SLC2 with 16b VLB). My suspicion is that with period correct (i.e. old & slow) storage these things will turn out to actually be pretty useful and our 'paper tiger' opinion is vs much faster storage which would not have been available when they were new. But that would take a lot of time I don't have, so for now it's just an idea.

Right on. I often forget to bring this point up.
With screechy ancient HDDs the caching controllers make much more sense.
Later tech, not even present day CF cards, but let's say HDDs from the end of the 90ies make them largely irrelevant.

pshipkov wrote on 2024-04-12, 03:26:

With screechy ancient HDDs the caching controllers make much more sense.

Imo those only made sense when there was no other way of adding normal ram into the computer. Smartdrive started shipping in late 1988.

rasz_pl wrote on 2024-04-12, 12:07:

pshipkov wrote on 2024-04-12, 03:26:

With screechy ancient HDDs the caching controllers make much more sense.

Imo those only made sense when there was no other way of adding normal ram into the computer. Smartdrive started shipping in late 1988.

IN 1988, 3600RPM HDDs that sounded like marbles in a mason jar were used on computers with ISA controllers . There was some I/O headroom left in the ISA bus at that point, but it ran out pretty soon. I think it ran out for me when I bought a 17ms WD Caviar 280 drive. The new high tech drive sounded like ball bearings in a baby food jar (woah! The future!) and felt a lot faster than the 30ms clunkers that ruled the previous year. Pretty much all my extended memory went to smartdrv at the time while in DOS. Smartdrv cache was way faster than cache hanging off the ISA bus could have been since I had a 20Mhz CPU. Anyway, I don't think I experienced an I/O jump like I got from that WD drive until I put a 10ms 5400 rpm quantum fireball drive on a PCI bus years later. That was for a number of reasons.

When I entered the VLB age in 1993, my video performance was so much faster, but I remember not seeing much of a change in storage. Why wasn't it any faster? Was it my 2 year old 200MB maxtor drive? Was it the cheap controller? Was it the Award INT13H handler? Probably all of those things. I had to dig through usenet to find hardware tips and PIO-4 was something I read about in Computer Shopper that might show up next year. Adding a weird IDE driver from rando-BBS was a risk . It might corrupt your drive. It was my only computer and I was writing software on it and my only backups were on floppy. Catastrophic failure was a possibility. The only two certainties about adding an IDE device driver to my config.sys were that 1) I was going to lose a UMB and 2) DOOM wasn't going to play any better. There were so many clear bottlenecks that needed $$$ to be addressed before worrying about the storage subsystem. And why get a new hard drive if I had enough space? That drive cost me $600 in 1991 and it was still big enough for DOS & OS/2. Real world performance limitations on hard drives could be mitigated by adding another SIMM. OS/2 wanted another 8MB anyway....

But today, with the benefit of hindsight, cheap high speed storage, and if you are patient, a wide variety of VLB controllers available on ebay at a fraction of their inflation adjusted 1994 prices , I'm enjoying looking for the storage combinations that could have worked better back in the day.

Many here relate to what you wrote, probably with some deviation in the time period.
For example:
Until early 90ies i perceived the computer as an opaque entity.
That progressed into a modular entity.
Then morphed into a set of individual components and subsystems (both software or hardware), that are subject to tweaks and optimizations.
Around '94 i was already an eager consumer for computer hardware ... with empty wallet, so my contribution to the global economy was very modest.

Just wanted to say that I have been watching this thread with much fascination despite having a VLB IO controller in only one retro system..

Anyone have any idea what the relationship is between the Cirrus Logic CL-PD7232-QC-AA chip and the XILINX chip on this Vesa extender for this device?

https://th99.dosreloaded.de/c/U-Z/20796.htm

Well that's a 'king puzzle. Seems to be an FPGA as glue logic for the accelerated IDE

BitWrangler wrote on 2024-04-16, 15:18:

Well that's a 'king puzzle. Seems to be an FPGA as glue logic for the accelerated IDE

This really is a King Puzzle.

The CL-PD7232 that is on this board has a similar number to the Cirrus Logic CL-PD7220 IDE controller, which is a rebranded Appian ADI/2.

I tried to search for CL-PD7232-QC-AA chip to find out what kind of device it is and google gave me the suicide prevention hotline.

The FPGA is likely just bog standard logic condensed into a smaller package. Pretty common on boards with limited space.

It is a CPLD there, something on the level of a much souped up 20V8 PAL chip. Judging from the connections which are mostly IDE control signals, it perhaps helps with the DMA modes.