IDE to SCSI Adapter - discrete logic - reverse engineering candidate?

[trag]

It's an interesting project, but I'm not getting emotionally invested in any new ideas, until I actually complete some old ones.   I'm mostly focusing on making a living these days.    It's going well, but takes a lot time.  I'm working for Roku and we just got the new 4K TV models to market.   Summer was bussssssssssseeeeeeeeeeeeeeeeeeeeeeeeeee.

As I understand it, Acard's converter cards contain a custom processor of some sort.    It may be a variant on an existing architecture, or it could be something their designers built from scratch.     If the latter, reading out hte attached ROM isn't likely to mean much.   It's probably full of object code for their custom processor.

I'm also not htat convinced that some of their products are ATAPI only or disk only.    The ones I've tried work great for both, and I don't see why they'd go to the work to reengineer it to remove abilties.   But I'm happy to read actual test results that prove me wrong.

My guess is they have one basic adapter design and they've been tacking on different SCSI/PATA/SATA protocols as necessary.

I have a book at home titled something like, The IDE and SCSI Protocols.   Looking at the command sets, I think Gorgonops is right a few posts up.  It's mostly goign to take the tedious work of going through and translating, with probably a few corner cases and special cases where things just don't translate easily or require additional/special handling.

 

[Trash80toHP_Mini]

Let the record show that 4MB/second is the equivalent of a 26.6666x-speed CD-ROM drive. Why do you need something super-fast if your goal is to drive laptop PATA CD-ROM drives, again?

OP was about developing something better than SCSI2SD in terms of throughput using IDE SSDs or IDE adapted SATA SSDs or something like that. Using a laptop slotloader in a SCSI Mac is but a side issue. 4MB/second is about what, Mac II to Quadra era SCSI on the Mac?

 

[NJRoadfan]

Those Adtron SCSI-to-CF adapters used an AMD Am186 combined with an off the shelf NCR SCSI chip, some RAM, and a ROM. Basically a x86 PC with SCSI acting as a bridge for the IDE device. It was common to do that before FPGAs became cheap/fast enough to do the job.

 

[Cory5412]

4MB/second is about what, Mac II to Quadra era SCSI on the Mac?

Think more: "Plus through Beige G3" - at least if you believe Wikipedia's PowerMac G3 page, about the db25 SCSI connector.

Some of the 7000/8000/9000 era machines had a secondary 10MB/sec SCSI channel. It's possible the Beige had two buses as well, but without the internal connector for the "slow" SCSI bus.

Point being, that performance should be "fine" for most things where SCSI is your only option (68k, NuBus PPC).

I'm not entirely convinced that there's something big we're missing out on here in terms of fast SCSI emulation/adaptors.

Every PCI Mac where it's even remotely important has good support for the SIL3112 SATA cards, and other IDE cards, for which there are SATA adapters that are cheap, plentiful, and known working, and for every non-PCI Mac or PowerBook there's typically other solutions that are Good Enough for that system.

This is an extremely well-solved problem, on the Mac.

There are a handful of late surviving RISC UNIX platforms where this is still a problem, and I think that those crowds are just forking over the money for acard adapters, or using SAS HBAs and external disk boxes.

 

[Unknown_K]

Slow compacts don't need fast SCSI HDs and as Cory said anything with a PCI slot has cheap and easy solutions for larger faster HDs (80/160 SCSI cards, SATA cards, or for the Beige G3 SATA to IDE adapters). I even shelled out a few bucks for a UWSCSI 320 PCIE card for one of my G5 towers.

What really needs a better solution would be Nubus equipped 68040 and early PPC desktop/towers. To get past the slow built in SCSI bus you need a ATTO SEIV (if you don't need it to be bootable) or a FWB Jackhammer and prices for those cards are much more expensive then a IDE to SCSI adapter. What would be cool is a small custom made Nubus card (assuming you can find the Nubus connector cheap enough in quantity) with a SATA interface to either a cheap SATA hard drive or smallish SATA SSD both of which are being discarded in the millions now for larger and faster SSD drives in laptops. You can either make the card small with 2 SATA ports or larger with board area to mount a laptop drive with a second connector for a full sized SATA drive mounted in the case. If you can make one for less then the price of a Jackhammer goes for it will sell.

 

[Gorgonops]

It's possible the Beige had two buses as well, but without the internal connector for the "slow" SCSI bus.

I'm fairly certain the Beige only has the *slow* bus. Unlike some of the bigger Beige PPC iron the internal and external connector are on the same chain (like a Quadra or older class machine), and that means the MESH integrated to the Heathrow ASIC is limited to 5MB/s.

Honestly I don't think there's really much future in SCSI-to-"a real device" adaptation. If you really need a high-speed solution and are a little patient you can get one of those ACARD adapters for optimistically what a truly high-speed remake would cost, and coming in right around the prices of those devices are things like the SCSI2SDv6 that are getting into the same speed ballpark as "pretty decent" contemporary-to-the-machine drives. I understand that, yes, there *are* people (literally dozens of them, maybe?) still doing things like video capture and whatnot on that original hardware, but for almost anyone else who just wants to use a machine of that vintage for "retro fun" SCSI2SD-style solutions are performant enough. Is the real audience for "blazing fast, way faster than the original!" drive performance for 1993 vintage machines really big enough to need a new-build device to fill it?

 

[Unknown_K]

I think most of the people who spend $100+ for a Jackhammer or ATTO SEIV are not doing video capture, they just like the speed increase and are hardware lovers in general. Still a SATA Nubus card would require some very specific skills in hardware interfacing and code writing few people still have. 

 

[trag]

I'm fairly certain the Beige only has the *slow* bus. Unlike some of the bigger Beige PPC iron the internal and external connector are on the same chain (like a Quadra or older class machine), and that means the MESH integrated to the Heathrow ASIC is limited to 5MB/s.

The way I remember it (need to check the developer notes) the Beige G3 actually uses a 53C96 or 53C94 cell for its SCSI support, but (and I may be confusing this with the B&W) there's a MESH cell in one of the big I/O chips that doesn't have its signal brought out to pins.   It's just there because some part of the OS expects to find it.

 

[Unknown_K]

The Beige G3 was sold with a third party 68 pin SCSI card if you needed it to run as a server with SCSI HDs. Internal SCSI was fine for scanners which is what was probably popular at the time along with external SCSI removable media that would not need a faster interface (at that time).

 

[Cory5412]

Still a SATA Nubus card would require some very specific skills in hardware interfacing and code writing few people still have. 

Yeah, I don't think anyone's advocating to build a NuBus SATA card. It would be neat to have, of course, but there's already solutions for those Macs.

The point being made is that no NuBus machines need that particular kind of performance boost.

Even the PowerPC Macs in general arguably don't need it and even G3s/G4s running Mac OS 9 arguably can't take advantage of the performance reasonably well, but it's at least an option there, versus trying to build something to max out the second SCSI bus on the higher end 7000/8000/9000 powermac boards.

An affordable SCSI-to-SATA adapter would be very neat, I think a lot of people would like it, but I don't think it's necessary.

 

[Gorgonops]

The way I remember it (need to check the developer notes) the Beige G3 actually uses a 53C96 or 53C94 cell for its SCSI support, but (and I may be confusing this with the B&W) there's a MESH cell in one of the big I/O chips that doesn't have its signal brought out to pins.   It's just there because some part of the OS expects to find it.

According to the dev note both internal and external SCSI are hanging off the Heathrow, there is no separate SCSI chip. (And if you look at the specs for Heathrow in the PDF you'll see that the built-in SCSI is based on MESH; MESH *is* 10MB/s capable in the stand-alone ASIC used in various "thousand-series" machines but it doesn't appear to be so in Heathrow.) External SCSI on the machines that had both was buried in the "Curio" ASIC. Considering what Heathrow has it in it I can't help but wonder if it's actually more of a "Curio" descendant than MESH. (Or maybe the SCSI in Curio is just a brain-dead version of MESH too?)

And yes, the "Paddington" ASIC that was used in the tray-loading iMacs and B&W G3 does apparently have the MESH cell from Heathrow still buried in it but not connected to anything. I've heard vague rumors that the Paddington ASIC actually ended up in some Beige G3s, either in prototypes or possibly even shipped, but I don't know what to make of that. I suspect it's confusion based on some Beige G3 models shipping with 10/100 ethernet *cards* installed. (Main difference between Heathrow and Paddington is the latter has 10/100 ethernet integrated instead of just 10m.)

 

[trag]

 (And if you look at the specs for Heathrow in the PDF you'll see that the built-in SCSI is based on MESH; MESH *is* 10MB/s capable in the stand-alone ASIC used in various "thousand-series" machines but it doesn't appear to be so in Heathrow.) External SCSI on the machines that had both was buried in the "Curio" ASIC. Considering what Heathrow has it in it I can't help but wonder if it's actually more of a "Curio" descendant than MESH. (Or maybe the SCSI in Curio is just a brain-dead version of MESH too?)

Thank you for the links.    You are indeed correct.  The Hardware Developer Note says that Heathrow contains a MESH controller.   Color me surprised.    I thought/expected it was basically CURIO with extra stuff brought inside.   Although, CURIO was an AMD chip, so maybe Apple didn't have access to all the rights/designs, or something and had to make Heathrow from scratch.

On the other hand....

Wandering into speculation country....

The separate Fast SCSI bus first showed up on the NuBus PowerMac 8100 and 9150.    It was based on the 53CF96 which was made by a number of folks.   I'm not sure where the design originated, maybe NCR.    But AMD also sold one and I think Zilog might have as well.   NCR mutated into Symbios (?) and later LSI Logic.

The separate Fast SCSI bus on the next generation, the PCI PowerMacs, used the MESH controller, in exactly the same package as the 53CF96.    I strongly suspect that the MESH is just a licensed version of the 53CF96.  Or, perhaps, Apple's home rolled version to avoid licensing fees?  In the volumes they were using, it may have paid to roll their own.   One of these days, I'm going to replace a MESH with a 53CF96 or vice versa and see what happens.

Meanwhile, the CURIO chip which first appeared in AV Quadras and was used in both the NuBus and the PCI PowerMacs, contained a 53C94 or 53C96 cell.  In all those machines the CURIO chip provided the slower internal/external SCSI bus, as well as serial ports and ethernet.

The 53C94/96 is closely related to the 53CF96.

So, I wonder, if Heathrow was an Apple design, and they put a 53C94/96 in Heathrow, would it be inaccurate for them to claim that Heathrow contains a "MESH-based SCSI controller".

I am also now curious to see what SCSI Probe says about the Beige's built in SCSI bus...

 

[Gorgonops]

I vaguely recall the Wikipedia page for the 53C9x series claims that MESH is indeed a lightly cooked version of that chip, so, sure, in a roundabout way Apple may well be being both truthful and elusive with the claim that the SCSI built into Heathrow is "MESH Based" even though it's actually a lower-spec core from the same licensed family.