Trifecta (rarest amiga hdd controller)

[alexh]

Not mine, ive been watching for one for years and years and never seen one but its outside my budget

https://rover.ebay.com/rover/0/0/0?m...2F132843666126

They are the rarest, one of the last, and arguably the fastest zorro 2 HDD controller.

[patrik]

Any details on exactly how fast fastest is and what makes it so fast?

[Paul_s]

Cool, didn't know they made one for the 2000 series. A500 version is probably based around the same design?

http://icd.com/amiga/index.html#trifecta

[alexh]

Originally Posted by patrik

Any details on exactly how fast fastest is and what makes it so fast?

If i had one I'd tell you but zorro ii is about 3.38MB/s theoretical max and trifecta was in magazines quoted at 10MB/s with phrases like "hidden DMA transfers"

[AMike]

Are you sure about that - most SCSI II controller reaches 7MB/s and more. For examples Blizzard 1230 or BlizzardPPC.

[alexh]

Originally Posted by AMike

Are you sure about that

Am I sure the magazine articles say that? Yes I'm sure. As does the ICD website. I'm sure some of it is marketing crap but i always wanted one to see for myself.

Originally Posted by AMike

Are you sure about that most SCSI II controller reaches 7MB/s and more

Am I sure that Zorro II SCSI controllers are limited to 3.8MiB? Yes.

The SCSI II controllers you list are not Zorro II they are connected directly to the 680x0 bus (i.e. PIO) and are only limited by the I/O timing and the CPU speed.

[patrik]

Originally Posted by alexh

If i had one I'd tell you but zorro ii is about 3.8MB/s theoretical max and trifecta was in magazines quoted at 10MB/s with phrases like "hidden DMA transfers"

Ok, even that fast, very interesting!

A reasonable guess is the 10MB/s is for transfers to/from its onboard memory, which would definately be "hidden" from the rest of the system.

I think the GVP cards does something similar to this, its not anywhere near as exciting as their memory is not anythwere near 10MB/sec and they are bottlenecked way lower at the SCSI end, like 2.3MB/sec.

[trixster]

Bboah and hardware database says it does not work in zorro III machines

[talybont]

Originally Posted by AMike

Are you sure about that - most SCSI II controller reaches 7MB/s and more. For examples Blizzard 1230 or BlizzardPPC.

Neither of those are Zorro II cards though.

[Toni Wilen]

Only older non-DMA GVPs (pre-Series II) had buffer RAM.

Trifecta does not seem to have any buffer RAM but because it has FPGA, it probably can prefetch/buffer multiple bytes internally so in theory it can do short (and useless) 10M/s transfers but Z2 bus is still the bottleneck..

Trifecta ROM dump is also still missing.

Bboah and hardware database says it does not work in zorro III machines

This could be also caused by some stupid code in boot ROM. It may not be hardware issue.

[alexh]

Originally Posted by Toni Wilen

Trifecta ROM dump is also still missing.

The vendor has agreed to dump the ROM for WinUAE before sale and has pulled the Auction.

They are an experienced Amiga user and say they can easily transfer files to and from the Amiga with the Trifecta plugged in.

Unfortunately I do not know how they would dump the ROM under AmigaOS?

I've seen a thread where they used TrackDOS to dump $E90000 -> $E98000 and another thread where they dumped $f00000 -> $f7ffff and $ea0000 -> $eaffff?

How do you determine the start address / size of the BOOTROM? Or are they all fixed location/sizes?

[Toni Wilen]

Great

Usually it is E90000 to E9FFFF if it is common 64k Z2 IO board. WB 2.0+ disk tools/showconfig or sysinfo can be used to find the board IO space. (if 128k, it is EA0000 to EBFFFFF) $f00000 is usually only used if it can boot under KS 1.2.

Note that some devices will hang if IO space is accessed without active IO (especially if it has pseudo DMA like functionality). In this case it is best to run this: http://www.winuae.net/files/b/expdump.zip and include the output. (photo is enough). Most commonly ROM is in upper half of IO space but it isn't guaranteed. (and bottom half is IO).

Also expdump output is needed if autoconfig data is not located in ROM data. Sometimes it is in ROM, sometimes in logic chips.

EDIT: Or simply read the ROM chip using ROM reader.

[patrik]

Originally Posted by Toni Wilen

Trifecta does not seem to have any buffer RAM but because it has FPGA, it probably can prefetch/buffer multiple bytes internally so in theory it can do short (and useless) 10M/s transfers but Z2 bus is still the bottleneck..

The onboard RAM could be faster than what is required for functioning as a plain Zorro2 RAM expansion, allowing the DMA controller to do faster transfers to/from it than what would be possible over the Zorro2 bus.

[Toni Wilen]

Originally Posted by patrik

The onboard RAM could be faster than what is required for functioning as a plain Zorro2 RAM expansion, allowing the DMA controller to do faster transfers to/from it than what would be possible over the Zorro2 bus.

It is possible. ROM driver disassembly will reveal if there is some "board internal" DMA.

[patrik]

Originally Posted by Toni Wilen

It is possible. ROM driver disassembly will reveal if there is some "board internal" DMA.

Does there necessarily need to be any special driver code for transfers to/from the trifecta local RAM? Shouldn't the hardware on the card access whatever address is given to the driver? So if given an address for its local RAM, never utilize the Zorro2-bus, but of course if given an address to RAM outside the board - act as a Zorro2 DMA-busmaster.

Imagine the only scenario where I think there would be a great benefit of fast transfers to local trifecta RAM: A 68000 machine with the majority of system RAM on the trifecta card.

A 68000 has no cache so it is slowed down a lot when locked out from memory access during normal DMA transfers on the Zorro/68000-bus (which say an A2091 would do). If the trifecta is able to interleave 68000 access to its onboard RAM with accesses from its onboard DMA controller, the DMA transfers would be basically free.

In that case, as most memory allocated would from the trifecta local RAM, its DMA controller would in most cases access its local RAM.

[Toni Wilen]

Surely it would have been marketed as fully DMA capable controller (instead of mysterious "hidden dma") if it also could do DMA transfers to/from chip ram etc..

[patrik]

Originally Posted by Toni Wilen

Surely it would have been marketed as fully DMA capable controller (instead of mysterious "hidden dma") if it also could do DMA transfers to/from chip ram etc..

Point taken!

[patrik]

Originally Posted by alexh

If i had one I'd tell you but zorro ii is about 3.8MB/s theoretical max and trifecta was in magazines quoted at 10MB/s with phrases like "hidden DMA transfers"

To be nitpicking, for a PAL A2000, it would be:
Motherboard oscillator frequency = 28.37516MHz
Zorro2(68000)-bus frequency = 28.37516MHz / 4 = 7.09379MHz = 7093790Hz
Zorro2(68000)-bus Bytes per transfer = 16bits / 8 = 2Bytes
Zorro2(68000)-bus cycles required for one transfer = 4

Theoretical bus speed = (7093790 * 2Bytes) / 4 = 3546895Bytes/sec = 3546895Bytes/sec / 1048576 = 3.38MBytes/sec

[mark_k]

Originally Posted by Toni Wilen

It is possible. ROM driver disassembly will reveal if there is some "board internal" DMA.

I've been looking at the Trifecta driver code.

I'm pretty sure it can use DMA, but only to on-board RAM. The driver (which can handle up to four Trifecta cards) seems to be very careful about associating the RAM on each board with the SCSI/IDE on that board.

Has the EPROM been dumped yet?

[alexh]

Not heard anything I'll ask again