Classic IDE DMA possible?

[SHADES]

Hi all,

I'm looking at the Zorro 3 specs and to me it looks like there is the ability to do DMA access of up to 50 MB/s non burst.

Is there some other reason as to why no one want's to manufacture a DMA IDE interface? One of the DMA modes for IDE is ATA33 which is well within the DMA spec.

I believe the AMIGA 3000 is limited to about 20 MB/sec due to a slower clock speed, the faster AMIGAs should be able to handle it though.

[Matt_H]

It probably could be done, but would the development cost make it worthwhile?

[Piru]

to me it looks like there is the ability to do DMA access of up to 50 MB/s non burst.

Unfortunately ZIII can't come even close to that theoretical maximum.

[mboehmer_e3b]

Hi,

> I'm looking at the Zorro 3 specs and to me it looks like
> there is the ability to do DMA access of up to 50 MB/s non
> burst.

Zorro III DMA is tricky, as both last Buster versions (rev 9 and rev11) have quite some quirks in it which must be taken care of. It is possible, but needs quite some development effort.

Moreover the Zorro III burst modes (/MTACK and /MTCR signals) do not work, and even if they would - most Zorro III cards, and CPU cards on A3000 / A4000 do not support burst, which is limiting bandwith anyhow.

Michael

[Crumb]

@mboehmer_e3b

Will you add real DMA to prometheus? And when do you plan to release the update?

[Karlos]

Piru wrote:

to me it looks like there is the ability to do DMA access of up to 50 MB/s non burst.

Unfortunately ZIII can't come even close to that theoretical maximum.

Surely speed alone is not the only motivation. Even if it managed 10MB/s, provided the CPU usage was nice and low, it would be worth it.

[SHADES]

Hmm, 20 MB/sec on A3000 was acuivable with a slower clock, ATA33 seems to me not an impossible target to achieve. A faster external clock could drive it. Doesn't a4000 have that ability?? Didn't know that burst modes didn't work! I was sure my Fastlane Z3 did burst and it can sustain 11MB/sec no problem. I know the hypermemory card used burst modes.www.amiga-hardware.com Now i'm thinking, I haven't seen any 100Mbit ethernet cards. I wonder why burst isn't working! Time to visit Haynie's page. About being worth while, you're probably correct there. Most H/w is failing now. I'm about to try fix my 4000 again which is grey screening and not booting. I have 3 cpu cards and 3 mainboards to try scavenge parts from. Hoping one is still able to function.

[Lemmink]

I think the major liminting factor here is cost not the technical feasility. Look at how much a FastlaneZ3 cost when it came out and at that time you could spread developmentcost over much more sold units then you could today.
Zorro3 is much, much much more complex then Zorro II from what I understand thats why even in the high time of the amiga only a few Z3 cards were made.
I would love to have an DMA Z3 IDE card or a Z3 100 mbit ethernetcard but at an estimated retailprice of 250 EUR (just an approximation) how much units do you think you could sell ?

[SHADES]

"how much units do you think you could sell ?"

This is a very valid point. I thoiught with todays mass chip manufacture, things like DMA IDE chips would be cheap, but then there's logic to Zorro 3 which may not be so that makes sense. Still they produce IDE cards that don't do DMA still, then again, there's much less logic needed to do PIO modes.

I'd say you are very correct here. Most mainboards are dying, why bother. I wonder if IDE cards on the Mediator would work properly

[Lemmink]

I wonder if IDE cards on the Mediator would work properly

Mediator can only do PCI <-> PCI DMA Transfers. Shoveling the Data from the GFX-Mem to FastMem through the 8 MB window by the CPU would make the whole DMA part worthless.

The only way for DMA for classic is either Zorro or G-Rex.

[mboehmer_e3b]

> I was sure my Fastlane Z3 did burst and it can sustain
> 11MB/sec no problem. I know the hypermemory card used
> burst modes.

Don't mix up memory burst on the card itself and burst access on the Zorro III side. The first one is up to the designer and completely decoupled from the bus interface itself.

Burst mode on Zorro III (using /MTACK and /MTCR) does not really work. If it worked you could do one Zorro III cycle and access 64 longwords, doing only one Zorro III address phase and 64 data phases.

> Now i'm thinking, I haven't seen any 100Mbit ethernet
> cards. I wonder why burst isn't working!

Well, 100MBit on Zorro III is possible - believe me :-)

Michael

[mikeymike]

5 years ago, yes, it probably would have been generally beneficial.  Now?  It's not like there's an expanding classic userbase anymore.

[SHADES]

"Burst mode on Zorro III (using /MTACK and /MTCR) does not really work. If it worked you could do one Zorro III cycle and access 64 longwords, doing only one Zorro III address phase and 64 data phases."

Is this a Buster 11 fault? from the looks of the specs, everything is set out quite nicely.
Haynie link to burst I also found this which was instruction on how to implement the zorro3 spec properly. Zorro 3 design How to design zorro 3 card. Looks like the use of /FCS is how to get in to the burst cycle.
I haven't seen anythuing on buster 11 yet, if it is a buster problem, I'm still looking. I guess it's irreleveant now, Zorro III is ancient compared to PCI express.

The mind really boggals with the what IFs on what hardware would be like now if Ami H/W was successful in market penetration, such a shame

[alewis]

About the same as VL-Bus.... probably. This was, imho, a superior local bus, and still is (40MHz+ bus speed vs PCI's 33MHz, for example). But it was up against Intel's muscle.

But this is all to similar to the endless "what would you like to see in a modern Amiga/what would the Amiga of today be like". In both cases, it would *not* be a classic Amiga, and unless had the same degree of hardware architecture (custom chipset) and an OS that exploited it, would not have a lot in common with an Amiga.

A thought. Whilst 100mbit ethernet would be nice, it would be rather pointless: Unless one had an ultra-wide SCSI controller (or a really good DMA capable SCSI-2 controller - and few were - with a good SCSI drive), what would be the point? The 8-10MB (on a good day) throughput of the ethernet (sans overhead) would be bottlenecked by the [relatively] slower hard disk.

And given the, on average, small size of Amiga files, would you see a real world increase in file transfer???