SX-1 + FMV needs a specific scsi.device ?

[Rochabian]

I'm quite sure the Gayle Emulation in SX-1 is not as perfect as in SX-32.
I own the two and never had problems with memory/disks on SX-32.
I have 2 SX-1 ( Rev 2.0 and Rev 3.0) and they have same problems with memory/disk/fmv. Many people complains about SX-1 stability with 8Mo of RAM (or 4 + FMV ), but SCSI.device 43.45 from BoingBag 2 fix the problem with 8Mo but disable the use of FMV within Workbench. As there is a software fix for 8Mo, i hope there could be a software fix for 4Mo + FMV. Problem is that we don't know what the 43.45 fixed to stabilize 8Mo......

@Toni
About WinUAE, i've tried to use Z2 Fast RAM, but with Z2 in autoconfig (00200000 to 005FFFFF), it say that my drive has a cheksum error on boot
Then if i set the real adresses (00600000 to 009FFFFF), the memory board is not shown at all, it boots to my HDD normally but in workbench i have no fast. And in the expansion board diagnostic from the CD32 boot, the card is not seen.

[alexh]

I can only imagine that it is timing related (a term used in digital logic to describe the amount of logic between registers and the time it takes signals to go between them relative to the clock speed) in the CPLD/FPGA which controls the IDE.

It is likely that IDE newer devices, particularly solid state storage are a lot faster to change their I/O than older devices which shouldn't be a problem if the timing was correct but might be an issue if it was marginal.

It might help to try a wider range of IDE devices. Older ones particularly. Perhaps a Conner 2.5" HDD from an A1200?

[Rochabian]

Originally Posted by alexh

I can only imagine that it is timing related (a term used in digital logic to describe the amount of logic between registers and the time it takes signals to go between them relative to the clock speed) in the CPLD/FPGA which controls the IDE.

It is likely that IDE newer devices, particularly solid state storage are a lot faster to change their I/O than older devices which shouldn't be a problem if the timing was correct but might be an issue if it was marginal.

It might help to try a wider range of IDE devices. Older ones particularly. Perhaps a Conner 2.5" HDD from an A1200?

I agree with you because old IDE 2"5 HDD drives tends to have less random freeze than SD/CF on SX-1.

Can these timings be adjusted in any way or a workaround with a special scsi.device , like it was done in 43.45, wich is rock solid with SX1 + 8Mo now ?

I've tried to understand what's inside the Don Adan open-source scsi.device and what can be adjusted in it but this is beyond my skills

[alexh]

Originally Posted by Rochabian

a workaround with a special scsi.device , like it was done in 43.45, which is rock solid with SX1 + 8Mo now ?

I am not familiar enough with the structure of scsi.device to comment accurately. But something as insignificant as the number of instructions between an event and reading the data associated with the event could mean the difference.

A diff of the (disassembled) working and non-working scsi.device files might show how it fixed it.

(Yes I know you said that v46.00 works for HDD and doesn't work for SSD but that doesn't help. Need two versions where one works and the other doesn't for the same media to point to where a delay might help)

I'd bet money it was fixed for SX-1 by accident. Possibly simply adding a few instructions which gives the bus time to settle.

[Rochabian]

The best example i have is the version 43.24
The non patched version freeze the CD32 before workbench appears, at every boot, FMV can't be used
The patched version (with patchstrip) boots fine and we can use FMV but with random freeze (within 3 minutes of playing video )

What does patchstrip to scsi.device, i don't know....

[alexh]

Originally Posted by Rochabian

What does patchstrip to scsi.device, i don't know....

Originally Posted by exkalibur

It works by jumping to the exit routine no matter what the date is, thereby bypassing the alert code completely.

It would appear patchstrip removes a "maximum system date" check from scsi.device 43.23 (and apparently 43.24) which were both time limited beta versions.

I can't believe patchstrip can be the reason you observed increased stability.

I suspect user error somewhere. Could you be thinking of a different version?

[Rochabian]

Sorry, my bad, you were right, i've redone the tests with 43.23 & 43.24

Original or patched versions 43.23 & 43.24 allow use of FMV.
With these 4 versions, freeze appears in less that one minute after launching a video
With 46.0 i can play at least a 30min video sometimes without problem.

As my tests was focused on 46.0 before, i think i'll try other blocks/FS/Maxtransfert settings with these versions to see if it can be more stable.

From what i can remember, there is only 40.40 (rom version, ultra slow with SX-1) 43.23, 43.24 & 46.0 that allow the use of FMV in Workbench
Other versions freeze the CD32 before workbench is fully loaded or freeze when you launch a video (43.45 BB version works great until you use FMV)
It will be hard to check what has been added /removed in those versions...

I've tried for "fun" the scsi.device 45.7 from 3.1.4 and 47.4 from 3.2.1 :
They dont load workbench with FMV

Those 2 versions on the same CD32 with same SD card with SX32 are working great, so there is definitely a bad implementation of Gayle in the SX-1....

[Don_Adan]

I think that FMV needs minimum transfer speed value, which reachest only v46.0 on some your HD. All others scsi.device are too slow on your config. I think that v46.0 is the fastest version available. You can compare speed results for all versions of scsi.device tested by You. And we can see if Im right.

[alexh]

Could stability be improved by increasing the buffers for that HDD?

[Rochabian]

I think i've found the good combo with your helps, guys

I've tried to add some buffers (1000) but it doesn't seems to change anything
Then, as Don suggested, i've done some speed tests, here is the results with Sysinfo :

With SD card Sandisk Extreme 120Gb in a SD2IDE (my main target for this setup ):
scsi.device 43.23 -> 1,453,663 B/s
scsi.device 46.00 -> 1,872,457 B/s

With IDE 2,5" 100GB FUJITSU from 2005 :
scsi.device 43.23 -> 2,586,947 B/s
scsi.device 46.00 -> 2,759,410 B/s

Don is right, it's the fastest scsi.device, at least with my config
Interesting thing, when with 43.23 i had freeze within 2min of video playing from SD, with my ide drive it has freezed after 30min.
Then with 46.0, no freeze at all for now. I played 2 files consecutively of 90min each without freeze, so it's really nice.

I think i"ll do some more tests but perhaps SD card is finally not the good choice with SX-1 (too bad because i've done a nice 3d printed SD card slot that hooks in the rear of the SX-1 for fast file transfer but i won't use it , or in my secon SX-1 with 8Mo perhaps) . Old IDE drives are faster. And there seems to be a relationship between speed and freeze, as said Don. More speed= more stability

The combo that seems most stable for now is :

SCSI.device 46 + 2"5 HDD IDE (100GB ) + FFS from 3.1.4 + Max Transfert 0xFFFFFFF + Block size 2048 + FMV + 4MB 60ns SIMM

Thank you guys for helping me to reach my goal. I can finally play MPEG file from HDD without random freeze

It doesn't explain at all why some SCSI.device completely freeze the CD32 when we access the FMV, it couldn't be only speed related but at least the version 46 works great, so it's fine.

Another thing i was wondering : Why a SD card that can output 200MB/s could be slower than a old IDE drive? Do we lose speed with IDE signal conversion from SD or it is CPU limited ?

[alexh]

Block size? Shouldn't that be 512 always? Maybe it means something else in AmigaOS but I was sure it was supposed to be 512 on all but the newest devices

And Max Transfer is wrong for normal Gayle IDE controllers it needs to be 0x1FE00

Have you tried a different SD card? Maybe you've been unlucky enough to get a fake card?

[Rochabian]

Originally Posted by alexh

Block size? Shouldn't that be 512 always? Maybe it means something else in AmigaOS but I was sure it was supposed to be 512 on all but the newest devices

And Max Transfer is wrong for normal Gayle IDE controllers it needs to be 0x1FE00

Have you tried a different SD card? Maybe you've been unlucky enough to get a fake card?

Yes i tried MANY SD Cards (at least 6 )from 32GB to 128GB, mainly Sandisk or Samsung with same results

MaxTransfer problem seems fixed in the FFS from 3.1.4 and allows larger block size too for large drive (it's not the physical block size if it's what you mean but the filesystem block size, as called in HDToolbox ). I think i've seen that in the changelog if i remember well

scsi.device 46 has no MaxTransfert Limit, according to Don

The SX-1 Manual mention a MaxTransfer limit of FE00 for some drives, but i had no benefit using this setting

I'm testing my IDE drive with the A1200 IDE 2,5" buffer from Amigastore to see if it can improve stability further

[Rochabian]

To answer shortly at the main question of this thread :

Yes it needs a specific scsi.device, and it's already there, it's the version 46.0 from Don Adan.

As said before only 3 versions are usable with this configuration :

40.4 (CD32 ROM version) -> works very slowly and limit drive to 4GB
43.23 or 43.24 -> works but limits drive to 128GB
46.0 -> works and allow drive (SATA to IDE) until 2TB, SD to IDE and CF to IDE are limited to 128GB