Amiga.org
Amiga computer related discussion => Amiga Hardware Issues and discussion => Topic started by: madsjm on July 13, 2007, 11:11:36 AM
-
I'm looking to speed up the disk IO my A4000.
Is the Buddha Flash Phoenix Edition any faster than the built-in IDE controller?
-
slower or same
-
Go SCSI!
--
moto
-
Yes... GO SCSI! It is simply the best!
Otherwise if you want keep IDE drive, check THIS (http://www.elbox.com/products/fast_ata_4000.html)
Alex
-
Thats all very good, but watch your cpu usage hit 100% with the powerflyer, no DMA.
SCSI all the way if you want fast HDD/CD access with no real cpu overhead. Though you have to go for the right SCSI solution as a lot of them are not DMA either.
-
And for Scsi, Phase 5 SCSI (from a blizzard card) is the only solution for scsi ?
APollo SCSI is without DMA, its better with IDE lol.
-
I have a GVP 2000 HC+8 controller, will that be faster than the internal IDE?
-
madsjm wrote:
I have a GVP 2000 HC+8 controller, will that be faster than the internal IDE?
"Faster" is a relative term. It's possible scsi.device and a relatively modern IDE Drive would give good results provided you had a decent cpu because Commodore's native controller does not use DMA nor FastATA modes.
The GVP board by contrast is a DMA controller (albeit a 24-bit one on the Zorro II bus). A good SCSI drive will yield good transfer results (2+MB/s) but what is more important in the equation is the DMA ability of the card, which means the cpu will not be so bogged down doing PIO transfers as with the Amiga's native IDE scsi.device. Of course, it's a different story if you are using scsi.device on an A3000 or A4000T, which have DMA scsi controllers built-in.
-
And don't forget about using Zorro II DMA controllers on a '030/040 machine: there's no way the SCSI controller can write to 32 bit memory (beyond 24 bit address range), so data has to be copied from slow Z2 RAM (that you hopefully have on the controller) to fast 32 bit RAM - might still be faster and more CPU efficient than IDE since the copy can take place in one piece and there's no constant polling on the interface.
The best solution is a Z3 SCSI controller (not many around) or an accelerator integrated one.
However, if you're not doing I/O intensive and CPU stressing tasks simultaneously, this discussion is more academic. The FastATA is probably the easiest way to improve speed.
-
Using the SCSI controller on a CyberstormPPC with good quality cables 2M in length or less and 10,000RPM drives and SFS filesystem I can get sustained disk to disk transfers ~8.5MB/sec. Whats really cool is that, while one large copy operation is happening, I can initiate another large copy operation and see only a tiny drop in overall throughput (none at all if the second COPY is using two disks not being used in the first COPY). Interestingly, the disks themselves can end up being your bottleneck. Also, not a single byte of chipmem is needed and CPU barely utilized (if at all). DMA is indeed a beautiful thing. :)
-
stopthegop wrote:
Using the SCSI controller on a CyberstormPPC [snip] I can get sustained [snip] transfers ~8.5MB/sec.
8.5MB/sec is quite slow. I've had burst rate of 40MB/s and sustained of 20MB/s!
Try turning on Synchronous mode.
-
Really?? What filesystem? I'll give synchronous a try, but I thought that would be slower... Why would synchronous mode be faster? Wouldn't that also limit things when doing multiple I/O operations at the same time?
-
Synchronous transfer is simply faster, and there's Ultra SCSI beyond that, too. ;-) (plus U2W, then U160, U320 - sigh...)
My old A3000 can do ~8 MB/s with a 15 year old 1 GB Toshiba, that's pretty much all you get out of async SCSI. 10kRPM drives can easily do 40 MB/s.
It also speeds up 'simultaneous' I/O transfers on several disks because it reduces bus load.
-
@Zac67
A couple of questions:
Can you switch on asynchronous mode on the CyberstormPPC/MarkIII without having to then reformat your existing scsi drives?
How do you swith on asynchronous mode on the A3000?