FastATA 1200 MK-IV Available For Pre-Ordering

[HammerD]

I know I have to run it in PIO, but I'm not happy about it.

If I had the option, I'd disable the 4000T SCSI and let the Deneb do DMA instead.  At least I can buy new fast USB devices locally, SCSI adapters and drives are rare these days.

The lack of DMA in the FastATA 4000 is almost a feature in certain configurations.

In my experience with Classics, Deneb, and DMA vs. PIO, you really don't gain much by using the Deneb in DMA mode.  When you are driving these devices so fast on the Classic systems your CPU is still being used quite a bit and DMA vs. PIO is about a 5% difference in CPU usage.

It's a general lack of CPU horse power and other sub-system slowness (like Zorro III) that are more of an issue.  My recent networking tests using a Deneb in DMA mode (OS 3.9) and PIO mode in AmigaOS 4.1 Classic showed only about a 5% drop in CPU usage in DMA mode.

[HammerD]

I cannot provide quantitative results at this moment, but qualitatively I experience a huge difference between PIO and DMA mode on system performance when transferring files to and from USB drives.  (Also, USB sound is much better for me under DMA.)

It would be nice to see those test results.  I only tested networking devices.

PS - a small hint, TLSFMem helped my system dramatically under AmigaOS 3.9.  You might want to test with and without TLSFMem running.

[HammerD]

erm... amiga networking is probably the worst test you could do to compare DMA vs non-DMA drivers because SANA-II is totally cpu driven. If you want to perform a decent test use an external USB2.0 hard disk, at least the cpu won't have to copy everything everytime.

-edit- moreover, comparing the performance of drivers with the goal of getting some information about DMA vs non-DMA performance is nonsense if you are using OSes with different cpu architecture. Since most people use OS3.x and not OS4.x the interesting test would be comparing DMA vs non-DMA in the same conditions.

If you are going to perform benchmarks of DMA vs non-DMA drivers you should:
1.- Try to use the same conditions:*Same OS version, same cpu, same HD.
2.- You should use something meaningful that uses DMA. Networking doesn't. Use an USB2.0 hard disk for example.

My goal was to measure RTL 8139 performance vs RTL 8029 performance.  It wasn't to test Deneb DMA vs PIO performance, although from the test it is clear that in the case of networking, it really is insignificant.

I can do some tests on hard drive access when I find the time...

[HammerD]

If the test was performed on a Classic with Mediator you'll probably know that all Mediators lack DMA support and in order to get data from PCI cards that require DMA the cpu always has to access video ram to get the data so in the end everything is moved by the cpu.

Yes I am aware of that.  The Deneb, however, was set to DMA mode in AmigaOS 3.9.  As you pointed out a networking test is not a good test for DMA...I'll do separate testing for hard disk and USB sound cards on the Deneb in DMA vs. PIO in AmigaOS 3.9 and 4.1 Classic.

[HammerD]

If the test was performed on a Classic with Mediator you'll probably know that all Mediators lack DMA support and in order to get data from PCI cards that require DMA the cpu always has to access video ram to get the data so in the end everything is moved by the cpu.

I did a quick test using the updated diskspeed program from Joerg Strohmayer:

AmigaOS 4.1 Classic, Deneb PIO Mode: 3.55 MB/sec, 96% cpu usage

AmigaOS 3.9 BB4, Deneb DMA Mode: 3.866 MB/sec, 41% cpu usage

So you are correct with hard disk transfers - DMA does make a larger difference.

Same disk in the same machine with the same partitions under AmigaOS 4.1 Classic and AmigaOS 3.9, using FFS2 (DOS\07).

[HammerD]

thanks for the test!

Could you please test AmigaOS3.9 BB4 with Deneb in PIO mode just for completeness? :-)

PS: I think you should get higher speed with a CSPPC:
http://www.e3b.de/usb/frame_usb_e.html

I use scsibench for testing HDs http://aminet.net/package/util/moni/SCSIBench

I've got some new test results up using scsispeed.

http://www.hd-zone.com/2011/07/scsi-and-usb-hard-drive-tests/

Basically under AmigaOS 3.9 you top out at 6MB/sec using the Deneb in DMA mode with a USB 2.0 hard disk attached.  Under AmigaOS 4.1 Classic with Deneb in PIO mode you top out at 5.4MB/sec.

With the Cyberstorm PPC SCSI controller and a Barracuda 9.1GB 7200 RPM UW SCSI drive I topped out at 15MB/sec on both AmigaOS 3.9 and 4.1 Classic, with 4.1 Classic coming in significantly higher at lower buffer sizes.

I'll update the test later with the Deneb in PIO Mode on AmigaOS 3.9.

CPU usage results are also there...