Promise Fasttrak/Marvel Interrupts.

[Brian Ellis]

Does anyone know how to change the IRQ of the RAID card? I've recently installed a TX-2 which is using IRQ 11, which is the same as Marvel, with obvious results.

I have IRQ 12 free but neither Marvel nor Fasttrak seem to be able to change. My BIOS does not seem to be able to allocate an IRQ to a specific slot, either. Windows Resources refuse to allow Marvel to change and there is not even a Resources tab on the RAID drive.


Nice please, someone??????

[The Scientist]

Brian,

If your BIOS doesn't allow manual resource allocation
then you are, AFAIK, probably reduced to playing the
card & slot juggling game...

Your AGP slot will almost certainly share resources with
at least one PCI slot. Sometimes its the adjacent PCI
slot, sometimes its the furthest one away from the AGP
slot - you might have to experiment a bit!

What OS are you using?

hope you get it sorted out,

Simon

[Brian Ellis]

Simon

Sorry, forgot to mention, I've already juggled

Thanks, all the same!

At the moment, 98SE, because of the Marvel.

[dchip]

Do you have whatever "plug and play" settings your BIOS has, turned on? I have not known the Promise cards to insist on a particular IRQ so it might be your BIOS creating the IRQ11 fixation. Plug and play releases the OS to bounce devices to an appropriate IRQ (not that conflicts don't occur with PnP). When I was using 98SE I used to install everying as PnP. Sometimes I still had to do the card juggle afterwards but always ended up with all devices settled in and working fine.

[Brian Ellis]

Dchip

Yes, I've also juggled with the PnP/PCI settings in the BIOS, but normally they are all enabled. Unfortunately, the m/b has 3 ISA slots (I needed them for my CD-R burner and scanner at the time I bought it), so it has only 4 PCI slots, all used. I have a couple of other tricks to try, up my sleeve, but they will be really a PITA.

[Brian Ellis]

By removing the PCI sound card (Live!) which was using IRQ 5 and replacing it with an SB16 in an ISA slot, also using IRQ 5, the IRQ of the Promise card immediately hopped to 3 (after using the installation diskette again). This was a nuisance because I used the two independent outputs of the Live! extensively (one to L/S, t'other to BOB). However, it made no difference to the performance and/or drops. This surprised me.

Nothing daunted, I put Live! back in. Lo and behold, ecce, the IRQ stayed on 3 again with no difference to the performance or drops, even though there are no IRQ or DMA conflicts. Sandra tells me she measured:
Buff read 166 Mb/s
Seq read 28 Mb/s (I expected more)
Random read 6 Mb/s
Buff write 51 Mb/s
Seq write 43 Mb/s
Random write 13 Mb/s
Av, access 8 ms
This is with a 30 Mb disk cache.

The CPU is a P III 450 (Slot 1) with an FSB speed of 100 Mb/s, RAM 384 Mb. The chipset is 440BX, The Sandra permance rating is PR541 and the Pentium performance rating is P622.

It is just possible that the drops were due to the input (I was using a VHS tape I use for benchmarking and it is now 8 years old). I'll try again with a clean video source and see what happens).



I think I can probably live with these figures

[Dr Mordrid]

The Fasttraks do not require any IRQ in particular, just taking what Windows gives them.

One thing to remember when shuffling cards: reboot WITHOUT the shuffled cards in the system BEFORE reinserting them and booting again. Otherwise Windows might remember the last boot's assignment. Believe it or not, this does happen.

Dr. Mordrid

[Brian Ellis]

I wonder whether Sandra isn't playing a nasty joke on me! Capturing to the Marvel hardware MJPEG, I was getting a typical average 1 dropped frame per minute. Out of curiosity, expecting bad things, I changed the capture to 640 x 480 RGB 24 or 16 bits. In both cases, I was dropping typically about 70% of frames in AVI-IO (set to replace dropped frames). Yet, playing back the same mucked up clips, there were 0 drops. Yet Sandra gives me good write speeds but poor read speeds.

I should perhaps mention before I installed the Fasttrak, I was capturing to a single UDMA-66 disk with almost zero drops, perhaps one per 2 Gb file on average, usually at a transition, so real-life performance (as opposed to Sandra) seems to have dropped with Promise, despite faster disks.

[Dr Mordrid]

RGB capture is very much disk and CPU intensive.

640x480 RGB is, on the Marvels etc., about 27.1 mb/s. Now...you need about 25-30% more than that for system overhead, so figure somewhere around 35 mb/s as a minimal requirement.

So far, so good. A pair of modern 7,200 rpm drives on a Fasttrak should handle that.

Now...for why the CPU has to be involved. It's because of the cards NATIVE capture format...which is YUY2. When you capture RGB this has to be dynamically CONVERTED from YUY2 on the fly, in software, which eats CPU cycyles like crazy. And don't forget that when capturing full frame a conversion from 704 wide rectangular pixel (YUY2) to 640 wide square pixel (RGB) also has to be done.

Capturing YUY2 is much easier on resources because no format conversion is necessary before laying it onto the disk. It also only requires 20 mb/s from the capture device for full frame.

This is why so many people downloaded and used Flying Dutchmans YUY2 patch for the Matrox drivers before it became a standard feature in the G450-eTV and in later driver builds.

Dr. Mordrid

[Brian Ellis]

Doc

Yes, I agree with you. I only tried the RGB to see what it would do, not because I was wanting to really use it.

I can now update on the drops. I've put in a "pure" video signal (virtually noise-free, taken directly from the camera looking out the window under good, sunny, lighting conditions, no tape). I'm still getting some dropped frames, averaging about 1 every 2.5 minutes, using the Marvel MJPEG hardware codec. I can live with that, but I would have expected better (and I was getting better than that before Fasttrakking the 'puter, virtually 1 in 15 - 20 min, as a guess). On downloading from a mini-DV tape I shot in India about 4 years ago, I find that the number of drops was about the same (1 in 2.5 min) plus an occasional one which occurred between shots when the camera had been powered down (never when it paused between shots). Of course, the latter drops are meaningless.

It would therefore seem that my VHS "benchmark" tape is introducing sufficient noise, or drop-outs, or tape distortions to affect tracking, or whatever to double or triple the number of dropped frames. Notwithstanding, I'm still not 100% satisfied with the TX-2 performance. Although I have not been able to provide any proof yet, my guess is that the Promise card is PCI-hogging.

[Dr Mordrid]

The last time I checked the CPU utilization on the Fasttrak TX4 it ran in the single digits.

This has to be a mainboard or Digisuite related issue as I've used my TX4 in multiple OS's on three mainboads;

Asus P3B-F
Abit BH6
ECS K7S5A

with the WinTV Theater, eTV, G400-TV, RT-2000 and others that require many times their PCI resources. I've seen no TX4 related issues at all.

The captures in these devices ran up to and sometimes over 30 mb/s and playback has often been more than double that...or more.

Dr. Mordrid

[Greebe]

Many of us have found Sandra to be unreliable at best when testing HD's. Use HDtach 2.61 instead.

[Sciascia]

Originally posted by Greebe
Many of us have found Sandra to be unreliable at best when testing HD's. Use HDtach 2.61 instead.

Don't use it if you are using a FT100. Unless BSOD's are your idea of fun. There is a supposed workaround in the Promise FAQ's, but it sounds like too much work. I/O Meter is supposedly a better choice, but it looks even more complicated. Maybe if you play around with it. Virtual Dub has a benchmark tool in the AuxSetup program. Not sure how good it is.

[Patrick]

HD Tach is good, but...

Many of us have found Sandra to be unreliable at best when testing HD's. Use HD Tach 2.61 instead.

(Edit: I typed this out before I read Sciascia's post.) It's possible the newer FastTrack cards are different, but the original FastTrack could not be tested with version 2.61 of HD Tach without locking up the computer. Version 2.52 works fine however.

Brian, just for the heck of it, try using a utility called TaskInfo2000 to see if it helps your dropped frames. This program allows you to set the process priority higher for anything running on your computer. Some people really swear by it. You can read about and download it from here:

http://www.iarsn.com/index.html#/download.html

[Brian Ellis]

Thanks, guys.

Patrick, I downloaded your suggestion and'll try it. However, my experience with this kind of thing is that they create more overheads than they detect, but am willing to give it a shot. Dunno when, as I'm pretty busy on other things (including weeds growing visibly in the garden: looks like playback at 1,000 fps )

In view of the comments, I won't try HDTach. I don't look for trouble: it finds me without looking