FrameDrops w/anything other than PC/VCR

[funsoul]

Hi Folks!
Have tried moving to the older vidtools and applied Flying Dutchman's YUY fix. The required throughput still exceeds my sustained on my h/d's...bummer

The problem this post is about, though is that when I capture using the PCVCR Remote, can capture (using MJPEG) at 704x480, 30fps, 44.1KHz stereo without any frame drops.

With virtualdub, am consistently dropping 2-3 frames per minute. Also tried using avi-i/o with the exact same result.

Any suggestions or anyone know how to take care of this issue?

Thanks!
-funsoul

[Dr Mordrid]

Sounds like you're at the ragged edge of your drives capabilities. Have you considered using the HuffYUV 2.1 compressor with your YUY2 captures? It takes the data rate down from 21 MB/s to about 11 MB/s. Just set your capture rez in the YUY2 setups and then select HuffYUV 2.1 in the Compression dialog. Try the YUY2 Fast setting first.

http://www.math.berkeley.edu/~benrg/huffyuv.html

Another possible problem would be if you're using a VIA chipped board, especially those with the MVP3 or Apollo chipsets.

Dr. Mordrid


[This message has been edited by Dr Mordrid (edited 10 August 2000).]

[funsoul]

Hi Dr Mordrid!
Thanks for the response!

Yes...am using ibm deskstar gxp eide drives. Even with all the tweaks I've seen...my sustained throughput according to the h/d benchmark is around 4.5meg/sec. Should I be getting higher results than that or is that about normal for eide?

My system is:
98se (was nt4 before...throughput is up over 40% with 98se)
pIII 500e
asus p3b-f mobo
256meg ram
pair of 27.2gig eide's
matrox g400tv

Am still not sure why I can capture using the PCVCR remote without any framedrops while other (better) proggies give massive drops (even if I reduce the screensize to 352x240). Asked this question in the 'official' matrox forum some time back and got a response that said it had something to do with how the 'non-supported' proggies do captures.

Any ideas how I can fix this?

Thanks in advance for any help or suggestions!
cya!
-funsoul

[Marijn]

Just a thought... Maybe it is an idea to check the number of fps that are captured while using PCVCR Remote, since you can't set the capture to exactly 30fps. See is there is a large difference in this number and the one you set in the other proggie. If your source gives you 29fps instead of 30, this maybe your problem.

Marijn

[Eric VanHeest]

I've noticed that I often drop (almost) exactly 1 frame per thousand (about 30 seconds worth of video) when using VirtualDub, but not (at least, not being reported) with PC-VCR. The interesting thing is that 1 frame per thousand is exactly the difference between 30.00 and 29.97 frames per second. I get this drop no matter what the capture speed is (i.e. using YUY2, or using HuffYUV compression, or flat 24-bit RGB, or whatever). When using VirtualDub, I have never noticed these dropped frames cause any problems whatsoever; even after a three hour capture, audio is still synchronized, and a missing frame every thousand is unnoticable (to say the least).

I have my suspicions that PC-VCR is either not telling me everything that's happening, or its frame-rate for capturing is much more closely locked to exactly what the Matrox card provides.

[Eric VanHeest]

-------------------------------------
Yes...am using ibm deskstar gxp eide drives. Even with all the tweaks I've seen...my sustained throughput according to the h/d benchmark is around 4.5meg/sec. Should I be getting higher results than that or is that about normal for eide?
-------------------------------------

Do "all the tweaks" include DMA for the drives (or are they striped together with a FastTrak or somesuch) ? 4.5 MB/sec is really slow for a 17 GB drive, as far as I've seen. Even the Maxtor 5400 17 GB drives, which are arguably one of the slowest drives of that size, bench at around 8-9 MB/sec minimum. 4.5 MB/sec sounds like PIO maximum for a 500 MHz machine. You might check your processor usage while benchmarking your drives; if it's 100%, DMA may not be functioning properly. A full speed benchmark on my systems (on a single drive, not attached to a FastTrak) uses up only about 5-8% of the processor with a Celeron 566, but in PIO mode, I get 5 MB/sec max, and processor usage is 100%.

On the FastTrak drives, however, DMA settings don't matter, since the FastTrak is essentially a SCSI interface.

Which benchmark are you using, by the way? I tend to use VirtualDub's setup benchmark, because it's simple and I can test different check-file sizes easily (and if you use the same benchmark program for all your tests, it makes comparisons much more valid).

The quest for more performance is endless !

[funsoul]

Hi again folks!

Thanks Marijn...will check that out but am pretty sure it's getting the full 30fps.

Eric- thanks for your reply. The tweaks I made are the ones that are on the 'official' matrox page.

Damn...something must be wrong on my side then. These drives (27gig...7200rpm) are supposed to be some of the fastest eide's around. Am not using fastrack or anything.

Am not sure what PIO is and (apologies) am not sure how to check DMA either (although am pretty certain it's enabled). Would you be able to give me some instructions or direct me to a webpage that will step me through what I should do? The extra throughput would be stellar!

Also...have been using the standard matrox h/d benchmark tool.

Thanks again
-funsoul

[Dr Mordrid]

1 frame drop every 30 seconds is indicative of capturing at the wrong frame rate. For NTSC video the proper rate is 29.970 fps, not 30 fps. Unfortunately PC-VCR has very poor frame rate control and often gives this result. That's why I capture with AVI_IO or some other program that can lock down 29.970 fps exactly.

The Matrox benchmark tool is almost useless except for making the registry entries that contain it's results. I run it once to create those and then plug in the results I get from SiSofts SANDRA benchmark. You can get a shareware version that includes the disk benchmark here;

http://www.sisoftware.co.uk/sandra

Plug the uncached sequential write numbers into the registry in place of HDBbenchmarks results.

One other thing to watch for with ATA66 or ATA100 drives is when using them on a lower spec'ed controller it may be necessary to put them into a compatability mode, either ATA33 or ATA66 depending on the model.

For the IBM drives this program is called "IBMATASW". It allows you to set ATA66 and ATA100 drives up for use on either ATA33 or ATA66 controllers as necessary. It can be found on IBM's HDD download page and supports the following drives;

Travelstar: 12GN, 18GT, 25GS, 20GN, 30GT, and 32GH

Deskstar: 22GXP, 25GP, 34GXP, 37GP, 40GV, and 75GXP

Dr. Mordrid



[This message has been edited by Dr Mordrid (edited 11 August 2000).]

[Eric VanHeest]

---------------------------------------------Am not sure what PIO is and (apologies) am not sure how to check DMA either (although am pretty certain it's enabled). Would you be able to give me some instructions or direct me to a webpage that will step me through what I should do? The extra throughput would be stellar!
---------------------------------------------

PIO is a method of transferring data from storage devices (including hard drives) that involves using the processor for each segment of data. As such, the processor consumes all of its time requesting blocks of data from the storage device and writing that data into memory itself. As a result, PIO transfer rates are highly processor-speed dependent.

DMA (direct memory access) allows the processor to direct the storage device to "send data to memory" directly, and then do other things while the device and memory are directly connected to each other. Of course, this also applies to getting data from memory and writing to the device as well. This mode of access does not depend on the processor nearly as much, with the typical result being that the bottleneck is usually the speed of the storage device instead (the maximum speed of the hard drive, in most cases).

Other than that PIO is very stable (the processor is involved at every step, so "accidentally" writing to the wrong area is extremely unlikely), there is no reason to use it, if the drive supports DMA (and all current drives do). Enabling DMA in Win98 is supposed to be very simple (Control Panel->System->Devices->Drives->Properties, and then check the "DMA" box), but I've seen situations in which it mysteriously refuses to accept DMA mode. I don't know how to tell whether the machine is "actually" using DMA or not, except to do a hard drive benchmark, which will be around 5 MB/sec using PIO and much higher using DMA.

(Under NT4 DMA is even more pesky to enable, involving some registry keys and/or a program called "DMACheck.exe" from Microsoft). Unfortunately I don't keep very good track of web pages, so I have little to offer in that area.

[funsoul]

Hi Eric & Dr Mordrid!

Woohoo! DMA did it for me. The matrox benchmark is 12.5meg/s...Sandra says it's actually 14.4.

Have downloaded the HuffYUV codec so will see how that goes (using virtualdub for the capture).

Now...where was that thread with how best to do vcds!?! hehehe

Thanks for all you assistance!
-funsoul

[Fred H]

Doc,
- which capture program do you run with YUY2/HuffYUV?
Probable VirtualDub (only what I can use)?

- SiSoft Sandra shows a test result over my dedicated video HDD:
<< Hard Disk (D: >> //15.3 Gb (oh boy, such a miniature? )
< Benchmark Results >
This Drive: Drive Index: 16448
Read Ahead Buffer: None
.......
< Benchmark Breakdown >
Buffered Read: 24 MB/s
Sequential Read: 23 MB/s
Random Read: 5 MB/s
Buffered Write: 25 MB/s
Sequential Write: 22 MB/s
Random Write: 7 MB/s
Average Access Time: 9 ms
Note: only 64 MB SDRAM, PA2013/VIA MVP3 (you don't like it, do you?)
Is it enough to capture YUY2/HuffYUV?
22 MB/s should cover a bitrate of 11 MB/s.
Definitely not full size (704x576 PAL)!
Even half size is slow on my system.

- Is the difference viewable to all people, considering YUY2 vs. MJPEG ?




[This message has been edited by Fred H (edited 18 August 2000).]

[Dr Mordrid]

90% of the time I use AVI_IO 3.02. When I want to do something weird I sometimes capture with VirtualDUB, but not often.

The remainder of the time I capture with something else I can't talk about using settings I also can't talk about ;-))

Dr. Mordrid


[This message has been edited by Dr Mordrid (edited 19 August 2000).]

[Michel Carleer]

Dr Mordrid,
When will you be allowed to talk about it, then?
He He,
Michka