HDV

The Future is Here and it's HDV

There is a lot of buzz around HDV. Lots of information and misinformation about mpeg transport streams (m2t), proxy files, and intermediaries. I decided to do some measurements of my own to see what the tradeoffs are.

You can capture HDv in its nativce M2T format but that's really not good for editing. M2T is an MPEG2 transport stream with a 15 frame GOP. You don't want to be editing this stuff. The other options are to use a product like GearShift to make DV proxies and these perform extremely well and take up the same amount of space as the M2T file. Or you can edit using an intermediary like CineForm AVI file which is very high quality but is also much larger than the original m2t file.

Codecs, Codecs, My Kingdom for a Codec

Here are the measurements that I took using the 2.3 codec from my copy of CineForm Connect HD:

Computer: P4 3.0Ghz, 1GB memory.
Source: 1 minute of HDV from Sony Z1U
Payback FPS: measured in Vegas 6.0c at Preview (Good/Auto) 720x540x32 size.

Original M2T:
Filesize: 201MB
Capture Time: 1:00
Payback FPS: 3

Connect HD CineForm (Medium):
Filesize: 768MB
Render Time: 01:35
Payback FPS: 8

Vegas 6.0c CineForm:
Filesize: 995MB
Render Time: 3:51
Payback FPS: 12

Vegas DV Widescreen Proxy
Filesize: 233MB
Render Time: 4:15
Payback FPS: 29.97

As you can see the file generated from Connect HD in Medium mode is 3.8x as large as the original and rendered slightly slower (1.5x) than real-time. The same CineForm codec file rendered from Vegas is 4.9x the size of the original and took almost 4x real-time to encode. (hence my comment 5x larger)

So it is very clear that while you can encode CineForm files using the codec shipped with Vegas 6, you can encode them faster and smaller when using Connect HD. It was also interesting that the CineForm file rendered in Vegas performed slightly better on the Vegas timeline (12 fps vs 8 fps) than the Connect HD file at that preview quality and resolution.

The DV proxy was almost the same size as the original file and was the only file that playback at the full 29.97 frame rate at Good/Auto. It should be noted that both CineForm and DV proxy played back at 29.97 at the normal Preview (Auto) 360x270x32 mode. The original M2T file could only playback at 12 fps even in this small preview mode.

You can draw your own conclusions. This is just what I measured. IMHO, if you want to work with CineForm files, it pays to buy Connect HD and get quicker smaller render files. Otherwise use DV proxies with GearShift.

AMD64 Dual Core ROCKS!!!

Since I did these tests on my P4 3.0Ghz PC, I have built a new PC around the AMD Athlon64 X2 4600+ and ASUS A8N-SLI Premium motherboard. I recreated the tests for the CineForm and DV Widescreen proxy and got these results:

Computer: AMD Athlon64 X2 4600+, 2GB memory.
Source: 1 minute of HDV from Sony Z1U
Payback FPS: measured in Vegas 6.0d at Preview (Good/Auto) 720x540x32 size.

Vegas 6.0d CineForm:
Filesize: 765MB
Render Time: 2:45
Payback FPS: 15

Vegas DV Widescreen Proxy
Filesize: 223MB
Render Time: 1:43
Payback FPS: 29.97

Comparing the Two computers we have the following results:

Computer / Format	CineForm (intermediary)	DV Widescreen Proxy
Pentium 4, 3.0Ghz, 1GB memory	3:51	4:15
AMD Athlon64 X2 4600+, 2GB memory	2:45	1:43

The AMD dual core did a significantly better job at rendering the DV Widescreen proxies at 2.5x faster and slightly better at encoding the CineForm intermediary at 1.4x faster. The AMD was also only mildly better at previewing the footage. 15fps is still not that great although it's better that 12. This is because only one core is used to preview so we are measuring a P4 3.0Ghz against a AMD 4600 (single core). It may have actually been better to purchase the 4800 for the best preview times.

Will those new 10,000RPM Drives Help HDV?

People often ask if buying a 10K RPM hard drive will give any noticeable increase in overall HDV performance. My guess is not very noticeable at all. I have a 10K Raptor as my C: drive but I use a RAID 0 of 7200RPM SATA2 drives for HDV editing. I really don’t feel any difference in my system and using a disk speed program here are my measurements:

WD RAPTOR 10K:
Linear Read = 68.32 MB/sec
Random Read = 5.02 MB/sec
Access Time = 5.35ms
Score = 12968

Price per MB: $2.19

WD SATA2:
Linear Read = 60.07 MB/sec
Random Read = 3.42 MB/sec
Access Time = 7.91ms
Score = 7639

Price per MB: $0.40

WD SATA2 in RAID 0:
Linear Read = 121.68 MB/sec
Random Read = 3.41 MB/sec
Access Time = 7.38ms
Score = 16429

Price per MB: $0.40

I believe the Linear Read measurement is probably the most important for reading video files. Based on those measurements the Raptor cost 5.4x more per megabyte but only a 12% gain in speed of linear reads over a regular drive. It is 30% faster at random read and access time so I guess that’s what you are paying the big bucks for. To me, it’s just not worth it by any measurement. If it were (2x) twice as fast maybe you could justify 5x the price. But at 12% linear and 30% random improvement, it just doesn’t seem like it’s money well spent. I would NOT buy a Raptor again unless they came way, way down in price.

I guess if you need the fastest drives regardless of cost, (and I do mean regardless of cost) you could RAID some Raptors and have a slightly faster file system but I would rather put that money somewhere else. My 500MB RAID of 7200RPM SATA2 drives cost $200. The same size RAID of Raptors would cost $1,097!!!

Bottom line: I doubt highly that you will see any significant performance improvement to justify the enormous cost. But some people swear by them so this is just one man’s opinion. ;-)