ATI All-in-Wonder series
Driver Version Tested: MMC 7.1, Win2K version (For this review, I mainly tested with AGP All-in-Wonder Radeons, but I also used one PCI AIW 128, and there is no difference as far as TV capture quality or MPEG encoding quality. They all use the same MMC version, so they all share the same MPEG encoder.) The AIW is the only product reviewed here that claims to capture real-time MPEG in software. All of the other products are either hardware-assisted, or off-line software encoders. This makes the board simultaneously the best bang for the buck, and also one of the most frustrating boards I've used. Despite cries of "hype!", the AIW can in fact encode MPEG-2 in real time, if your computer is fast enough. If you only want to capture full-size I-frame video (roughly equivalent to MJPEG in quality), you can probably do 15 Mbit/s on a Pentium III-500. The second you shift to IP-frame or IPB-frame video, though, you begin to need some very serious horsepower. On my 1.2 GHz Athlon with PC-133 SDRAM, it dropped frames when trying to do full-D1 IPB encoding. Later I tried the board on a 1.4 GHz Athlon with DDR SDRAM and it handled full-D1 encoding using only 80% of the CPU. Note that multiple CPUs don't help: a dual PIII-800 I tried this board in fared worse than a single PIII-933. Also, the operating system doesn't matter: the board uses the same amount of CPU for MPEG encoding whether it's running under Win9x or WinNT/2K. With ATI's encoder, the major impacts on CPU usage are frame size and whether you're using I-frame or one of the delta-frame GOP formats. Things that don't matter (much) are: increasing the motion search accuracy, increasing the motion search range, or increasing the target birate (!). If you can encode at 2 Mbit/s without dropping frames, you can probably encode at 15 Mbit/s as well, if you have the disk bandwidth. Disk latency is also not important, apparently: I tried capturing to a single ATA-33 disk (benchmarked at 13 MB/s) and to a pair of ATA66 disks in a hardware RAID 0 array (35 MB/s) with no change in CPU usage. If your system is slower, this board is still usable. One option is to capture to full-frame 15 Mbit/s CBR I-frame-only MPEG, which is effectively the same thing as 13:1 MJPEG compression. Another option which will get you similar results is to capture to AVI with a codec like PICVideo MJPEG; this affords you the option of using lower compression ratios to get even better quality. PICVideo, for example, can go as low as 2.5:1. A final option is to use a smaller frame size. On one 933 MHz system, I captured to VCD format (MPEG-1, 352x240, 1.15 Mbit/s) using only 30-40% of the CPU. When I set it to SVCD format (MPEG-2, 480x480, 2.5 Mbit/s), it used 80-90% of the CPU. On the subject of video CD, the AIW does produce compliant VCD files, but it doesn't have a specific SVCD setting. You instead have to produce a regular MPEG-2 file that has an SVCD-compatible frame size and bit rate, and then re-multiplex it with a tool like TMPGEnc into an SVCD-compatible program stream. If you choose to capture to AVI, I suggest using VirtualDub instead of ATI's utility, because VirtualDub is much faster. Even on my 1.2 GHz box, I couldn't do full D1 with HuffYUV within ATI's utility, but with VirtualDub I did it with half the processor's cycles to spare. Also, VirtualDub can do 48 KHz audio, which ATI's software doesn't allow. Unfortunately, ATI doesn't expose its MPEG codec as a DirectShow filter, so you can't capture to MPEG with VirtualDub; you have to use ATI's capture utility. Once you've got a high-bitrate I-frame MPEG or an AVI, you can transcode down to IPB MPEG in software with TMPGEnc or whatever better offline encoder you have around. The AIW has RCA audio inputs, but these are simply passed through the card untouched to your sound card's line-in. In fact, you can completely ignore the AIW audio inputs and connect your source directly to your sound card. Because the AIW depends on your sound card for capturing audio, your video's audio track quality depends on your sound card, not on the AIW board. Make sure your sound card has a line in (some only have a mic in), that it has all the sampling rates you'll want (44.1 KHz, 48 KHz, etc.), and that it's a reasonably high-quality unit, otherwise things like video lip synch can be affected. (Due to encoding latency or jitter.) There's an odd little misfeature in the AIW's capture applet: when capturing to a compressed AVI flavor, the program apparently can't ask the DirectShow filter what kind of compression rate it's getting. The program (wrongly) assumes the output format is uncompressed, so when it reports "video recording time left on disk", it starts off too low by a factor of the compression rate, and counts down proportionately slowly. For example, if you're getting 10:1 compression, it will report 1/10 the amount of actual recording time on disk, and will count down one "second" per 10 real world seconds. This is no doubt a limitation in Microsoft's DirectShow infrastructure, and therefore not really ATI's fault. It's easy to compensate mentally for it, once you know what's happening. The AIW has a cute Tivo-like ability: it can record MPEG and play it back on the fly, with pause and fast-forward abilities. Naturally it doesn't look nearly as good as with a Tivo, and it requires an extremely fast machine unless you want to put up with very small frame sizes. This board has some kind of "soft" Macrovision detection. I have personally tested several different AIW boards in many different machines, and the only time I ever got a Macrovision hit was occasionally under WinME. Under Win2K, there seems to be no problem. I suspect that this is a driver issue, so upgrading to Win2K to use those drivers, or downgrading to older Win9x drivers might help. There are many patches floating around on the net that purport to disable Macrovision detection with this board, but I haven't tried them, and I hear that they often fail to work. I can't help you find working patches if you need them, so don't ask. OutputFrame 0, Frame 1, Bitrate and Quantization Data. In the frame grabs, notice how little artifacting there is. Blockiness is almost completely absent. I'm told that the encoder is based on Ligos' GoMotion technology. The same engine is in the Premiere version of their LSX-MPEG encoder, and in some Matrox products. In the chart, notice the extremely high Q level beginning at around 41 seconds, and the precipitous dip in the bitrate beginning at around 55 seconds. I can only call this an encoding error. In later testing, I found a particular clip that would reliably fool the encoder into breaking the video up horribly unless you cranked the bit rate way up, even though the clip should have been easy to encode. These faults do not happen often; you just have to be sure that if you're encoding a video with this board for posterity that you check its output carefully before calling it a wrap. Bottom LineIf you have the PC to run it, this is the best MPEG encoder for the money you will find. If your PC is too slow for the video you want to capture, you'll think this is the worst MPEG capture card in the world. Even so, it still makes a reasonable AVI capture board. When considering this board's value, compare it to the non-AIW version of the same board. At the time of this review, there was an $80 difference between the All-in-Wonder and the plain Radeon version of each ATI card. That's about what you'd pay for one of Hauppauge's better WinTV boards. Even if you compare to Hauppauge's cheapest WinTV board, you have to factor in all the extra software ATI bundles. Not bad for $30 extra!
Usability: 10 |
| Updated Sun Feb 26 2006 02:07 MST | Go back to MPEG Encoder Reviews | Go to my home page |
