pdvdben

PowerDVD 1.30 System Diagnostic

Benchmark results from DDTESTER.EXE v1.30 (higher scores are better)

`Celeron 300A @ 450MHz`			`Trident 9850AGP (AGP2X DIME)`				`Cirrus GD-5465 AGP`		`Stealth II S220 PCI`
`ASUS P2B, 128MB PC100`			`83MHz MCLK`		`93MHz MCLK`		`600MHz RAMBUS clock`		`126/63MHz clock`
`resolution`	`Depth/rate`	`refresh`	`YV12` `overlay`	`YUY2` `overlay`	`YV12` `overlay`	`YUY2` `overlay`	`YUY2` `overlay`	`UYVY` `Overlay`	`UYVY*` `overlay`
`800`	`8bpp`	`60Hz`	`184fps`	`120fps`	`206fps`	`139fps`	`-`	`-`	`NA`
`800`	`8bpp`	`85Hz`	`170fps`	`113fps`	`189fps`	`128fps`	`83fps`	`68fps`	`NA`
`800`	`16bpp`	`85Hz`	`149fps`	`100fps`	`170fps`	`118fps`	`-`	`-`	`85fps`
`800`	`16bpp`	`120Hz`	`-`	`-`	`-`	`-`	`-`	`-`	`62fps`
`800`	`32bpp`	`85Hz`	`117fps`	`80fps`	`139fps`	`94fps`	`NA`	`NA`	`NA`
`1024`	`8bpp`	`60Hz`	`179fps`	`128fps`	`199fps`	`141fps`	`-`	`-`	`NA`
`1024`	`8bpp`	`85Hz`	`157fps`	`113fps`	`179fps`	`128fps`	`-`	`-`	`NA`
`1024`	`16bpp`	`85Hz`	`132fps`	`97fps`	`154fps`	`113fps`	`-`	`-`	`84fps`
`1024`	`32bpp`	`85Hz`	`75fps`	`49fps`	`90fps`	`66fps`	`NA`	`NA`	`NA`

NA = not available or not supported

- = not tested

`Pentium MMX @ 250MHz (83x3)`			`Permedia 2v 8MB`	`V2100 4MB (125/63MHz)`	`Virge 325`
`ASUS T2P4 (i430HX) 48MB EDO`			`PCI 41MHz`	`PCI 41MHz`	`PCI 41`
`resolution`	`Depth/rate`	`refresh`	`YUY2*` `overlay`	`UYVY*` `overlay`	`YUY2` `overlay`
`800`	`8bpp`	`60Hz`	`NA`	`NA`	`72fps`
`600`	`16bpp`	`60Hz`	`-`	`-`	`67fps`
`800`	`16bpp`	`85Hz`	`27.4fps`	`86fps`	`NA`
`800`	`24bpp`	`60Hz`	`NA`	`NA`	`68fps`
`800`	`32bpp`	`85Hz`	`28.1fps`	`NA`	`NA`
`1024`	`16bpp`	`75Hz`	`26.9fps`	`96fps`	`NA`
`1024`	`32bpp`	`75Hz`	`25.5fps`	`NA`	`NA`

NA = not available or not supported

- not tested

The V2100 4MB (Stealth II S220) and Permedia2 provide overlay EMULATION. Neither chip directly supports hardware overlays. (Although the Verite2x00 is fully capable, the Stealth II lacks sufficient RAM for overlay emulation in 32bpp mode.)

Although not shown, the K6/2 & FIC system never exceeded 130fps with the fastest card (Trident985) in the round-up.

How to properly read the benchmarks

For each test (YUY2, YV12, etc.) DDTESTER first allocates a Directdraw surface. PowerDVD supports both overlays and non-overlay (offscreen plain) surfaces. Apparently, both PowerDVD and DDTESTER prefers Directdraw resources in this order: 1) YUV overlay, 2) YUV offscreen, 3) RGB primary, 4) GDI (DIB). Both PowerDVD and DDTESTER always request the best-reported Directdraw capability. If this is unsuccesful, even if lesser Directdraw resources are available, DDTESTER terminates abnormally.

DDTESTER's screen-benchmark measures overall video-subsystem throughput, in the context of motion-video display. The "fps" measurement represents how quickly the system can write whole video-frames (720x480, 360x480, 720x240, YUV or RGB) into the display adapter's RAM. (I assume VSYNC is off, because the fps scores are much higher than the monitor's refresh rate.)

The color format (UYVY, YUY2, YV12) affects the benchmark results. Packed YUV formats (YUY2, UYVY, YVYU, YVYU) supported by DDTESTER are all 16bpp.

One YUY2/UYVY/YVYU frame consumes 720 x 480 x 16bits = 691200 bytes.

One YV12 frame consumes 720 x 480 x 12 bits = 518400 bytes.

-- YV12 and other YUV fps scores are not directly comparable!

 
YUY2-130fps == 130 * 691200 bytes = ~90MB/sec (more bandwidth)
YV12-150fps == 150 * 518400 bytes = ~78MB/sec (higher frame rate)

--The two scores above indicate 1) data-bandwidth (in MB/sec) is better with YUY2. 2) YV12 is still faster for drawing to the display. PowerDVD always selects YV12 if available, for better or worse.

What the benchmarks mean

All other things being equal, faster scores are better, but only to an extent. Modern PCI/AGP video cards provide bus-master capability (host-initiated, target device executed), so the CPU doesn't actually write each byte. In other words, the CPU will setup a hardware-assisted block transfer, and the bus hardware between the peripheral and motherboard core-logic will execute the block move, freeing the CPU to do other work

Higher scores are better for two main reasons.

As long as the PCI bus and main-memory are hogging the bus with video-data transfers, your CPU has reduced access to main-memory. Your PCI/AGP video card shares the system RAM with the CPU, hard drive, and all the other cards in your PCI/ISA slots. These devices all consume some processor time and bus bandwidth. So in general, the faster your video card can accept data, the better. This leaves the bus available for other devices.

In DVDs, the MPEG2 data-rate, and hence the processing requirements for decoding, vary over the course of a movie. DVD employs variable bit rate (VBR) MPEG-2 coding, ranging from 2Mbits/sec up to ~9Mbits/sec (for a full 2-hour movie, avg MPEG2 video-rate ~3.5Mbit/sec.) A borderline PC may barely maintain 30fps during the "average" portions of a DVD-movie. But scenes with lots of motion are coded at higher bitrates, and playback will break-up or skip on the borderline PC. A faster PC has the "headroom" to deal with the peak data-rates of a DVD-movie.

Miscellaneous notes:

+ video cards with very high memory bandwidth (RivaTNT, G200) are generally bus-limited (transfer rate is limited by the AGP/PCI bus throughput.) This means the card's performance does not vary (much) as a function of the Windows display mode (1024x768x32bpp 100Hz vs 800x600x16bpp 85Hz.) The tested hardware shown above is clearly limited by video memory speed.

+ there is anecdotal evidence showing how poorly Super7 motherboards implement AGP. This manifests in 1) compatiblity issues between combinations of some chipsets (Via MVP3) with specific video adapters (i740, Savage3D, for example.), 2) reduced AGP throughput compared to P2/Celeron platforms.

+ "half-width" and "deinterlaced" benchmarks use half-sized frames (360x480.) They are only meaningful if PowerDVD runs in half-resolution mode (not enough video RAM for 720x480.)

Take a look at some other DDTESTER scores over at http://members.xoom.com/virtuamedia/dvd/ddspeed.html