Video for CPUs - The Software Decode Reference Thread

Discussion about old PC hardware.

Re: Video for CPUs - The Software Decode Reference Thread

Postby Standard Def Steve » 2016-7-04 @ 02:42

Added 2160p/60fps H.264 at 105 mb/s.

H.264 is so much easier to decode than HEVC that even the Core 2 Quad can handle it fairly well, despite the massive bit rate. And unlike the HEVC decoder, H.264 doesn't seem to need SSE4 for decent performance.
Standard Def Steve
Oldbie
 
Posts: 903
Joined: 2012-9-15 @ 08:04

Re: Video for CPUs - The Software Decode Reference Thread

Postby noshutdown » 2016-7-04 @ 09:38

Standard Def Steve wrote:Added an MPEG1/VCD chart!

about the mpeg1 test, which video card did you use on the 486 and 586 machines? and whats your idea on choosing the player? what do you think of xing, wmp and ffdshow?
in the days when win95 was still popular and xing was better known than wmp6.4, i remember getting ~24fps on a amd486-133, and ~42fps on pentium-120, in xing's built-in benchmark, both using a cirrus5446 videocard. not sure if it made any use of hardware decoding, or just some output accelerations.
noshutdown
Oldbie
 
Posts: 970
Joined: 2010-7-23 @ 17:04
Location: China

Re: Video for CPUs - The Software Decode Reference Thread

Postby elianda » 2016-7-04 @ 17:59

For the MPEG1 test:
Why don't you use the regular playback via ActiveMovie on the old machines?
It is important to be sure to have information about the following points first:
Do you watch at 100% video size or do you scale? Scaling is an expensive operation.
Do you use Overlay capabilities (introduced with e.g. S3 Vision868, Trio64V ...). Does your set graphics mode is overlay capable?

On low end machines like the 486 ActiveMovie defaults to fullscreen at 640x480 at 256 colors for playback in software, video at 100% scaled in a rectangular area in the center. It reaches nearly watchable fps even on a 486DX2-66.

If you play from CD-ROM or HDD it is also important to know if the data transfer is using DMA, otherwise transfer blocks playback. It is highly recommended to use SCSI as subsystem.
I guess you use some PIO mode transfer from the CD-ROM on the 486, which would explain the 2 fps. It looks like most of the CPU time is eaten by the transfer and not by the actual video decoding.
Retronn.de - Vintage Hardware Gallery, Drivers, Guides, HQ Videos.
Youtube Channel
FTP Server - Driver Archive and more
DVI2PCIe alignment and 2D image quality measurement tool
User avatar
elianda
l33t
 
Posts: 2225
Joined: 2006-4-21 @ 16:56
Location: Halle / Germany

Re: Video for CPUs - The Software Decode Reference Thread

Postby Standard Def Steve » 2016-7-05 @ 19:18

noshutdown wrote:about the mpeg1 test, which video card did you use on the 486 and 586 machines? and whats your idea on choosing the player? what do you think of xing, wmp and ffdshow?
in the days when win95 was still popular and xing was better known than wmp6.4, i remember getting ~24fps on a amd486-133, and ~42fps on pentium-120, in xing's built-in benchmark, both using a cirrus5446 videocard. not sure if it made any use of hardware decoding, or just some output accelerations.

I used a Hercules Dynamite Pro VLB 2MB card in the 486 and a Radeon 7000 PCI in the Pentium.

I used the Roxio VCD player because, well, I happened to find it in the random programs folder on my file server. Which means that I must've liked it one point in time. :happy: I don't really know Xing (although the name rings a bell; I probably used it years and years ago).
I'll retest MPEG1 playback with Xing.

elianda wrote:For the MPEG1 test:
Why don't you use the regular playback via ActiveMovie on the old machines?
It is important to be sure to have information about the following points first:
Do you watch at 100% video size or do you scale? Scaling is an expensive operation.
Do you use Overlay capabilities (introduced with e.g. S3 Vision868, Trio64V ...). Does your set graphics mode is overlay capable?

On low end machines like the 486 ActiveMovie defaults to fullscreen at 640x480 at 256 colors for playback in software, video at 100% scaled in a rectangular area in the center. It reaches nearly watchable fps even on a 486DX2-66.

If you play from CD-ROM or HDD it is also important to know if the data transfer is using DMA, otherwise transfer blocks playback. It is highly recommended to use SCSI as subsystem.
I guess you use some PIO mode transfer from the CD-ROM on the 486, which would explain the 2 fps. It looks like most of the CPU time is eaten by the transfer and not by the actual video decoding.

Hmm. I never even noticed ActiveMovie. :blush: I'll retest MPEG1 with ActiveMovie on the 486 and Pentium as well.

I was viewing the video full screen on all of the machines. On the 486 that meant 16-bit 800x600; the Pentium ran at 32-bit 1024x768. The Win2K machines ran at 32-bit 1920x1080.

I know for sure that both Roxio and PowerDVD were using overlay output on the Radeon 7000. It may not have been with the Dynamite Pro. The video looked like it was being drawn differently (tearing) although that could've been due to the very slow CPU.

I completely forgot about PIO mode on the 486. I bet you're right about PIO sucking the life out of the CPU. The ancient VLB IDE controller in that machine probably doesn't support any sort of DMA mode.

I also wonder how much time the CPU is spending just on the 224kb/s MP2 audio stream. That rig can't even decode 64kb/s MP3s in real time!
Standard Def Steve
Oldbie
 
Posts: 903
Joined: 2012-9-15 @ 08:04

Re: Video for CPUs - The Software Decode Reference Thread

Postby elianda » 2016-7-06 @ 01:28

On the TSENGs Overlay capabilities were introduced with the ET6000. So the Dynamite Pro is a not so suitable choice.

How about the Diamond Stealth 64 Video VLB ?
Retronn.de - Vintage Hardware Gallery, Drivers, Guides, HQ Videos.
Youtube Channel
FTP Server - Driver Archive and more
DVI2PCIe alignment and 2D image quality measurement tool
User avatar
elianda
l33t
 
Posts: 2225
Joined: 2006-4-21 @ 16:56
Location: Halle / Germany

Re: Video for CPUs - The Software Decode Reference Thread

Postby Standard Def Steve » 2016-9-13 @ 03:20

The latest version of Chrome seems to greatly improve VP9-HTML5 video performance. In fact, HTML5 video performance in Chrome 53 appears to be almost as efficient as DirectShow! When I get the time I'm going to have to retest all of the machines that were in the old VP9 performance chart.

But for now, here are the Chrome 53 numbers from my main system:

Image

The GTX 970 does not accelerate VP9 decoding, so the new version of Chrome is still using the CPU to do all of the work. Same method of testing as before: Allow the buffer to fill for 30 seconds, then measure CPU load.
Standard Def Steve
Oldbie
 
Posts: 903
Joined: 2012-9-15 @ 08:04

Re: Video for CPUs - The Software Decode Reference Thread

Postby noshutdown » 2016-10-25 @ 02:35

Standard Def Steve wrote:I used the Roxio VCD player because, well, I happened to find it in the random programs folder on my file server. Which means that I must've liked it one point in time. :happy: I don't really know Xing (although the name rings a bell; I probably used it years and years ago).
I'll retest MPEG1 playback with Xing.


i found xing and i got an impressive 26fps in its built in benchmark with following platform:
a-trend 1425b mainboard(sis496)
512k onboard cache, wb mode and 2-1-2 timing
64mb fp dram 0wait
amd5x86-133 at default
nvidia riva128 pci
win95osr2

note that xing runs its benchmark with sound disabled, and decoding mpeg audio alone takes a lot of cpu(more than a dx4-100), so no wonder you got 2fps with dx2-66. however you can downgrade sound quality and channels to quarter cpu usage on audio decoding.
noshutdown
Oldbie
 
Posts: 970
Joined: 2010-7-23 @ 17:04
Location: China

Re: Video for CPUs - The Software Decode Reference Thread

Postby noshutdown » 2017-11-23 @ 15:16

tested xing mpeg player on following machine:
amd386dx-40
cyrix fastmath 83d87-40
the infamous macronics 83c306/305 board, which has no onboard cache but 8kb integrated cache in chipset
16mb dram
stb cirrus5434 2mb
win95sp1

and the scores:
1. 16bit color desktop, hardware acceleration: 2.8fps
2. 16bit color desktop, no hardware acceleration, quality decode mode: 1.2fps
3. 16bit color desktop, no hardware acceleration, speed decode mode: 3.3fps
4. 256 color desktop: all three settings scored an identical 4.0fps

as for image quality, only 16bit desktop with hardware acceleration or quality decode mode looks good, while the rest all look like shit.
i would also try if the ati mach64 performs better...
noshutdown
Oldbie
 
Posts: 970
Joined: 2010-7-23 @ 17:04
Location: China

Re: Video for CPUs - The Software Decode Reference Thread

Postby noshutdown » 2017-12-03 @ 15:35

noshutdown wrote:i would also try if the ati mach64 performs better...


i tried the mach64 but it turned out to be no good, xing results are simply 0.1-0.2 fps slower than cirrus in all settings. i also tried atiplay but it simply crashes whenever i open an mpeg file.
noshutdown
Oldbie
 
Posts: 970
Joined: 2010-7-23 @ 17:04
Location: China

Previous

Return to General Old Hardware

Who is online

Users browsing this forum: mwdmeyer and 14 guests