MS Dual core patch = +15% encoding performance

[ngnr]

I just got myself a new AMD X2 4600 and by accident I found a few mentions of a MS patch to Win XP that increases performance on dual core CPU's in some instances by preventing threads migrating across the cores and reducing performance. I haven't seen anyone mention it on here so I installed the patch and did a little testing of encoding speed using CCE SP2. Immediately after installing I noticed what I thought was about a 20% speed increase in encoding while testing on a file I had just encoded previously before patching!

Finding that kind of surprising, I tried another file, first with the patch turned on and then disabled and the CCE speed with the patch off was 5.33 and on was 6.03, about 13% faster. And while I can't be absolutely sure (no actual testing yet) I think the CPU's are actually running at a lower utilization with the patch on. Has anyone else seen this? In case anyone else wants to try this the MS kb article is here

http://support.microsoft.com/kb/896256

I'd be interested in your results to confirm what appears to be happening on my machine, just to convince me I haven't made some sort of mistake. I'll be doing more testing over the next little while make sure I'm not dreaming. It's pretty amazing.

[Capmaster]

Thanks for the tip, and thanks to Dr. DOS for tipping me off about this topic 8)

I installed it and saw a 6.5% improvement in my video encoding benchmark

[ngnr]

Excellent! I'm glad you could replicate it and thank you for posting.

What I've found with a bit more testing is it seems to vary by file and type although I don't quite understand that. I have one particular divx that shows the most improvement, about 25% that I'm going to retest, and some of my own DV caps that show almost none although I haven't done a lot of testing. In no case have I seen any worse speeds with the patch on so I'll probably just leave it on and take whatever I get. If you do more testing and care to post results I'd be interested. Seems odd that there's not a lot more known about this with the proliferation of dual core CPU's out there now.

[Capmaster]

about 25%

Wow!!!

Now I'm really interested. I think I'll toggle it off, get some baselines, and then turn it on and benchmark a few CPU-intensive apps and post it here.

Edit: I tested it with Photoshop 7, Riva FLV Encoder 02.00.0004 and Nero Vision Express 3.

For Photoshop I opened a large 3016x1164 .jpg, quadrupled the size to 6032x2328 pixels so I could actually load down the CPU, then performed a Colored Pencil filter on it.
Before: 13.83s
After: 13.82s

For FLV Encoder I converted a 62.7MB MPEG1 to .flv format (YouTube format).
Before: 1min. 24.81s
After: 1 min. 25.56s

For NVE I opened the same 62.7MB MPEG and exported as another MPEG, VCD format.
Before: 1 min. 24.68s
After: 1 min. 24.68s

Apparently not all apps misbehave where the patch would help. Clearly TMPGEnc benefitted, but the other multithreaded apps didn't.

Maybe some CPUs are more prone to being helped by the patch

[ngnr]

Honestly, I'm not sure what this thing does. I have one file that I tested (AVI-DIVX, convert to MPEG2 in CCE, all after a restart) that tests as follows:

Patch off = time 2:50 CCE speed = 4.94
Patch on = 3:00 CCE speed = 4.81
Patch off = 2:53 CCE speed = 4.86
Patch on = 2:49 CCE speec = 4.99

Not very consistent.

Another file, tested under same conditions, AVI-DIVX

Patch on = 2:18 CCE speed = 6.10
Patch off = 3:01 CCE = 4.61
Patch on = 2:18 CCE = 6.11

Which seems to be a consistent test result (didn't do the "off" test again, too late!). Got me confused. Got to get some sleep and try again later.

[Capmaster]

Strange. You wouldn't ever expect to see the times increase I saw it on FLV Encoder and your data agrees.

It would be interesting to see if anyone else has tried the patch and done a similar comparison.