• Thanks: 1

Topic: SVP Observations & issues..

First, well done on a really nice set of filters.. I've been using SVP on and off for a little while now and feel that it's now at a point where I can use it all the time.. Having used it previously it was quite buggy when seeking etc.. 

There are a couple of issues that seem to give SVP some grief however. I had initially installed 3.11 but it seemed to have a serious mem leak. Playing a vid for a couple of mins it would eventually stutter. SVP index would go to 0.02% or something. At the same time, memory usage on my AMD 7970 3gb card would max out at 3gb, and stay there for anywhere from 10 secs to 2-3 minutes, which it would eventually fix itself but after a few minutes it'd go back to stuttering again.

Also, 1440x1080i (anamorphic) DVB-T streams would simply crash the app that plays them. However, disabling GPU-acceleration (open cl) causes it not to crash. (Tried using media portal, jriver media centre and progdvb).

So I'm now back on 3.06 ..

Re: SVP Observations & issues..

Vladdo
At the same time, memory usage on my AMD 7970 3gb card would max out at 3gb, and stay there for anywhere from 10 secs to 2-3 minutes

There's another feedback from "happy" 7850 owner.
May be it's the same issue hmm

Is "a serious mem leak" in system memoryor in video mem?

1440x1080i (anamorphic) DVB-T streams would simply crash the app that plays them. However, disabling GPU-acceleration (open cl) causes it not to crash.

Could you make a short sample of raw stream?

3 Reply by Vladdo (edited by Vladdo 21-04-2012 03:49:13)

  • Thanks: 1

Re: SVP Observations & issues..

Hey chainik, thanks for your prompt reply smile

At the time, when i was using 3.11, I believe it was using all my vid memory (3gb on a 7970) as MSI Afterburner shows how much GPU memory is in use. I'm now using 3.06 so I can't replicate the bug as it doesn't appear to be affecting 3.06.

Here's a dump of my svpmark score.. Pretty much identical to the 7850 numbers. I'm using 12.3 whql cat drivers.


Test summary
-----------------------
  Date: 2012-04-21T11:41:16
  CPU:  Intel Core i7-2600K @3392 MHz [8 threads]
  GPU:  AMD/ATI Radeon HD 7900 [ver.1.4.1703]
  Mode: FHD + GPU [17 threads]

Overall scores
-----------------------
  Synthetic CPU:                  MC2003
  Synthetic GPU:                  MG3438
  Real-life:                      FG2603

Details: synthetic
-----------------------
  CPU: compose (single-threaded): 583
  CPU: compose (multi-threaded):  2400
  CPU: search (single-threaded):  433
  CPU: search (multi-threaded):   1719
  GPU: system -> GPU transfer:    531
  GPU: GPU -> system transfer:    747
  GPU: calculations:              1808
  GPU: total score:               3780

Details: real-life /FHD
-----------------------
  decode video:                   13.99x (335.8 fps)
  48 fps - vectors search:        1.28x (61.3 fps)
  60 fps - frame composition:     3.41x (204.6 fps)
  48 fps - [SVP] fastest:         5.90x (283.2 fps)
  48 fps - [SVP] simple 1:        4.31x (206.8 fps)
  60 fps - [SVP] good:            2.23x (133.9 fps)
  60 fps - [SVP] high:            1.79x (107.4 fps)
  60 fps - [SVP] highest:         0.82x (49.2 fps)
  72 fps - [SVP] simple 2:        3.74x (269.4 fps)


Now, i've just spent the last hour trying to replicate the other 1440x1080i bug. This is where it gets interesting. If I load the sample into MPC-HC, it plays fine, no issues. However, when i play the same video in progdvb (http://update-progdvb.com/download/ProgDVB6.84.2Pro.exe), it crashes the player. Bear in mind that it did the same thing in media portal and jriver media center (all TV recording apps). Maybe you could download progdvb and see if you can replicate the bug. You don't need to worry about the TV side of it, just load the .ts file.

Acceleration off
http://farm6.staticflickr.com/5452/6951678886_0d8e483df4_b.jpg

Acceleration on
http://farm8.staticflickr.com/7110/7097762003_2e07627042_b.jpg

Here's the sample .ts file.
http://dl.dropbox.com/u/19297064/21%20A … erience.ts

Hope this helps.. and thanks

Re: SVP Observations & issues..

Vladdo
i've just spent the last hour trying to replicate the other 1440x1080i bug

I should ask for the error message first smile This's simple - just patch .exe file of progdvb with that tool and it should be OK. MPC-HC executable is already patched so it works.


I'm now using 3.06 so I can't replicate the bug as it doesn't appear to be affecting 3.06.

3.0.6 and 3.1.1 can easily live together in different installation folders wink
The main thinks that is different in 3.06 and 3.1.1 GPU processing is "Motion vectors interpolation mode" option so try to switch it from "bicubic" to "bilinear". Anyway I think it's some kind of driver bug for 7xxx chips  hmm

  • Thanks: 1

Re: SVP Observations & issues..

Nice work there Chainik.. the old '4gb tool' fixed the crashing problem in progdvb.. SVP must use a heap of mem to render 1080p..  hmm  hmm  hmm

I've installed 3.11 and used the hotfix, so will test that over the next few days.. it wouldn't surprise me at all if it was a driver bug.. AMD drivers are rubbish these days sad

Re: SVP Observations & issues..

Vladdo
SVP must use a heap of mem to render 1080p..

It's a modified Avisynth feature  big_smile - more processing threads requires more memory. In case of i7 the default value is 15 threads so 2GB limit can be an issue.

AMD drivers are rubbish these days
It took half an year for NV to make OpenCL 1.1 performance comparable with OpenCL 1.0  big_smile

Re: SVP Observations & issues..

Vladdo
New 12.4 really improves speed on 5xxx and 6xxx. What's about your issue? smile

  • Thanks: 1

Re: SVP Observations & issues..

It's still quite slow when using that benchmark.. might try a reformat at some stage and see if it fixes the issue.

Re: SVP Observations & issues..

The only 7770 in our database has awful scores below 5570  hmm but this's with 12.3.
May be it's our issue but I can do nothing with this now cause I don't have access to any GCN card sad

10 Reply by Vladdo (edited by Vladdo 02-05-2012 11:56:05)

  • Thanks: 1

Re: SVP Observations & issues..

What is the benchmark codebase? 3.11? or 3.06?

I find that 3.06 works quite well with no slowdowns etc.. but 3.11 turns into a slideshow after a few mins..

Is there a 3.06 benchmark?  perhaps I could run both and compare?

Re: SVP Observations & issues..

It's 3.0.6-based only.

Have you tried "bilinear" option in 3.1.1?

  • Thanks: 1

Re: SVP Observations & issues..

No.. i haven't tried bilinear.. will do..

Here's my updated svpmark with 12.4 drivers..


Test summary
-----------------------
  Date: 2012-05-02T22:42:50
  CPU:  Intel Core i7-2600K @3392 MHz [8 threads]
  GPU:  AMD/ATI Radeon HD 7900 [ver.1.4.1720]
  Mode: FHD + GPU [17 threads]

Overall scores
-----------------------
  Synthetic CPU:                  MC2182
  Synthetic GPU:                  MG3694
  Real-life:                      FG2777

Details: synthetic
-----------------------
  CPU: compose (single-threaded): 601
  CPU: compose (multi-threaded):  2667
  CPU: search (single-threaded):  446
  CPU: search (multi-threaded):   1847
  GPU: system -> GPU transfer:    673
  GPU: GPU -> system transfer:    840
  GPU: calculations:              2295
  GPU: total score:               4503

Details: real-life /FHD
-----------------------
  decode video:                   16.07x (385.6 fps)
  48 fps - vectors search:        1.43x (68.8 fps)
  60 fps - frame composition:     3.12x (187.1 fps)
  48 fps - [SVP] fastest:         6.19x (297.2 fps)
  48 fps - [SVP] simple 1:        4.47x (214.4 fps)
  60 fps - [SVP] good:            2.41x (144.7 fps)
  60 fps - [SVP] high:            1.91x (114.8 fps)
  60 fps - [SVP] highest:         0.90x (53.7 fps)
  72 fps - [SVP] simple 2:        4.01x (288.9 fps)

Re: SVP Observations & issues..

Vladdo
Here's my updated svpmark with 12.4 drivers..

Yeah, it looks better, but they (AMD) obviously have more work to do  big_smile