Topic: Should I download Version 4.0, or stay with 3.17?

I have been getting the pop up to update to Version 4.0 for almost two weeks now, but never got the impulse to do it. But today I thought about it, but then seen that its not a full version, a Technical Preview or something, and it clearly says that very few 4.0 features are added in the TP version.

So the question is, should I stay with old 3.17 or upgrade to 4.0 TP? I haven't donated anywhere, but would still like to know the amount to get the full 4.0 version, if it is available that is. This page doesn't help me to get any information on the availability or the pricing! https://www.indiegogo.com/projects/real … c#/updates

Thanks in advance.

Re: Should I download Version 4.0, or stay with 3.17?

If you're curious how SVP 4 TP fare, you can always try it without uninstalling SVP 3.1.7

I've installed SVP 4 TP twice, and uninstalled it too twice.

The pricing and other information are available at the bottom of story page, or at the TP download page itself.

Re: Should I download Version 4.0, or stay with 3.17?

mashingan
May I know why you uninstalled it?

Re: Should I download Version 4.0, or stay with 3.17?

It's still Technical Preview and the provided interface still basic so I can't edit the setting like I do with SVP 3.1.7.

Since I mainly use SVP 3.1.7 and the SVP 4 will be unused so I just uninstalled it.

The finished product will be lot better, you should try it when the devs finally release it big_smile

Re: Should I download Version 4.0, or stay with 3.17?

Okay, then sticking with 3.1.7, until final 4.0 hits.

Re: Should I download Version 4.0, or stay with 3.17?

It really depends on your individual needs. I personally make use only of the features that already exist in the Technical Preview of SVP 4, and have already seen better results with the smoothed video when compared to 3.1.7, as well as better GPU acceleration overall. So for me using SVP 4 TP is a nobrainer, and I've already upgraded to it on multiple computers with no major issues.

So if you make use of the features in SVP 3.1.7 that haven't been implemented in the Technical Preview then wait. However if you aren't a poweruser and only requires the motion interpolation bit then you really don't waste anything on not trying SVP 4 TP out. SVP 4 requires 3.1.7 to be installed and working anyway, so it's entirely possible to have the two of them installed at the same time and just launching the SVP Manager of your choice.

You can even change the running SVP Manager mid-playback to either revert to v3.1.7 or try out v4 TP on the fly.

Feature list comparison:
http://i.imgur.com/RE2mHPc.png

Re: Should I download Version 4.0, or stay with 3.17?

ithehappy wrote:

Okay, then sticking with 3.1.7, until final 4.0 hits.

I saw this thread and got excited. Checked my email, checked the download page, checked google... Then came back here and saw that it's not out yet. Lol I guess for those of us who paid, we'll get notified via email anyway when it comes out.

Re: Should I download Version 4.0, or stay with 3.17?

Keep 4.0 installed, and just switch between both versions.


4.0 gives me better results/performance on my hardware, but I still switch to 3.1.7 to tweak certain genres (animation, for example).

9 (edited by xenonite 26-08-2015 16:24:43)

Re: Should I download Version 4.0, or stay with 3.17?

SVP 3.1.7 can be configured to provide the exact same performance and "quality" as SVP 4, but the reverse is not true.
SVP4's current settings is optimised for low-end hardware and producing temporally aliased images, which you may or may not like.
So, if you have old/slow/low-power hardware, you won't be able to use any advanced interpolation present in SVP 3.1.7 anyway, so using SVP 4 would not make you lose anything. However, if you have reasonably recent hardware (with a CPU TDP > 80W) AND you like the 'smoother' (read: reduced temporal aliasing) images produced by a 'high-end' configuration of 3.1.7 (see the comparison below), then you shuld stay with SVP 3.1.7 (at least for now).

To give you an idea of the differences, I have created these simple high-contrast 2D images (to reduce the attachment size and better illustrate an exaggerated example of the differences). Imagine this simple translational movement as representing a difficult motion to interpolate (big displacement, occluded pixels, non-uniform lighting, etc), from image a to image b:

                                                      Image a:
a

                                                      Image b:
b

Ps. Continue on next post because of 3-attachment limit (need 4 images to display the difference)

Post's attachments

a.png, 2.28 kb, 512 x 512
a.png 2.28 kb, 429 downloads since 2015-08-26 

b.png, 2.28 kb, 512 x 512
b.png 2.28 kb, 345 downloads since 2015-08-26 

10 (edited by xenonite 26-08-2015 16:34:16)

Re: Should I download Version 4.0, or stay with 3.17?

(Continued from above)

Then these images approximately show the different 'style' of interpolated (simple frame doubling) frames produced by the two SVPs:

                                                        SVP 3.1.7
SVP317

                                                          SVP 4
SVP4

Basically, SVP 4 produces 'jumpy' images (like watching a 24fps video on a 48hz el-cheapo lcd with very slow pixel response times of >20ms), while SVP 3.1.7 can be configured to produce more 'wavy' artifacts, but actually moves the objects in the frame to produce a new frame with the objects in-between their respective locations in the previous and next frames (which I find to be the whole point of motion interpolation, but some {most?} people seem to prefer the SVP 4 way).

This does, however, only affect difficult motion areas. In places without much temporal aliasing to begin with, both SVPs perform very well and generate a proper interpolated image.

Post's attachments

interp317.png, 8.52 kb, 512 x 512
interp317.png 8.52 kb, 427 downloads since 2015-08-26 

interp4.png, 2.11 kb, 512 x 512
interp4.png 2.11 kb, 310 downloads since 2015-08-26 

11 (edited by Nintendo Maniac 64 26-08-2015 20:49:06)

Re: Should I download Version 4.0, or stay with 3.17?

xenonite wrote:

SVP4's current settings is optimised for low-end hardware

Your definition of "low-end" hardware must not be the same as mine, because on my 2.4GHz Core 2 Duo without any GPU acceleration nor HW decoding, I can get much less artifacting with SVP 3.1.7 due to the fact that SVP 4 insists on using a higher-end interpolation mode like "Adaptive" combined with a low-quality shader setting like "2. Sharp" along with a low-resolution vectors grid like 28px; meanwhile in SVP 3.1.7 I can choose to instead use 1m + Standard + 16px.

Thread I made regarding the subject:
http://www.svp-team.com/forum/viewtopic.php?id=2699

xenonite wrote:

if you have reasonably recent hardware (with a CPU TDP > 80W)

TDP does not equal performance - and easy example is Broadwell which is only 65w TDP yet has slightly better IPC than Haswell and has the fastest iGPU Intel has ever made.

Heck, even the i7-3770k has a TDP of only 77w.

12 (edited by xenonite 26-08-2015 22:41:54)

Re: Should I download Version 4.0, or stay with 3.17?

Nintendo Maniac 64 wrote:

TDP does not equal performance

Yes, in general you are correct.

What I was trying to explain, was a easy metric that people (who do not know about the relationship between architecture, IPC, transistor dimensional scaling, core count and clock frequency) could use to quickly determine if their PC has the potential to run SVP 3.1.7 at a higher quality than SVP 4.
With all "relatively recent" (i.e. released with or after the Sandy-Bridge generation) CPUs being gimp'ed by consumer opposition against powerful processors, all of the 'high TDP' processors are performing relatively similar (compared to the generational improvements before), especially in SVP (a 5-10% difference would not reliably make a higher quality level attainable).
With that in mind, neither clock speed, nor model number, nor core or cache count, nor any other single parameter that is listed on the physical packaging or on the official online Intel "Tech specs" listing correlates as well with the CPU's potential SVP performance as TDP does.

Nintendo Maniac 64 wrote:

[...] easy example is Broadwell which is only 65w TDP yet has slightly better IPC than Haswell and has the fastest iGPU Intel has ever made.

True, but if someone were to try to use the IGP while running SVP on the CPU (perhaps even using the IGP for SVP), then there might not be enough TDP budget left for a heavy SVP 3.1.7 configuration.

Also, I never said that if CPU A's TDP > CPU B's TDP >= 80W; THEN CPU A's performance = CPU B's performance * f(CPU A's TDP / CPU B's TDP), where f is a linear or sub-linear function.
I understand that AMD doesn't have access to Intel's advanced process nodes, so their TDPs will be significantly higher simply due to the manufacturing node being optimised for low-TDP parts (<50W) and their CPUs being designed to run at high clock speeds, meant that they had to use a significantly higher voltage than the process was originally intended for (and power scales with the square of voltage). Also, TDP figures are not comparable between manufacturers. So, yes, TDP is not a perfect performance measure by a long shot, but what other single 'spec sheet' number is? Maybe price, lol.

Anyway, I simply thought that someone who doesn't know whether to choose SVP 3.1.7 or SVP 4 could quickly look up the TDP of their processor, since it is the single most important parameter influencing modern consumer purchases (allegedly). If their PC is less than about 4.5 years old and they have a CPU TDP > 80W, then they should be able to use higher quality settings in SVP 3.1.7. And that was all that I meant by it.

Nintendo Maniac 64 wrote:

Thread I made regarding the subject:
http://www.svp-team.com/forum/viewtopic.php?id=2699

Thank you for pointing me to that thread; I never would have believed that a Core 2 Duo can run SVP at all, let alone as well as you described.
Very interesting indeed.

Ps. It seems that Chainik has indicated that, once SVP4 is fully released, you will indeed be able to fully configure all of its settings, just like you can with the current SVP 3.1.7.
So then the question becomes moot, SVP 4 will become better than 3.1.7 in all aspects; a.k.a. progress without regression! big_smile

Re: Should I download Version 4.0, or stay with 3.17?

xenonite wrote:

I never would have believed that a Core 2 Duo can run SVP at all, let alone as well as you described.

The main key is to only use 720p videos and have Motion Vectors Precision set to "Two Pixels"; 16px Average 0 also helps out decently as well (good balance between smoothness and performance).

Re: Should I download Version 4.0, or stay with 3.17?

Nintendo Maniac 64 wrote:

16px Average 0 also helps out decently as well (good balance between smoothness and performance)

Cool. I believe the computational complexity increases quadratically, i.e. O(N^2), while the quality degradation is extremely subjective (such as very small blocks actually degrading the image quality), but probably only degrades linearly with increasing size.
Now you have given me ideas; I am busy disabling most of my cores through the bios and underclocking to 1.8GHz to see what the best image is that I can produce under those circumstances.  lol

15 (edited by Nintendo Maniac 64 20-09-2015 21:36:00)

Re: Should I download Version 4.0, or stay with 3.17?

Another thing I forgot - using D3D Fullscreen and only using Bilinear (normal, non-PS 2.0) in MPC-HC also helps reduce CPU utilization...which means getting away with higher settings in SVP.

Take note that the settings I posted are optimized for minimal artifact since the 25fps content I watch is highly susceptible to said artifacts (motorsports).  For 24fps and 30fps I normally use Uniform + Simple Lite instead which is actually smoother but gives more artifacts, though they're not really noticeable on the 24fps and 30fps content I watch.

Another key thing is that, since you're doing software decoding of the video stream itself, using a codec that's light on the CPU (i.e. MPEG2, FLV1, etc) can allow you to use higher settings in SVP.  In particular, I can get away with using Uniform + Standard for 30fps, 25fps, and 24fps content in such situations.  Heck, if I don't plan on watching a video for a while, I'll even sometimes re-encode a video, such as a 1080p VP9 video from YouTube, into 720p FLV1 @ Quantizer 2 (the filesize is like 6-8x larger, but it doesn't matter when I don't plan on keeping it anyway)


This is all assuming 720p videos and 60hz monitor refresh rate of course.


EDIT: Just remembered one last thing - you can also save a few more percent of CPU utilization by making sure that yous "Default Sampling Rate" in the Windows control panel matches the sampling rate of the audio stream in whatever video you're watching.

EDIT 2: For most use-cases H.263 is probably the best choice in terms of re-encoding since, at Quantizer 2, if only has 2% more CPU utilization than FLV1 on SVP-intensive content and 8% smaller filesize, plus for other devices it has compatibility with the MP4 container not to mention HW decoding capability if available.

EDIT 3: ...except if you're encoding directly from VP9 via Avidemux, then there's some weird issues with SVP not working correctly with specific codec-container combinations.  The ones I find work are Mpeg4 ASP (xvid4) with MPEG Quantization or Mpeg4 ASP (ff), both in an MP4 container (and only an MP4 container, using MKVtoolnix after-the-fact does not work correctly) or FLV1 into an FLV container (which does work with MKVtoolnix, so that'd be the most flexible).  Just remember that Avidemux doesn't like Opus audio, so you usually have to deselect it via "Audio" -> "Select Audio"; for MP4 you'll need to provide an external AAC/M4A/MP4/MP3/AC3 audio stream for Avidemux to add to the video.

EDIT 4: Just realized another one - setting processing threads in SVP to "5 threads" seems to give the lowest CPU utilization without resulting in decreased SVP performance (like if you set "2 threads", SVP won't run at full speed even when your CPU utilization is like 30%).

EDIT 5: Go figure, it turns out that non-PS 2.0 bilinear in MPC-HC is actually less CPU-intensive with weak GPUs, so use that instead!