Conservative auto-selection of SVP profiles?

Nintendo Maniac 64
overly conservative

50% tolerance

Pentium G3258 is detected as 2g when in reality it handles 4g just fine

I hardly believe
What is "4g" profile for FHD for example? Those settings allow "SVP Reference system 4" (note "4" at the end ) to playback 1920*1080 video at ~90% CPU load.
"Reference system 4" is 6-cores Phenom II X6 at 3.0 GHz and it gets 1589 SVPmark scores (87 fps in "good" sub-test).
Your G3258 gets 1144 scores and 57 fps.
So may be it can, under some conditions, play 720p with 4g settings but definitely not FullHD.

Does this mean that it only selects settings where the CPU utilization never goes above 50%?

Ok, so it's not as conservative with 1080p, but it's still somewhat.  Using a more typical 1080p video (which in this case is 30fps rather than 24fps), I get ~80% CPU utilization on profile 3c with "Decrease frame size" set to 'Disabled' and all hardware acceleration disabled.  Note that I am interpolating to 90hz.

More typical 1080p video in question:
https://www.youtube.com/watch?v=5I-nQtsJuhE

EDIT: CPU utilization is still only 84% if I use the exact same configuration but with the VP9 version of the video + MPC-HC v1.7.8.

EDIT 2: lolwut, 3c uses arguably higher settings than 2g?  Uh huh.  Well, using the highest settings from both 3c and 2g, I do indeed get 90% CPU utilization.

EDIT 3: This still begs the question though, why are the default 720p settings more conservative than the default 1080p settings in a specific profile?

I'll agree with you on 3g since it does indeed have everything at default or higher.

The problem is that SVP detects my CPU as 2g, and that in particular has "Motion vector grid" set to '14px. Average 1' and the "Wide search" is set to 'Average (default)'; by comparison 3g uses "12px. Average 2 (default)" and 'Strong'.

Also 2g has "Reduce grid step" set to 'Disabled (default)' and 3g has it set to 'By two with local refinement', but my CPU can handle the 4g setting of 'By two with global refinemen'* with performance to spare.

*I just noticed that SVP 3.1.6 + hotfixes does indeed say "global refinemen" without the 't' at the end.  Is this a typo perhaps?

...and it gets 1714 when you include the GPU.  Remember that SVP is recommending profile 2g, not 2c.

EDIT: Derp, I just realized that I never ran the 720p benchmark, I only ran the 1080p one!  I should probably do something about that...

EDIT 2: I have now done something about it.  However, it doesn't seem like my results are appearing just yet.

According to the SVPmark program itself, it says my FullHD 1080p final score is 1119 while my HD 720p final score is 2340.

Well then as long as the value is independent of frame size the recommended profiles will have conservative 720p settings.

EDIT: I understand how the profiles are selected and stuff, and I don't disagree with the 2g selection.  My issue is that, the 720p settings in the 2g profile seem conservative while the 1080p settings in the very same 2g profile seem accurate.  Optimally both the 1080p and 720p settings in a single profile should be accurate should they not?

EDIT 2: While I was at it, I ran SVPmark for FullHD 1080p and got 936 for Synthetic CPU.  Looking at the online results, that puts me on-level with desktop Sandy Bridge i3 CPUs.

Last thing, it would seem that the "Fast SVPmark" in SVP's automatic profile selection is moderately conservative as well.  With my CPU OC'd @ 3.8GHz, the Synthetic CPU for FullHD 1080p in the full SVPmark gives me a score of 1106, but "Fast SVPmark" only gives me a score of 982.

This would explain why, with my 3.8GHz overclock, I am able to do 3c at 1080p with "Reduce frame size" set to 'Disabled (default)' and still have ~20% spare CPU headroom.

Well it seems to be moderately consistant since I did it like 10 times and the highest value I saw it give me was exactly 1000, but I'm guessing you need at least 1001 for it to recommend 3c.

EDIT: And to clarify 982 is the lowest value I saw.

There are a few improvements that could be done to the profiles configuration.

- Making it less conservative; but then there's the risk that it crosses the line and asks too much on some computers

- Allowing different configurations based on the output display size (768p laptop vs 1080p TV output); but then there's the risk of making configuration too complicated. 4k TVs, in particular, may start to cause problems.

- In some cases it gives weird aspect ratios by default on my TV. I'll have to see exactly what refresh rate it is detecting when this problem happens.

- For 120hz displays, using madVR really becomes non-practical since it has only 8ms to process each frame. "Decrease frame size" performs resize functions before processing SVP, see if it's possible to process madVR there so that for a 24fps video on 120hz display, madVR processes only 24 frames with expensive algorithms and SVP has higher quality frames to work with.

Hmm... G3258 has a very good price per money... but that is not very good CPU.

In gamming, This CPU is best CPU I think. Because gamming is depends on GPU. But SVP is depends on CPU.

SVP's GPU limitation(?)(Sorry for bad English ) is ATI HD6850.

Because HD6850, 270X and 280X are not difference in SVP. When I use.

But When I change a CPU (FX6300 -> i7-4790K) I can see higher performance.

Source file : Blu-Ray 10bit AVC FLAC ( Uniform, 23. Complicated, 24 to 60, 8px, Half pixel, Search Radius & Wide Search are highest)