Here is an end of year update on recent progress and activity with VP9.
Encoder
A rate distortion bug was uncovered that hurt lossless mode and potentially some other very low Q encodes. Tests showed compression gains averaging about 8% for lossless encodes with some clips as high as 25%.
Rate control, RD and other improvements have yielded gains in quality averaging ~4% across our various test sets in VBR mode (though only ~1% for the smaller CIF derf set).
Some further small across-the-board encoder speed gains.
Tiled encoding of new content in YouTube. (Recode of older content Q1).
(480p, 720p 2 tiles, 1080p 4 tiles, 4K 8 tiles : (360p 2 tiles coming in the new year))
Decoder
Desktop Chrome M32 decode supports independent decode of column tiles using different threads / cores, though the loop filter for the whole frame is currently still running on a single thread.
VP9 decoding support now REQUIRED in Android KitKat Compatibility Definition Document (CDD).
Decode support added to latest version of Chrome on Android (4.4.2 update + Clank M33):
Up to 1080p/30 decode on Nexus 5
Up to 720P/30 decode on Nexus 7 (2013 version) and Nexus 4
Up to 480P/30 decode on Nexus 7 (2012 version)
HW progress
VP9 decoder RTL (codename “G2”) has been completed and launched in WebM (http://www.webmproject.org/hardware/vp9/). RTL has already been shipped to a number of SoC vendors developing chipsets for tablets and smart TVs, all preparing to ship silicon in 2014.
G2 decoder is a single core design and can do up to 4K@60fps decode with:
500MHz clock frequency (TSMC 28nm)
2.35Mgates silicon area
--
You received this message because you are subscribed to the Google Groups "WebM Discussion" group.
To unsubscribe from this group and stop receiving emails from it, send an email to webm-discuss...@webmproject.org.
To post to this group, send email to webm-d...@webmproject.org.
Visit this group at http://groups.google.com/a/webmproject.org/group/webm-discuss/.
For more options, visit https://groups.google.com/a/webmproject.org/groups/opt_out.
-Adithya
I am working on a multicore software implementation of VP9 decoder. And I'll have design benefits if I assume that a 1080p clip in YouTube will "always" have more than one column tile.
I see in the decoder code that the maximum width of a tile can be upto 4k pixels. Can you please advice if in the context of YouTube content, there will always be >1 column tiles? (its 4 as per your mail, But just want to confirm if there will be corner case clips with just a single tile, or is it a safe design assumption that every 1080p clip will have 4 tiles). Thanks a lot!
-Adithya