Bitboys Licenses Graphics Processor Technology from Ericsson
- Thread starter Aeros405
- Start date
Re: Bitboys Licenses Graphics Processor Technology from Eric
So they licenced not only FLIPQUAD but also PACKMAN. Oh boy's!
This will certainly mean that they also use the POOMA texturing system for (sort of) trilinear filtering, otherwise the PACKMAN texture compression would make no sense. I hope this texture filtering looks good enough but I fear that the quality of the filtering will be quite low on the new G3x and G4x hardware from Bitboys even compared to the bilinear filtering on other hardware.
So they licenced not only FLIPQUAD but also PACKMAN. Oh boy's!
This will certainly mean that they also use the POOMA texturing system for (sort of) trilinear filtering, otherwise the PACKMAN texture compression would make no sense. I hope this texture filtering looks good enough but I fear that the quality of the filtering will be quite low on the new G3x and G4x hardware from Bitboys even compared to the bilinear filtering on other hardware.
Ailuros said:
Hi Ailuros;
no problem. Look here: http://graphics.cs.lth.se/research/mobile/
In this PDF you can find an good description of all the new things Ericsson invented.
The rather big presentation [26MB] is also really nice : http://graphics.cs.lth.se/research/mobile/akeninestrom_ppt.zip
They have included a few mpg's to show the effects of Flipquad and POOMA in "realtime".
SiBoy said:
supposing stuff isn't healthy on forums like this. way too much of lurkers might understand that as a fact. (really, believe or not, it does not depend much who posts it. if it's here, it's a fact to some ppl.)
to what comes to failing on mobile, I don't estimate it yet. Still, NEC licensing G34 straight away doesn't really make it look that way. Bitboys Desktop parts has basically nothing to do with their mobile line up. Their ideas / technology behind these lineups is completely different.
Well I can understand that there are totally different requirements for the PC desktop and for mobile/PDA and I doubt it's any different at any other IHVs be it ATI, NVIDIA, IMG or whoever else.
Bandwidth, size and power consumption are a consideration for them all.
For the time being - keeping the above restrictions in mind - I doubt we'll see full unoptimized trilinear, no TC, and high sample adaptive AF on these accelerators in games anytime soon (despite what featurelists claim).
Whatever differences accelerators will show in real time, we'll see them when devices with them start to ship more widely.
Considering FLIPQUAD, it looks like a pretty reasonable idea. What I don't like about that method or Quincunx as another example is the fact that additional filters apply on all the scene and not just polygon edges/intersections. Additionally any form of Multisampling should be accompanied with fast performing at least 2xAF, otherwise I'd rather have plain old Supersampling instead.
Doesn't tell me much either. Could be many reasons why NEC decided to, which frankly I'm not even aware of. I'd step into the same dead end while wondering for example why Qualcomm chose Imageon in the recent past.
Bandwidth, size and power consumption are a consideration for them all.
For the time being - keeping the above restrictions in mind - I doubt we'll see full unoptimized trilinear, no TC, and high sample adaptive AF on these accelerators in games anytime soon (despite what featurelists claim).
Whatever differences accelerators will show in real time, we'll see them when devices with them start to ship more widely.
Considering FLIPQUAD, it looks like a pretty reasonable idea. What I don't like about that method or Quincunx as another example is the fact that additional filters apply on all the scene and not just polygon edges/intersections. Additionally any form of Multisampling should be accompanied with fast performing at least 2xAF, otherwise I'd rather have plain old Supersampling instead.
Doesn't tell me much either. Could be many reasons why NEC decided to, which frankly I'm not even aware of. I'd step into the same dead end while wondering for example why Qualcomm chose Imageon in the recent past.
Nappe1 said:
Sounds like you are agreeing with me.
Just making the observation that when BB reinvented themselves in the mobile space (which could have meant they saw some leverage from their previous work in this new market), they instead had to buy the "ideas / technology" elsewhere.
SiBoy said:
actually no. I see it as they are such a small company and their resources are limited, there's no use of re-inventing something, that can be lisenced. (most likely with same or even less expenses than inveting something similar by themselves. Plus, using available tech instead of making another texture compression algo. Also, if you invent something new, you have to try to get APIs to support it. in case of licensing, the rights owner is already doing this, because he likes to make some money of the licenses and no one is going to buy tech that has no API support.)
And as they stated several years ago, "NIH thinking in this business is way too common." and what else is better way to start fixing situation than start it on your own company?
Frankly it does sound a bit of an oxymoron when an IP selling company, licenses IP from others. 
(cross-licensing though is quite common)
Jokes aside it sounds like a smart move to me; I'd be very surprised if future Sony/Ericcson mobile devices wouldn't contain any BB accelerators at all.
(cross-licensing though is quite common)Jokes aside it sounds like a smart move to me; I'd be very surprised if future Sony/Ericcson mobile devices wouldn't contain any BB accelerators at all.
PACKMAN isn't only texture compression they support.
DXT1 (which is S3TC, right?) is also supported by G32,G34 and G40, acording their product comparison chart.
http://www.bitboys.com/comparison.php
DXT1 (which is S3TC, right?) is also supported by G32,G34 and G40, acording their product comparison chart.
http://www.bitboys.com/comparison.php
Simon F said:
I think they already have S3TC license. Axe supported DXT1 as well as G30 also supports it. (the previous gen. chip.) They just didn't thought it was big enough thing to make press release. Again, BB is the first one supporting PACKMAN and FLIPQUAD, so that again is a bigger news than supporting several years old (already substandard status reached) DXT1.
OK, I've gone and dug up the "technical sketch" for Packman...
The quality of Packman is slighly lower than than of DXT1. The advantage, however, is that it is cheaper to implement. Although still 4bpp, the blocks are reduced to 2x4 pixels and thus only require 32bits of storage. This is useful for mobile devices which possibly only have a 32bit bus. Also the decode appears to be very simple.
Nappe1 said:
The quality of Packman is slighly lower than than of DXT1. The advantage, however, is that it is cheaper to implement. Although still 4bpp, the blocks are reduced to 2x4 pixels and thus only require 32bits of storage. This is useful for mobile devices which possibly only have a 32bit bus. Also the decode appears to be very simple.
