25. We realise that there's ongoing research for smaller manufacturing
processes such as 90nm and 65nm in the more distant future. Die shrinks obviously
suggest the ability to increase clock speeds and of course more headroom for
further feature additions. Are things such as a stencil buffer and/or cube map
support amongst the priorities for future incarnations of MBX like MBX2 or MBX3?
John: The feature set will increase for future generations but it's too early
to say specifically how. That said, the types of features you indicate would
26. Considering that power consumption and heat dissipation will always
be a major concern with embedded devices, how long would you estimate that the
integration of advanced programmable shading capabilities into those types of
devices will take?
John: The main lesson of the last 40+ years is that increased silicon integration
and features is relentless; support for full shading support will inevitably
come, but there are various opinions on when.
27. Are any steps being taken to ensure that MBX remains the market
John: We are building further alliances and working on expanding our partnerships
to further tier one silicon companies. With the partners we have - Intel, TI,
Renesas etc - we are already in a very strong position and fully intend to build
28. Do you feel threatened by the competition with regard to 3D capable
PDA/mobile devices, even with large IHVs such as NVIDIA/ATI/Bitboys OY etc.
entering those markets? (see MediaQ acquisition by NVIDIA).
John: We feel vindicated as much as threatened by the emergence of potential
competition in this area. We would never write-off ATI or Nvidia, who obviously
continue to be very successful in the PC space. However no other graphics company
has yet shown the ability to bring the same performance and features to the
29. Would the assumption that in some cases ImgTec is metaphorically
pulling the carpet out from under some competitors’ feet be an exaggeration?
John: Well, at the minute PowerVR MBX is the only high-performance choice for
30. FSAA4free(tm) is part of MBX's featureset. Considering Anti-aliasing
comes essentially for free in the x resolution, while having a minor fillrate
penalty in the y resolution, one would assume that MBX uses Supersampling. Is
it safe to assume that the same marketing term will make its appearance in future
products too, yet not necessary with the same sampling approach?
David: I can't say at this time.
31. Is there any interest for addressing the mobile and integrated/chipset
market in the future, under the presupposition that potential clients/licensees
will request such IP?
John: I assume you mean the PC integrated chipset market - we believe we have
the perfect technology for this market given our ability to deliver performance
and features without using vast amounts of bandwidth, so yes, we do see it as
a potential market.
32. The recent breakthrough from the Metagence division concerning
Linux ports on multithreading capable SoCs didn't go unnoticed, which would
suggest ongoing research for Linux for all platforms. End users have already
voiced interest if future PC products will be accompanied with Linux drivers
at launch. Anything you'd like to tell them?
David: Linux is a key area of interest for us and we intend to support it appropriately
across our technology divisions.
33. One of the recent ImgTec patents deals with a copy protection
scheme. Could you elaborate what its purpose and target is?
John: The patent deals with another element of DRM and how a publisher protects
their intellectual property through distribution to the end user.
34. What is your opinion on virtual texturing? Wouldn't it make sense
to use a similar approach on a deferred renderer, especially for the cases where
textures flood the local memory?
John: Virtual texturing is an interesting technique and one which tile based
rendering can take full advantage of, given that a tile based renderer can look ahead
a full scene to what textures are required (and in which tile).
35. Do you think that 32bit floating point accuracy in pixel shaders
are or will become a necessity soon?
John: That depends on the target market. 32bit precision is, as ever, a trade-off
between size, performance and quality. I would expect to see a range of solutions
for pixel shader accuracy for some time to come.
36. Judging from recent benchmark results (such as HL2 f.e.), do you
think that applications/games are tending to become more shader limited and
far less bandwidth limited than before? Do you predict that under those presuppositions
future PowerVR products to be able to keep pace with the competition?
John: Different benchmarks have tended to stress different performance elements
and while pixel shader performance is an important factor, bandwidth will remain
top, or very close to top, of the list of factors affecting overall performance.
One thing to note of course, is that if pixel shading is the bottleneck, a deferred
texture engine only shades visible pixels and is therefore much more efficient
than an immediate mode renderer in this respect too.