PowerVR Shows The Way at ARM DevCon 2004

Early-Z

Sounds good on paper, but I get suspicious when I hear too bold claims. Current high end PC sollutions are still performing only essentially 2xAA samples per clock and use sample loops to achieve higher sample densities. Given the limited die area and the claimed power consumption I severely doubt that falanx has been able to achieve 4xAA (whatever it is) in a single clock. Much much more even 16xAA.

DeltaChrome comes to mind with it's original 16xMSAA claim, which ended up being limited to 2xRGSS after all; and wait that's even a PC part...

 

That would be me.

Our Mali cores use tile-based rendering (albeit quite different from that of ImgTech), with which we found that 1-cycle 4X RGMS was very cheap to implement. The 16X AA mode does come with a performance penalty, though.

As for how real Mali is, we have one announced design win so far, with a company named Zoran (press release).

 

is the 16AA mode also RG?

 

I know..... I have good memory sometimes.

 

Hang on a sec so what you're saying is that, the MBX on board the SH3707 includes the VGP?

 

Congratulations arjan de lumens and everyone at Phalanx! Is Nokia going to be using the Mali cores?

 

Technical aspects aside, Falanx' screenshots, PDFs, Demos and such are looking much more appealing / polished from an artwork point of view than what PowerVR offers. Same with Bitboys marketing stuff. Scandinavians are really strong in that kind of design.

 

before Swedes and Norwegians gets here to correct you, I think you propably ment Nordic people instead of scandinavians

as finland really isn't part part of scandinavian countries. Even though, we do have a tiny peninsula like area in the northwest reaching the northest parts of the skands.

 

My question would be if it's a deferred renderer; one could call tile based a large multitude of architectures and not only today's designs.

Single cycle 4xRGMS sounds highly promising, yet I don't seem to see anything concerning anisotropic filtering. I realise that it's about small handheld devices, but even then I'd rather have 2xRGMS + 2xAF. Non-adaptive and not performance optimized AF though might throw performance back to 2xSSAA level though.

What made me really sit up is the VS2.0 compliance of the geometry processor. Any chance we could get any further details on that one? (it never hurts to ask, even if it's unlikely to get an answer ).

 

W/o being sure, I think Intel is the only or amongst the few licensees that hasn't a VGP or VGP Lite aside any MBX/MBX Lite.

 

I would doubt it. Which Renesas chip are you referring to?

As far as I know they licensed MBX for two devices, one targeted at car navigation, one for SH-Mobile3. In the former case at least, Renesas made a big deal about using their own 4-way SIMD floating point coprocessor (attached to their CPU, not to the MBX) for transforms, etc.

At least until they get rid of their homegrown CPU and switch to ARM I would expect this approach to continue.

 

Phhht, I don't know, that was months ago! However, given Simon's replay I think I'll assume it does.

 

I've only heard of one embodiment of MBX/MBX-Lite that doesn't have a VGP.

[UPDATE] I've been told there are couple that are "sans" VGP <shrug> [/UPDATE]

 

Most recent announcement about SH-3 mobile:

http://www.renesas.com/fmwk.jsp?cnt...=/company_info/news_and_events/press_releases

SH-73180:
http://www.imgtec.com/news/Release/index.asp?ID=204

SH-7770 (SH-4a):
http://www.imgtec.com/news/Release/index.asp?ID=189

SH-3707:
http://www.extremetech.com/article2/0,1558,1667467,00.asp

I think the first announcement about the license agreement between IMG and Hitachi was in February 2003 and it must have been prior to the Hitachi/Mitsubishi=Renesas merging.

It of course nowhere mentions in any of those or any other announcements whether a VGP Lite is included, yet it still doesn't mean much. In fact it's the first time I see frequencies mentioned in the first link above. SH-Mobile3/MBX Lite@54MHz.

What exactly is "attached" in a SoC anyway?

 

Read the paper from last year's ISSCC.

 

Couldn't find anything related at least not on Renesas' homepage (only older press releases from past years or I didn't know where to search).

In any case the only other related note I was able to find was the following:

http://www.renesas.com/fmwk.jsp?cnt...=/company_info/news_and_events/press_releases

I haven't a single clue what they could mean with "both graphics engines", unless they mean the PowerVR M24VA, which might have found it's way into that one too.

http://www.powervr.com/Downloads/Factsheets/PowerVRM24VA.pdf

 

SiBoy:

Renesas just took over the SuperH licensing business from STMicro, so I don't think they're planning on replacing it with something else.

Ailuros:

The Renesas statement seems to mean that there's geometry acceleration for both the 2D and 3D graphics engines of the SH7770.

 

If there would be only a seperate 2D/3D engine on a MBX:

http://www.powervr.com/Products/Graphics/MBXLite/index.asp

For video/mpeg acceleration I'm pretty confident that semis have also licensed the prementioned M24VA and/or M2VX from IMG for multimedia applications.

 

They appear to have added a distinct 2D engine to the SH-4A core for drawing consistent, anti-aliased lines in maps. For the SH7770, though, it does seem like they could've used the MBX's 3D to do the 2D.

 

Based on what?