View Full Version : vivante V's IMG
Been, looking at vivante's Graphics IP solutions, they seem to have started to get a bit of traction lately, albeit with 3rd tier companies (fujitsu aside). If the data is accurate they look quite competitive against SGX.
Ti's implementation of SGX530 in Omap3530 is rated at 10M triangles per sec, and according to some work done by Arun, it takes around 100mW @ 100Mhz.
Vivante's OpenGl es2.0 GC600 also does 10M triangles per sec, but this time @ 200Mhz clock. However it is quoted as having an active power of only 63mW @ 200Mhz, which is substantially less than the figure that Arun extrpolated for SGX530 @ 100Mhz.
http://www.vivantecorp.com/p_mvr.html
So same performance at twice the clock, and 50% less power
Its also 4.0mm2 fully routed at 65nm, which seems highly competitive too.
I note that Vivante are also claiming to have the world's smallest OpenGl es2.0 GPU @ 2mm2.
http://www.ocpip.org/pressroom/schedule/speaking/papers_presentations/Vivante_DATE_Slides.20090416.pdf
In November IMG claimed to have the smallest such core, the SGX520, @ 2.6mm2.
http://www.imgtec.com/News/Release/index.asp?NewsID=411
Comparing Vivante's marketing claims, which are probably projections from their IP, to the worst-case numbers of one real-world SGX product won't lend much insight.
Vivante's architecture does seem interesting, though.
Simon F
18-Jun-2009, 09:09
I note that Vivante are also claiming to have the world's smallest OpenGl es2.0 GPU @ 2mm2.
http://www.ocpip.org/pressroom/schedule/speaking/papers_presentations/Vivante_DATE_Slides.20090416.pdf
/me looks at bottom left of page 4 and nearly coats monitor with coffee....
Ailuros
18-Jun-2009, 09:48
/me looks at bottom left of page 4 and nearly coats monitor with coffee....
Is that an iPhone?
Laurent06
18-Jun-2009, 09:56
Didn't Intel also use some iPhone picture in their initial presentation of the Atom?
Is that an iPhone?
...running Windows Vista?
Comparing Vivante's marketing claims, which are probably projections from their IP, to the worst-case numbers of one real-world SGX product won't lend much insight.
Vivante's architecture does seem interesting, though.
I'm as big a fan of IMG as it gets, but I'm not doing totally unreasonable compares.
SGX520 isn't in silicon (or if it is, its not in the public realm), last I heard circa Nov last year, via IMG was that it was "in design".
Vivante on their website quote both synthesis and fully routed areas, its not clear from IMG's table and/or marketing whether its synthesis or routed areas they are quoting. It is also of course not clear how accurate vivantes figures are.
Fujitsu has just announced an OpenGl es.20 Soc, which I assume is vivantes as they are a major backer, but nothing in there that says which variant they are using.
http://www.fujitsu.com/sg/news/pr/fmal_20090310.html
With regard to using iphone in their PR, it seems to be habit that EVERYONE has to have a picture of it somewhere in their bumf.
Nice of them to put at the bottom of page 4 "Stock photos (not actual products)"
the last page of that PR .pdf does boldly state:-
"Vivante guarantees our GPU to meet or exceed stated performance in silicon
Vivante guarantees our GPU to meet or exceed stated area in silicon"
Simon F
18-Jun-2009, 11:10
Vivante guarantees our GPU to meet or exceed stated area in silicon"
LOL!
I get the impression someone is not totally re-assured by the wording of the guarantee.
:smile:
Simon F
18-Jun-2009, 12:31
I get the impression someone is not totally re-assured by the wording of the guarantee.
I just wish I had a target like that :-)
Rob Evans
18-Jun-2009, 20:25
I just wish I had a target like that :-)
LOL!!
Ailuros
19-Jun-2009, 13:40
I just wish I had a target like that :-)
Wouldn't we all wish to have as flexible targets at our work (irrelevant of field...)?
Ailuros
19-Jun-2009, 13:45
SGX520 isn't in silicon (or if it is, its not in the public realm), last I heard circa Nov last year, via IMG was that it was "in design".
I thought iPhone 3GS uses it?
That has not been confirmed anywhere, my guess/assesment is that its SGX530/535.
The teardown today shows that the date code on the app pro chip in the 3GS as 0919, which puts it in the first week of May. I don't think it could go from "design" (remember the statement said SGX540 was in tape-out, but SGX was in design) In Oct, to being incorporated in an App Pro chip in May.
http://www.rapidrepair.com/guides/iphone-3g-s-repair/iphone-3g-s-dissasembly-repair-guide.html
http://www.imgtec.com/news/Release/index.asp?NewsID=400
I previously said this statement was in Nov, it was actually Oct.
They did announce 520 had passed opengles2.0 testing 20-Nov, but I don't know if that was in FPGA form or some test silicon.
http://www.imgtec.com/news/Release/index.asp?NewsID=411
Ailuros
19-Jun-2009, 16:50
As I said before in another thread, Anand cannot possibly have gotten most of his details from IMG's public whitepapers in his 3GS article. He has been guided by an IMG employee for sure IMHLO for that article and I don't think it's a lucky guess from Anand's side either. Frequency maybe yes.
Besides a 520 matches Apple's keep it thin an light for the iPhone much more than any other current SGX variant.
Vastly OT for this thread but nonetheless interesting:
http://www.youtube.com/watch?v=rGof1tPqIBE
For what it is worth when I start debuggin code on the latest iPhone I see this line going by ...
Unable to read symbols for /bla/bla/bla/bla/Extensions/IMGSGX535GLDriver.bundle.
So it does look like it is 535
.. or perhaps this is a common driver for all SGX versions or something along these lines.
So how many SOC vendors are there? They're all using ARM cores with their own GPU, memory controllers, etc.?
Which one has the biggest market share?
It sounds like the performance characteristics, including power consumption envelope, are not transparent so it's not easy to benchmark them with publicly available info.
Ailuros
21-Jun-2009, 05:49
.. or perhaps this is a common driver for all SGX versions or something along these lines.
Could be; besides SGX535 has 2 TMUs vs. 1 TMU on SGX530. Both have 2 ALUs or USSE "pipes".
For the 3GS, all we have to worry about is the single USSE configuration.
http://www.anandtech.com/gadgets/showdoc.aspx?i=3579&p=3
Ailuros
21-Jun-2009, 06:00
So how many SOC vendors are there? They're all using ARM cores with their own GPU, memory controllers, etc.?
Pretty much yes. The first question is a bit tricky to answer since I wouldn't place firms like Imagination or Vivante here amongst "SoC vendors" since both are merely designing and selling IP apparently. For graphics the firms that currently are active in the market (and I'm sure I've forgotten some):
Imagination Technologies (IP only)
NVIDIA (SoC manufacturing & some selected IP)
ARM (graphics IP from the former Falanx group)
Qualcolmm (former ATI handheld including the former Bitboys OY)
Vivante
I think Toshiba had also something cooking but I'm not aware where the graphics come from...
Which one has the biggest market share?
Considering the success of MBX and today's SGX for two generations from the list above IMG. With Intel, Texas Instruments and Apple under wraps for both generations it's a simple answer.
It sounds like the performance characteristics, including power consumption envelope, are not transparent so it's not easy to benchmark them with publicly available info.
Well any IP selling company can include in it's marketing material whatever claims; the closer their "estimates" are in the pre silicon stage to the final silicon stage the better for them and their potential customers.
That said if Vivante's claims are damn close to reality it's truly a pretty impressive design.
Imagination Technologies (IP only)
NVIDIA (SoC manufacturing & some selected IP)
ARM (graphics IP from the former Falanx group)
Qualcolmm (former ATI handheld including the former Bitboys OY)
Vivante
I think Samsung has its own IP for graphics too.
Ailuros
22-Jun-2009, 13:22
I think Samsung has its own IP for graphics too.
Which one? The one they're using in their SoC for the iPhone?
Toshiba's graphics have been designed in-house too, I believe.
Which one? The one they're using in their SoC for the iPhone?
The one in the S3C6410 possibly.
http://www.samsung.com/global/business/semiconductor/support/brochures/downloads/systemlsi/s3c6410_datasheet_200804.pdf
4M Tri/second
Open GL ES 1.1/2.0
I think Arun said he spoke to Samsung at a trade show last year and they indicated to him that it was their own solution.
Note that the published S5PC100 docs simply confirms the existence of a "2D/3D" block in the block diagram. and makes zero reference to the graphics capabilities in the narrative either relating to performance or standards compliance information.
http://www.samsung.com/global/business/semiconductor/support/brochures/downloads/systemlsi/s5pc100_brochure_200902.pdf
Ailuros
23-Jun-2009, 07:14
The one in the S3C6410 possibly.
http://www.samsung.com/global/business/semiconductor/support/brochures/downloads/systemlsi/s3c6410_datasheet_200804.pdf
4M Tri/second
Open GL ES 1.1/2.0
I think Arun said he spoke to Samsung at a trade show last year and they indicated to them that it was their own solution.
Note that the published S5PC100 docs simply confirms the existence of a "2D/3D" block in the block diagram. and makes zero reference to the graphics capabilities in the narrative either relating to performance or standards compliance information.
http://www.samsung.com/global/business/semiconductor/support/brochures/downloads/systemlsi/s5pc100_brochure_200902.pdf
That's weird; so the most likely reason why they've originally licensed anything SGX was due to Apple requesting it?
That's weird; so the most likely reason why they've originally licensed anything SGX was due to Apple requesting it?
Or maybe its to do with :-
"4M Tri/second"
For more info on the "samsung" IP, check out this thread:-
http://forum.beyond3d.com/showthread.php?t=47776
Ailuros
23-Jun-2009, 12:44
Or maybe its to do with :-
"4M Tri/second"
For more info on the "samsung" IP, check out this thread:-
http://forum.beyond3d.com/showthread.php?t=47776
Quoting theoretical triangles rates is in my mind as useful as the meaningless GFLOP rates in the standalone GPU market lately.
Arun's worries about the fillrate might be a serious drawback of the chip; the quoted bilinear fillrate with alpha blending is abysmally low.
The real question is if Samsung actually uses those cores for their own phones after all. Isn't the OMNIA HD using TI's OMAP3 SoC?
Yes, the Omnia HD (8910) is Omap3. It predecessor, the 8510 is Omap2.
never really heard it being used much, designed into some acer phones:-
http://www.pocketpicks.co.uk/latest/index.php/tag/s3c6410/
and some photo frames
http://www.chumby.com/pages/media_090108_samsung
I also recall hearing something about people hosting an android phone on it.
vBulletin® v3.8.6, Copyright ©2000-2013, Jelsoft Enterprises Ltd.