For the potential customers, duh? They're selling IP after all.
PowerVR Athena ? ....a few thoughts
- Thread starter Megadrive1988
- Start date
For the potential customers, duh? They're selling IP after all.
What is high end now, will be in every other cellular phone 3 years down the line.
Cheers
R580 has bilinear TMUs, whereby when I interviewed them last time they said in between lines that SGX has twice the trilinear fillrate then MBX, which suggests trilinear TMUs.
Again those 1.2GPixels/s are effective fillrate. IMHO those are 400 MTexels/s trilinear fillrate; if it would have bi-TMUs like MBX it would of course need 4 TMUs to reach 800MTexels/bilinear-400MTexels/trilinear.
I assumed that SGX uses the "fast trilinear" technique used on Kyro. Kyro IIRC ran up to twice as fast using fast trilinear versus normal (proper?) trilinear.
Blitzkrieg
Newcomer
xxx, have they actually ever managed to get a high end core made by any customer?
Gubbi, couldnt you always just specifically develop for mobile?
Gubbi, couldnt you always just specifically develop for mobile?
No idea. Actually Kyro/Kyro2 would have been high-end if a smaller process was used. The way it was, it was "upper midrange", but the design was definitely capable.
The problem is, no company is ready to invest such huge money/resources without a 100% certainty that they'll blow nV/ATI out of the water on release. And that's of course not that easy.
KYRO2 was released in early 2001 roughly around the time the NV20 hit shelves. Not only did the K2 lack in features for a complete high end competitor, but I never saw it as anything else than an excellent GF2MX competitor. The fact that it managed to embarass the GTS in certain cases, does not mean in my book that it was competing with anything more than a GF2 MX either. As I said times and times again ST Micro should have released the original STG5000/KYRO3 @180nm and 166-175MHz. There you would have had twice the pipelines, twice the fillrate, double the memory bandwidth and a strong T&L unit with the same clockspeed as the K2. Do you really think that a hypothetical K2 clocked at say 250-275MHz would had been able to beat that? Especially since I'm willing to bet that STG5xxx was capable of Multisampling and times faster higher sample AF.
The real answer to the question above, was the Dreamcast chip. For it's time it definitely was a high end offering, albeit a console design.
[
Well thus I said "would have been if" 
Theoretically, ST could have decided they want to seriously compete, transfer the design to the top-notch process of the day, double the pipes and voila, there you'd have had your high-end chip. I don't know how feasible that would have been, but what I'm implying is that the basic design as such was very well high-end capable. I think
Where's Simon when you need him?
I'd like to know if it was theoretically possible to double the design up with minor effort, provided there were customers interested.
Theoretically, ST could have decided they want to seriously compete, transfer the design to the top-notch process of the day, double the pipes and voila, there you'd have had your high-end chip. I don't know how feasible that would have been, but what I'm implying is that the basic design as such was very well high-end capable. I think
Where's Simon when you need him?
I'd like to know if it was theoretically possible to double the design up with minor effort, provided there were customers interested.
Well thus I said "would have been if"
Theoretically, ST could have decided they want to seriously compete, transfer the design to the top-notch process of the day, double the pipes and voila, there you'd have had your high-end chip. I don't know how feasible that would have been, but what I'm implying is that the basic design as such was very well high-end capable. I think
Where's Simon when you need him?
I'd like to know if it was theoretically possible to double the design up with minor effort, provided there were customers interested.
Vaporware STG5500/K3 on 150nm @250MHz never made it to shelves, since ST Micro put the graphics division up for sale just about when it had it's tape out.
There were fundamental differences between Series3 and Series4, with the latter not only being a twice the K2-pipeline something and that's never a "minor effort".
PowerVR Series 1 PCX could be considered high end, although it needed a good CPU - unlike Voodoo, it had no triangle setup engine.
Most GPU profits come from low to mid range cards. AFAICS the main purpose of a high end card is to improve brand image, helping to sell more mid to low end cards.
From memory, Kyro 1/2 did not have:Theoretically, ST could have decided they want to seriously compete, transfer the design to the top-notch process of the day, double the pipes and voila, there you'd have had your high-end chip. I don't know how feasible that would have been, but what I'm implying is that the basic design as such was very well high-end capable. I think
- cube mapping
- fixed function t&l
- usably fast aniso
- DDR support - not needed at Kyro 1/2 speeds
Most GPU profits come from low to mid range cards. AFAICS the main purpose of a high end card is to improve brand image, helping to sell more mid to low end cards.
:lips sealed:Where's Simon when you need him?
I recall it being announced (at least insofar as online previews are concerned) in early 2001 around March or so after NV20's February PR blitz, although I believe general availability (for Kyro II) was a couple of months later -- I was only able to pickup a Hercules 3D Prophet 4500 in August 2001 through a UK e-tailer. Of course, who could forget the gentlemanly admiration bestowed upon the little TBDR that could from those Saints of Understatement at NV -- spread the love, man!
loekf
Regular
Sigh....
SQRT(8) = ~2,8, assume 60 um / pad => 46 on each side, so that makes 4 times 46 = 184.
But of course 8 mm2 is just the GPU or 3D core.
Add a memory controller, bus interface etc. However, for embedded stuff if you ask me (I could know...) 8 mm2 in 90 nm is quite costly still.
BTW, is there some kind of sensible limit to what you actually need in a portable/embedded application in terms of 3D ? Take a phone with its QCIF+ or QVGA display. At some point enough is enough I would say.
loekf
Regular
Check. NEC and ST did something with Kyro, resulting in real products. At one point in time ST stepped out.
It looks to me MBX is/was quite successfull. However, if you want to integrate it in a true SoC ehh... it's not that nice.
I wonder though about the impact of ARM buying Falanx, leaving IMG on its own.
Considering MBX PRO was never realized due to it's large size, I'm not so sure the SGX530 won't have a similar fate; unless some partner wants to use it for something more demanding than a PDA/mobile device.
NEC had nothing to do with KYRO.
It's my understanding that IMG won more licenses on it's own for MBX, than through ARM. I'm not aware of the details of the ARM/IMG deal for MBX, but to have a middleman for licensing within a technology partner scheme might gain you more "customers" in the end overall but it also sounds to me as less revenue.
NEC had nothing to do with KYRO.
It's my understanding that IMG won more licenses on it's own for MBX, than through ARM. I'm not aware of the details of the ARM/IMG deal for MBX, but to have a middleman for licensing within a technology partner scheme might gain you more "customers" in the end overall but it also sounds to me as less revenue.
To help me understand this quote could you tell we what a "true" SoC is and how it would differ from the OMAP2s, PNX4008, SHMobile3s, etc. "SoCs"
Perhaps a "true soc[k]" is something you put on your foot which would indeed make it much harder to integrate.To help me understand this quote could you tell we what a "true" SoC is and how it would differ from the OMAP2s, PNX4008, SHMobile3s, etc. "SoCs"
