PowerVR Series5 still alive and it's "high end"
- Thread starter rake
- Start date
asicnewbie said:
A Power VR chip wouldn't give any help to Intels bottom line. The reason they might get involved goes beyond immediate revenue. Some of Intels biggest rivals are meeting at one place, the Playstation 3. This location has to be starting to look like a "den of thieves" from Intels perspective. How unsavory it must be to see the PS3 not only use non-Intel architechtures, but also be using new cutting edge architechtures (think Cell) that could cross-pollinate into other markets. Does the PS 3 winning the next console generation hurt Intel? In my opinion it has the potential to cause some problems because new architechtures gain momentum and the companies involved make money allowing them to continue to improve them.
It's Microsoft doing the real dirty work. Intel would just be making swords and shields to let the eager Microsofties go fight the barbarian hordes that are always trying to threaten the Wintel empire.
Sage said:
Actually, in the situation ImgTech was, it would seem to make excellent sense to be agressive in terms of targeted process technology. As has already been pointed out, quite apart from all the normal causes of delay, their business model can add significantly to that if they do not have licensees already lined up.
And if so, it would make sense to design for a process technology sufficiently into the future that there is a window between when you have a saleable design and when it should hit the market, a window that is wide enough for finding liscensees, striking the necessary deal, and time for setting up all the necessities for producing product. Given the signals coming from for instance TSMC a year and a half ago, targeting 0.09um could well have seemed like a good idea if you wanted to aviod getting to market with trailing edge process tech.
Since DX9 will be with us for quite some time, I still think this approach would make one hell of a lot of sense in their shoes. It would ensure that their design has a longer market window, and is also certain to hit the market being as competitive as possible. The only drawback would be that the part couldn't be introduced any earlier than the process allows. But this would only be a real problem if there had been a generational feature shift during the delay, which wouldn't seem very likely with DX9.
Entropy
Richthofen said:
Unfortunately, the "market-share" numbers you are referring to are an aggregate overview of the entire graphics-chip market, which includes a disproportionate amount of both 2D-only and low-emphasis 3D chips. These numbers do not break the market down into observable segments, such as the 3D-only market segment, in which chips like R3xx and nv3xx -based products are the dominant purchases. Since nVidia has effectually shipped nothing competitive in this segment since the GF4 last August (when R300 shipped), it is extremely likely that in this specific segment ATi has enjoyed an almost total domination. Of course, since it has been effectively non-competitive in this segment of the graphics chip market for awhile, nVidia has slashed prices on its lower end mainly 2D chips in order to accelerate sales into market segments in which it could compete, and IMO, this is why the overall aggregate numbers appear to show ATi making only small gains over their previous position. There's simply no way they've made anything but remarkable headway in particular market segments relative to R3xx purchases.
Hellbinder[CE said:
The return of modifier volumes? Less general than stencil buffers, but faster.
Can anyone tell me whether Doom 3's shadows could be done with modifier volumes? Bearing in mind that PVR performance excels with stencil buffers anyway, just how fast could Doom 3 then run on PVRS5?
And what would massive D3 performance do for S5's sales? Could also be PVR texture compression used in MBX (IIRC not the same as VQ used in S2), but I assume TBR would not be required. May also be very high (internal) colour accuracy, but possibly not much point. 64bit "internal true colour" could still be of use in some legacy games.
One thing that needs to be seriously improved from Kyro is anisotropic filtering, as it halved frame rates with not a lot of visible improvement. Obviously PVR will have been working on this, but IMHO this is a fairly big unknown wrt S5 performance and image quality.
(WARNING: personal speculations to follow)
a) I don't see any reason why there should be a separate geometry chip.
b) That ancient old parameter bandwidth claim can be effectively treated.
c) At this stage and if we're refering to a high end PC graphics sollution, nothing less than PS/VS3.0 makes sense.
No idea. Simple math:
KYROII = 175MHz * 32 Z/stencil units = 5.6GPixels/sec stencil fillrate (not being incurred by pixel fillrates).
Now speculate on a possible clockspeed, and a natural increase in on chip Z/stencil units and do the math yourself. The max available stencil fillrate for a NV35 (and correct me if I'm wrong somebody should be 3.6GPixels/sec); and before it leads to any misunderstandings, I'm just trying to verify that TBDR's have an advantage with stencil ops. How and to which degree it can affect final performance in a game like Doom3 (always future products, it's senseless to take the K2 under consideration for D3 IMO), is unknown.
If I did read Kristof's replies in his interview correct about PVR TC, it should be present in future products apart from MBX. I'm still not in the clear about it's specifics though (albeit discussed extensively here in these forums).
*ahem*
http://www.pvrgenerations.co.uk/cgi.../2002/kristof1102&printer=0&pagenum=1
Finally I've thought of financials/possible partnerships/potential lisencees myself quite often; it might not be an answer but if there isn't at least one reliable interested party lurking in the background, then the risk ImgTec is taking with possibly fabbing the prototypes themselves, the involved costs and resources absorbed, is more than just high.
***edit:
I don't see why 16x sample MSAA should be impossible as an example, yet if we're talking just about ordered grid sampling patterns then errrmmm.... *yawn*
a) I don't see any reason why there should be a separate geometry chip.
b) That ancient old parameter bandwidth claim can be effectively treated.
c) At this stage and if we're refering to a high end PC graphics sollution, nothing less than PS/VS3.0 makes sense.
No idea. Simple math:
KYROII = 175MHz * 32 Z/stencil units = 5.6GPixels/sec stencil fillrate (not being incurred by pixel fillrates).
Now speculate on a possible clockspeed, and a natural increase in on chip Z/stencil units and do the math yourself. The max available stencil fillrate for a NV35 (and correct me if I'm wrong somebody should be 3.6GPixels/sec); and before it leads to any misunderstandings, I'm just trying to verify that TBDR's have an advantage with stencil ops. How and to which degree it can affect final performance in a game like Doom3 (always future products, it's senseless to take the K2 under consideration for D3 IMO), is unknown.
If I did read Kristof's replies in his interview correct about PVR TC, it should be present in future products apart from MBX. I'm still not in the clear about it's specifics though (albeit discussed extensively here in these forums).
*ahem*
http://www.pvrgenerations.co.uk/cgi.../2002/kristof1102&printer=0&pagenum=1
Finally I've thought of financials/possible partnerships/potential lisencees myself quite often; it might not be an answer but if there isn't at least one reliable interested party lurking in the background, then the risk ImgTec is taking with possibly fabbing the prototypes themselves, the involved costs and resources absorbed, is more than just high.
***edit:
I don't see why 16x sample MSAA should be impossible as an example, yet if we're talking just about ordered grid sampling patterns then errrmmm.... *yawn*
Similar threads
- Locked
