Gamespy leaks Xenon specs (?)
- Thread starter Titanio
- Start date
PC-Engine said:
Then I feel sorry for you...
london-boy said:
R&D is not cheap...
It isn't, but they already paid for that. It's a fixed cost, which will be paid off with profits made from selling the product later on in its life.
How much a drive costs to manufacture, in practice, is what my question was. Can't be 4x more expensive than constructing a DVD drive, they work the same afterall, only the BDROM one needs a more expensive lens and probably more precise components.
london-boy said:PC-Engine said:
Then I feel sorry for you...
london-boy said:
R&D is not cheap...
It isn't, but they already paid for that. It's a fixed cost, which will be paid off with profits made from selling the product later on in its life.
How much a drive costs to manufacture, in practice, is what my question was. Can't be 4x more expensive than constructing a DVD drive, they work the same afterall, only the BDROM one needs a more expensive lens and probably more precise components.
If that's what you're asking then you have at least two factors that affect manufacturing cost.
1. Difficulty to manufacture which arises from complexity and/or immature production process.
2. cost of actual components and/or raw materials.
PC-Engine said:
I know, i'm just not too sure that at a manufacturing stage (before all the premiums for being new and shiny and Sony and all that), that will come up to 4X or more the cost of making a DVD drive that's all.
PC-Engine said:Then I feel sorry for you...
Don't worry, it's not my only means of DVD playback.
But I DO still use it.
London-boy is right though, it's an expense that they've already put down for, and have been putting down for for years.
As far as immature manufacturing processes, that was actually one of my prior points; that blu-ray inclusion in PS3 will allow Sony to become the manufacturing leader in terms of process as far as blu-ray goes early on in the media's lifecycle, simply by the volume they will be moving through PS3.
Manufacturing in volume doesn't equal production maturity. Yields determine maturity. You could manufacture a million units per month but if only half of those pass QC then it means you're losing a lot of money because your yields are only 50%...works the same way in semiconductors.
PC-Engine said:
Obviously, but that's just one case scenario. Yields could be more or could be less.
To be honest i don't think a bloody BDROM drive is such a high precision device that Sony will have problems with it for long. It's not a bio-mechanical-nano-whatever... thing.
PC-Engine said:
I agree with you, but just as with semiconductors, the sooner the manufacturer begins producing in volume, the sooner they work out those 'kinks' in process yield. Plus, I agree with london-boy that for a manufacturer the size of Sony, who has been making optical drives forever, I'm sure the 'learning curve' as far as process will be a very shallow one.
Question is when do they start? Do they already have production issues resolved? Have they gotten the price down to within twice of what a 12X DVD drive cost? What is the cost of the blue laser diodes? Is the supply of the laser diodes high enough? How reliable are the drives?
PC-Engine said:
Well i guess we'll have to wait and see. Besides, they have about 12 months more or less to get it right, and that's a long time.
Megadrive1988
Veteran
aaaaa00, DaveBaumann and DeanoC: ok you guys tried to clear up the whole thing about next generation console (and PC) GPUs not having X amount of pipelines with Y amount of TMUs attached. therefore, Xenon GPU is not going to be 16:2 or 24:2 or 32:1 or 48:1 etc. It has a certain amount of ALUs or processing resources. things are totally different than the traditional 4:1, 4:2 pipeline era that most of us remember with Nvidia. and the 8:1, 16:1 era we remember with ATI
I am assuming that the PC R520 Fudo (not Xenon GPU) still has traditional pipelines. - most likely 8:24:1
(8 vertex shaders, 24 pixel pipelines with 1 TMU each)
that would give R520 Fudo 32 total pipelines compared to 22 in R420 Loki
(which had 6 vertex, 16 pixel).
some are even thinking that R520 might even have 32 pixel pipelines and 8 or even 12 vertex shaders. okay, 12 vertex shaders and 32 pixel pipelines is not likely but, *possible*, if ATI doubled everything for Fudo. so assuming Fudo still has pipelines we are looking at anywhere from 32 total (8v, 24p) to 44 total (12v, 32p) pipelines in R520 Fudo. I'm thinking either 8:24 as most expect, or maybe 8:32.
but even that maximim concievable amount for R520 Fudo is ( i am assuming ) less than Xenon R500's reported 48 shader ALUs--according to the leaked document-- which can be used for either vertex or pixel processing--also according to the leaked Xenon document.
I realize the actual Xenon GPU spec could well be different from that leaked document. maybe more than 48 ALUs. heh, but I don't expect any answers from you guys regarding the amount of processing resources inside Xenon GPU.
anyway, I am trying to get it through my head that Xenon GPU is no longer about pipelines, or even fillrate, but how many shader ops can be processed per clock. am I headed in the right direction regarding Xenon GPU (and the probably not-unified shader R520 Fudo) ?
I am assuming that the PC R520 Fudo (not Xenon GPU) still has traditional pipelines. - most likely 8:24:1
(8 vertex shaders, 24 pixel pipelines with 1 TMU each)
that would give R520 Fudo 32 total pipelines compared to 22 in R420 Loki
(which had 6 vertex, 16 pixel).
some are even thinking that R520 might even have 32 pixel pipelines and 8 or even 12 vertex shaders. okay, 12 vertex shaders and 32 pixel pipelines is not likely but, *possible*, if ATI doubled everything for Fudo. so assuming Fudo still has pipelines we are looking at anywhere from 32 total (8v, 24p) to 44 total (12v, 32p) pipelines in R520 Fudo. I'm thinking either 8:24 as most expect, or maybe 8:32.
but even that maximim concievable amount for R520 Fudo is ( i am assuming ) less than Xenon R500's reported 48 shader ALUs--according to the leaked document-- which can be used for either vertex or pixel processing--also according to the leaked Xenon document.
I realize the actual Xenon GPU spec could well be different from that leaked document. maybe more than 48 ALUs. heh, but I don't expect any answers from you guys regarding the amount of processing resources inside Xenon GPU.
anyway, I am trying to get it through my head that Xenon GPU is no longer about pipelines, or even fillrate, but how many shader ops can be processed per clock. am I headed in the right direction regarding Xenon GPU (and the probably not-unified shader R520 Fudo) ?
Megadrive1988 said:
Shader ops per clock is IMHO the most interesting stat going forward. Of course what exactly is a shader op is itself a hard thing to pin down.
Especially with long shader architecture (i.e SM 2a and higher) a relatively low fillrate but high shader op architecture could be very pretty.
But fillrate is still important for things like particles and the like...
Fillrate gives much bigger numbers than ops/clock (gigapixels vs. say 64-256 ops/clock) so fillrate is vastly more important.
Similar threads
