Hey, nVidia still has like 8-9 hours left if you go by their HQ's time in Californy..... 
AMD: R8xx Speculation
- Thread starter Shtal
- Start date
Bouncing Zabaglione Bros.
Legend
Aha! I voted next year. Too much BS and lack of consistent information from Nvidia if they were going to make it out by the end of the year. The claimed schedules just didn't match up with what we were hearing (or not hearing).
Now they are saying March, so either they are lying to us or they have no idea what is going on and when they can fix the chip.
Now they are saying March, so either they are lying to us or they have no idea what is going on and when they can fix the chip.
Ah yeah, the mixed signals. Alibrandi said something about the lines of "waiting for drivers" because they want GPU and GPGPU both perfect, somehow suggesting that GF/Qdro/Tsla all need better drivers then they were working on for the past year.
The PCPer article from the 30th has nvidia saying that there is no relation between the GF and Tesla cards, seems to me that they could launch one or the other since not everyone has to wait for the same drivers? It must be a PR nightmare having one side of the company say they're completely different products and the other side saying they're not.
Talking to someone else (who's actually tasked with getting cards to customers) he said it's still a matter of yields and actually reaching "satisfactory" clocks. That doesn't bode well to better availability than Cypress to me.
Last edited by a moderator:
Bouncing Zabaglione Bros.
Legend
Fermi would be worse though. A more complex design, with more process issues, that is larger (thus more likely to get hit by die defects), that has already had lots of problems. It's got to be at least as fast and not much more expensive than 58xx/59xx. It's harder and more expensive to build, and if clocks are poor, it's not going to be fast enough to warrant any kind of price premium. Sure, Nvidia will BS about DP and Physx, but those are not the fundamentals in the gaming market.
And then you have to ask what Nvidia are going to do for all segments other than the high end.
Groo The Wanderer
Regular
I can't believe they named a part after my daughter like that, I'm just sincerely touched!
Remember what became of the chip they named after Fudo? Don't be so confident, even if the naming was well intended.
-Charlie
Groo The Wanderer
Regular
Ah, finally, some consistent info with multiple trustworthy sources to back it up. If you recall, Theo said there was A1 (although I seem to recall it saying A2 in the past) silicon in Santa Clara in May, and that they couldn't begin work on the drivers until they had silicon in house (article seems to have gone poof there).
Step forward, and you can see Alibrandi saying that work on the drivers is still ongoing, so that is the reason for the delay. There, simple logic.
Remember, they never claimed a year for their black friday launch.
-Charlie
http://img.photobucket.com/albums/v455/lixianglover/001345118.jpg
redwood die.
Deadend source: http://www.xtremesystems.org/forums/showthread.php?t=242333
EDIT: Looks like the *5* subseries is now much more buffed up, akin to 4670, and the 6 subseries should have GDDR5.
redwood die.
Deadend source: http://www.xtremesystems.org/forums/showthread.php?t=242333
EDIT: Looks like the *5* subseries is now much more buffed up, akin to 4670, and the 6 subseries should have GDDR5.
Last edited by a moderator:
Jawed
Legend
http://sa09.idav.ucdavis.edu/docs/SA09_AMD_IHV.pdf
Despite the terrible typeface used on the diagrams throughout this presentation, the information in there is great.
Jawed
Despite the terrible typeface used on the diagrams throughout this presentation, the information in there is great.
Jawed
The problem is, it was supposed to come as 32nm product, and as we know, TSMC (and GF, for that matter) has cancelled 32nm bulk process. Only "real option" is either 40nm (too hot?) or 28nm (too late?), or if Chartereds (now part of GF, right?) 32nm process is bulk and suitable for GPUs?
32nm for the GPU in 2010? When was this roadmap made? Also whose 32nm process would they be talking about? Their own or TSMC?
I would say a better analysis of it would be based on overall performance and for mobile parts its as important to have good performance for every watt 'spent'. Overall its probably better to look at it from the perspective of overall margins. If a mobile chip has better performance characteristics for the platform then its margins would be higher.
I would say the major expense would be from actually redesigning chips designed for a bulk process and making them suitable for SOI. So long as the chip performs better any increased cost for the chip itself relative to the competition can be offset by increased margins.
i believe Nvidia was investigating this a while ago but im not sure how far or what conclusions they drew from the idea.
If it's the same 480 stream processor variant used in Llano the redesign work could be shared for the most part (the GDDR5 interface might be a big deal, but seeing as AMD pioneered it in the first place and they have experience with running HT links at similar speeds on SOI for a long time it might not be a big deal either ... I dunno).
Page 8 gives a lot away and should gie you a fairly good indication why there are two rasterizers.Hm, dual thread dispatchers in Cypress?! The arch diagrams in this presentation are much more detailed.
OMG that pcb is empty and small! Looks dirt cheap (if even redwood board looks so small how does a cedar board look like?)
Anyone figured out approximate die size yet?
Similar threads
