I am sure Dave B will disagree with me on this one...

Your thoughts are similar to a lot of the speculation being bandied about on the boards. Although you did manage to make it sound more ominous!!

Just to clarify - you do have more robust sources for this info than the usual public gossip right?
 
Not just is something rotten in Denmark, the aroma is coming from all points of the compass. Some serious lying going on somewhere.
 
FWIW, I remember reading an interview with someone at ATI approximately a month back and they were talking (well, mentioned) actual R520 board production. The implication was that they were busy designing the actual PCB and the engineer was having a little laugh about it (not sure why, but it came off sounding like "yeah, we have it...we
are just tooling around with it *heh heh heh*").

Naturally, I am too lazy to dig up this interview.
 
Well, I hope I didn't sound too doom and gloom about it. It isn't the end of the line for ATI at all, but it could be a pretty significant speed-bump for them (assuming my information is correct). For all the talk the past few years about ATI releasing their smaller parts on a new process, then migrating the more complex parts to that new process when it is producing well, seems to have gone out the window with the R520.

While the high end is nice for ATI, they are getting killed in the sub $200 range by NVIDIA. While the HyperMemory boards are starting to make an impact, the 6200 and 6600 series have seemed to have taken a large amount of sales. It makes you wonder what the RV5x0 series are doing, and if they will suffer some of the same problems as the R520, or if they have been delayed because of the issues. I'm sure folks in Canada know, as well as a few of their partners, but we certainly don't know for sure.
 
Josh, are the bits of info re the number of pixel pipes and the yield problems from the same source?
 
This doesn’t reconcile with the information leaked about the R500. It seems ATI is more than capable (if the rumors are correct) of producing a power efficient GPU on this process node. I know that it is from a different design team as that of the R520 but as far as placing the blame on the use of standard cells I would think that these would not be proprietary and ATI would use the same on both.
 
Nelg: Well the R500 core chip is 150-160M, it really is a multichip design imo (with the NEC EDRAM+ROPs chip). That's quite different from a single 350M chip, which has about 4-5x the complexity. I agree the blame cannot be put on "standard" cells though.
MuFu said:
Josh, are the bits of info re the number of pixel pipes and the yield problems from the same source?
And in addition to this, does the "only 16 pipelines out of 32 working on most chips" come from the same source, or a different one?
Interesting article anyhow though, thanks!

Uttar
 
Uttar said:
Nelg: Well the R500 core chip is 150-160M, it really is a multichip design imo (with the NEC EDRAM+ROPs chip). That's quite different from a single 350M chip, which has about 4-5x the complexity. I agree the blame cannot be put on "standard" cells though.
R500 core is 232 MTransistors.
 
Half the time people are saying 32, half the time 16 and sometimes 24, is there any concensus on which it really is yet ?

Or is it down to semantics with the card being able to process 32 pixels at the same time and this being read as describing the physical setup of the chip, ie 32x1 ?
 
Someone brought up a good point in another forum. As we (up til now) deal with quads, why are we assuming a 32-pipe board will go down to 24 or 16 pipes? Why not 28 or 24? Surely ATI isn't trying to avoid confusing the customer. :p
 
Back
Top