I am sure Dave B will disagree with me on this one...
- Thread starter JoshMST
- Start date
Could just be NV types struggling to understand what the competition is doing on imperfect information. They've created (apparently) a 24-pipe SM3.0 card on 110nm. From their pov, it is perfectly rational to assume ATI could do 32-pipes on 90nm --so if they didn't, if they are beginning to hear that it is a 16-pipe card from some of their "friends in common", then they think "Ahha! Tragic design flaws are causing such yield problems they are only getting 1/2 their pipes to work!"
"Oihhhhh" :?
(The way I go after virtually each of Josh's articles.)
Just think about it from a production pespective. If you'd assume ~300 mm^2 die size for a theoretical 32-pipe, SM3 GPU on a 300mm / 90nm process, you'd literally need several hundred defects per wafer to get the described scenario, and all in the 'right' places, too.
(The way I go after virtually each of Josh's articles.)
Just think about it from a production pespective. If you'd assume ~300 mm^2 die size for a theoretical 32-pipe, SM3 GPU on a 300mm / 90nm process, you'd literally need several hundred defects per wafer to get the described scenario, and all in the 'right' places, too.
I'm thinking more along the lines that these reports are right....just a few months late.
The R520 was originally supposed to launch in April, at least at the beginning of the year all the signs pointed to it....I think they had trouble then with the process and that's why we're not going to see the R520 until later this month.
But I think these problems Josh is writing about are in the past and were the reason for the delay rather than a reason for another one.
The R520 was originally supposed to launch in April, at least at the beginning of the year all the signs pointed to it....I think they had trouble then with the process and that's why we're not going to see the R520 until later this month.
But I think these problems Josh is writing about are in the past and were the reason for the delay rather than a reason for another one.
Well, Dave obviously knows what exactly the R520 is, so his hints can be frustrating.
I sincerely don't doubt that the 16 pipe scenario could in fact be true. If ATI did a lot of internal architectural work, and got the chip to run at 600+ MHz, then I don't see why it couldn't seriously spank the G70. The timing of it all has seriously confused me though, and the info I was given about the design issues (then the Goldman Sachs release) certainly fit into what we are seeing. Whether it is right or wrong... I have no idea. I am just going on information that was given to me, and held up to the light of public information. Not all of us have Dave's contacts, and if we did we wouldn't be able to talk about it anyway.
Man, I got slammed by Faud. How embarrasing is that?
I definitely don't think that the R520 is going to be the next NV30, but I still don't think that it will be an easy chip to produce.
I sincerely don't doubt that the 16 pipe scenario could in fact be true. If ATI did a lot of internal architectural work, and got the chip to run at 600+ MHz, then I don't see why it couldn't seriously spank the G70. The timing of it all has seriously confused me though, and the info I was given about the design issues (then the Goldman Sachs release) certainly fit into what we are seeing. Whether it is right or wrong... I have no idea. I am just going on information that was given to me, and held up to the light of public information. Not all of us have Dave's contacts, and if we did we wouldn't be able to talk about it anyway.
Man, I got slammed by Faud. How embarrasing is that?
I definitely don't think that the R520 is going to be the next NV30, but I still don't think that it will be an easy chip to produce.
About on par with the last time I woke up late and was a little too rushed getting my kids ready for school and ended up noticing I was still in my pajamas right after I had dropped my daughter off.JoshMST said:Man, I got slammed by Faud. How embarrasing is that?
No, I take that back...it's worse.
Bouncing Zabaglione Bros.
Legend
JoshMST said:
R300 wasn't an easy chip to produce. In fact, ATI's competition publicly said it would be impossible to produce with the required performance. What's more important (regardless of how difficult R520 might be to manufacture), will ATI be able to make them in quantity anyway? Delays and respins are designed to do exactly that.
JoshMST said:
Does he? I doubt he'd say that even now. He obviously is relying on some information, but I doubt he's placed it in the "100% verified" column, and wouldn't without some independant lovin' on the test bench.
I can't entirely discount yet the posibility that we've all been played big-time, but we'll know soon enuf.
allnighter
Newcomer
I wouldn't feel bad about it. After all Fudo also brings out the info he's been given. Being it somebody yanking his chain again or him reading these and other message boards and calling them his "usually very reliable sources"- bottom line it's the Inq. So far Fudo has explored all the possible pipeline configurations and probable clock speeds so he can say he knew it all the time.JoshMST said:Man, I got slammed by Faud. How embarrasing is that?
geo said:
Actually I've seen the 32-piper rumour mostly from those evangelizing that R520 is going to spank the G70.
Inq seems to believe that the R520 is doing just fine and Nvidia is trying to embellish whatever negative rumours there may be about ATi's ability to produce. Isn't it as likely that ATi is still working out some kinks and are trying to minimize the negative public opinion?
There is also Inq's theory that ATi is stockpiling a monster G70 killer and is waiting to see G70 before setting clocks - has there ever been a case where a company has needed to reduce performance of a product to match the competition?
Then there's the other theory about ATi holding back R520 and launching Crossfire so they can get rid of X850 inventory. Wouldn't the prices and demand for X850 cards drop just as much when G70 debuts whether or not R520 is out as well? Why not just cut X850 prices and launch Crossfire with R520, since they are apparently sitting on an arsenal of R520's just biding their time? That approach would certainly damage G70 sales a lot more and the Crossfire brand would be just as prominent.
trinibwoy said:
Right, so on the one hand some NV types are hearing in that one ear. . .and in the other ear they are hearing hints that all anyone is actually seeing so far is 16-pipers ("anyone" being partners, ISV's, etc). One could put those two threads together and end up with a design/yield problem theory.
The thing that really killed me about Faud's writeup is that he said my info was from NVIDIA, but then goes onto say at the bottom that ATI is not actually mass producing the R520 with TSMC at this moment, and that they are obviously having issues with 90 nm. Which in turn lends a little credence to my article about ATI having issues with the R520 design on the 90 nm process...
I don't get it.
I don't get it.
My info was not from NVIDIA, and it was not a secret email type "Deepthroat" affair. The person I got the info from is very reliable, and usually the things I hear from them I keep under my hat. This was the exception.
So in short, Brian Burke did not email me and say, "Hey Josh, I heard ATI has some major issues with R520 and it leaks like a sieve."
So in short, Brian Burke did not email me and say, "Hey Josh, I heard ATI has some major issues with R520 and it leaks like a sieve."
Ah, so it was Derek.JoshMST said:
More news from Xbit - http://www.xbitlabs.com/news/video/display/20050609093014.html
So R520 is functional but the only problem is yields. My question is how can ATi be so confident of having the flexibility to choose clockspeeds at a whim once G70 performance is known.
So R520 is functional but the only problem is yields. My question is how can ATi be so confident of having the flexibility to choose clockspeeds at a whim once G70 performance is known.
Similar threads
