R520 to have 300 Million Transistors??

Really hate quoting the Inq .. Anyways..

------------------------

ATI'S FUTURE R520 graphics chip could have as many as 300 million transistors on it, it has emerged.
We don’t have the exact number yet, but in a conversation with a source, we were reminded that with every single chip generation ATI ended up with twice as many transistors as the previous chip.

If we take into account that the R420 chip codenamed Loki and commercially known as X800 series ended up with number close to 150 millions of transistors, the 300 million figure may be close to the mark.

ATI decided not to go for Shader Model 2.0b with its new generation and saved some 60 million transistors since otherwise it would have 225 million, the number that Nvidia ended up. This also made the ATI card less advanced but easier to produce.

We are prettu sure that Fudo, R520 will end up with Shader Model 3.0+ not 4.0, as this one will be out with Longhorn late 2006. If you think about it we will be talking about R620 at that time and I am sure that this one will have Shader Model 4.0 support.

We all know that this chip will use 110 nanometre marchitecture as ATI never likes to take the chances with a non proven process and high end chip while we will see some of chips built with 90 nanometres in 2005.

Imagine what can you put into 300 million transistors, many new pipelines and much more memory as well as some wicked shaders.

R520 comes in the first half of 2005 anyway so I would expect to see it sometime in May if not earlier, just a year after the R420.
 
Every single chip generation ... except R300->R420.

*cough*

So has everyone settled on 60M as the transistor savings of SM2.0b over SM3.0, or has Jen-Hsun's conference call 60M figure somehow become official (even though it doesn't seem to take into account the media processor)?
 
R520 has prolly not tapped out yet and should not be for 3 month if May release is expected.. It is way too soon for reliable informations i think.
 
AlphaWolf said:
Fuad "here's some stuff I made up and posted like it was news" Abazovic strikes again.

Actually, in this instance he was pointed to the idea that ATI have doubled couts over revisions (of course, ignorning R300->R420!), although I spoke with the guys that pointed that out to him today and I think they were having a bit of a larf to see whether or not it would end up at L'Inq! ;)
 
I'd love to have a competition - see who can create the most mysterious, outrageous, factually relevant bit of 'insight' that can be stolen and printed up by the Inq as a revelationary new finding from their inside sources :)
 
g__day said:
I'd love to have a competition - see who can create the most mysterious, outrageous, factually relevant bit of 'insight' that can be stolen and printed up by the Inq as a revelationary new finding from their inside sources :)

r520 will only do fp 32. It wont support fp24 or fp16.

It will also have a 512bit bus
 
FP32? Pffft. Try FP48 throughout. ATi doubles its shizzle every generation, baybee!
 
jvd said:
nope ram will only be 700mhz . 512 megs on it. With perhaps 756 on a refresh

Uh, I guess one of the engineers working on the chip came into the video games store in Jersey where you work on his lunch break and told you all this. :LOL:

Besides, I'd like to know how you plan on reaching 768MB with an even number of memory devices... I mean, 48MB memory chips being fairly uncommon and all. :LOL:
 
madshi said:
The Baron said:
Yes, of course they'll use a .11u process. Since it has low-k and all.
Since when does .11u have low-k? TSMC doesn't offer low-k for .11u, or does it?

It was my understanding too that 110nm is purely aimed for mainstream parts and for performance IHVs have the choice between low-k 130 and 90nm at TSMC.

I've of course no idea if something has changed in the meantime.

Guden,

I think the boys were merely joking. Who in his right mind can take the INQ seriously anyway?
 
Back
Top