RV870 rumor thread !

Burn said:
dual cores .. it mean that a '' X2 '' version could have 4 cores ...
Click to expand...

What benefit would there be to a dual-die design? Taking the RV870 core and merging it with another, to give double the sp's etc., but a single memory controller/shared cache? Removing the bridge communications? Single memory setup? But why wouldn't you just extend the original die in the first place, give it +50% more... more performance increase for less investment and design?

20% boost doesn't sound like a 'new gen' type increase. I guess I'm spoiled going from 3870 to 4850. :lol:
 
noko said:
Wider ram would help too, maybe not 512 bit but something greater then 256 bit wide.
Click to expand...

AlexV said:
Why exactly?
Click to expand...

A wide bus is what nvidia is currently doing to achieve performance, though as far as i understand it, it also makes things more expensive and a bit harder to implement witch is why ati has been using increasingly fast ram. I suppose at some point it might be necessary to make something higher than 256-bit for good performance
 
@ 655Mhz each that would be.... 5TF. :)

r870jq2.jpg


:D

They need to drop PLX chips and make a new one all together. have 80 lanes. (4x16 + 16 for PCI-E)
 
Last edited:
Hatecrime69 said:
A wide bus is what nvidia is currently doing to achieve performance, though as far as i understand it, it also makes things more expensive and a bit harder to implement witch is why ati has been using increasingly fast ram. I suppose at some point it might be necessary to make something higher than 256-bit for good performance
Click to expand...

I don't think so.

Currently 4870's GDDR5 is 3600Mhz and has plenty of bandwith to boot.

GDDR5 is expected to scale to 7Ghz, but even a "pessimistic" 5Ghz next May or so would be enough.

5Ghz 256bit GDDR5 =~ 2.5Ghz 512bit GDDR3 (AFAIK there's nothing really official that is going to be 2.5Ghz at all)

So at 5Ghz you can have equivalent bandwidth at half bus width, saving a lot on the chip costs itself. GDDR5 might be expensive, but heck, graphics RAM for a highend has almost always never been cheap.
 
Burn said:
It is believed that this GPU could perform 1.2 times better than RV770, purely based on the paper-specs.
Click to expand...
I am soooo tempted to make a joke about that 1.2 (which actually means 20% faster), but I'll pass :p
 
Treeckcold57 said:
:lol: Why are you think it will have two dies on one GPU? Oh please. :p
Click to expand...

Why not, R700 was rumored to be 1-4 cores. IF I was ati i would do it that way, why make different chips for each pricepoint, when you can start grouping them together. The other thing could be native dual cores, and 2 of them. Also IF these specs are true, to get 2x the performance over the R700 it will have to be 4 of them, or 2 @ 1300Mhz shader clock. :)
 
SSXeon said:
Why not, R700 was rumored to be 1-4 cores. IF I was ati i would do it that way, why make different chips for each pricepoint, when you can start grouping them together. The other thing could be native dual cores, and 2 of them. Also IF these specs are true, to get 2x the performance over the R700 it will have to be 4 of them, or 2 @ 1300Mhz shader clock. :)
Click to expand...

Ah...speaking of makes few flavors of card. Well said.
 
I still dont understand what people are talking about when referring to multi-core GPU's. They are already massively parallel (800 stream processors... 800 processors that are identical... multi core what?). Multi die is X2 series; I can't see the production costs of dual die in a single package being worth it unless its a radical die shrink for power, cooling and space saving.
 
caveman-jim said:
I can't see the production costs of dual die in a single package being worth it unless its a radical die shrink for power, cooling and space saving.
Click to expand...
A dual die GPU is absolutely ideal. It has the potential of being nearly as fast as a monolithic die containing the same transistor count. But with one huge advantage, yields. Doing a 2 billion tran single die vs. a 2 billion dual die package will reduce costs in a big way.
 
Yep, but this is also why it can't be a native dual core. It will still need to be a multi chip product to have the same benefits with yields.
 
Fuad said:
Comes next year

After the success of its RV770 and refreshed DirectX 10.1 cards ATI wants to continue its winning streak with a DirectX 11 card.

We still don’t know the codename, but let's assume that the next generation performance / mainstream chip is codenamed RV870. This chip will likely have DirectX 11 support; at least this is what our sources believe at this time.

There is an important indication that RV870 might launch before DirectX 11 becomes available but this can only be a good thing for ATI, but if we were betting people we would suggest that such a chip should be available a year from now, in late Q2 or early Q3 2009.
Click to expand...


Link
 
Skynet said:
A dual die GPU is absolutely ideal. It has the potential of being nearly as fast as a monolithic die containing the same transistor count. But with one huge advantage, yields. Doing a 2 billion tran single die vs. a 2 billion dual die package will reduce costs in a big way.
Click to expand...

unless I misunderstand how a dual-die package is created, I dont think this is true.
 
You must log in or register to reply here.
Back
Top