AMD: Volcanic Islands R1100/1200 (8***/9*** series) Speculation/ Rumour Thread

[UniversalTruth]

Yeah, and while waiting a quarter for pricing to become normal, Nvidia will be ready with their new generation...

There are many reasons for AMD to listen, and not to learn the hard way.
The process is already quite old, yields are high, AMD itself with this performance are quite late. They simply have to do it in this way. Otherwise they still leave the door open for people to think what to choose...

 

[rapso]

The actual memory controller should shrink, the PHYs (that's the part usually marked as memory interface on die shots) not. The PHYs are responsible to create the actual signals on the external pins. They need to be able to drive much more current at very high frequencies than the internal on die connections. That's why they need to be relatively large (and tend to grow with higher frequencies). But the PHYs are not en-/decoding the ECC information. As said, they don't really "know" what data or commands they are driving to the pins or are receiving, they just "translate" the signals from the memory controller or the memory between the lower clock speed memory controller and the high speed external interface.
That's what one usually strives for and why the address space is interleaved between the memory channels.
As said, it's interleaved between channels anyway.

thanks for clarifying that, I guess I was talking about this part all the time, using wrong terminology.

The interleaving granularity is probably much smaller, could be as small as a cache line size (64 byte) or a a very small integer multiple of it.

I would have thought the reason to have separate controller was to split the work load. having tiny granularity sounds close to have just one unified controller. but from that perspective, it would indeed balance. yet I'm still not sure about efficiency.

(isn't the cache line size on GPUs 128byte? )


I wish there was some ATI guy here to tell what ...

 

[rapso]

Perhaps the underlying elements of Tahiti's memory controllers were reused from Cayman's design for time-to-market reasons, and thus haven't benefited from advancements to efficiency since then.
This is just my speculation though.

your speculation sounds like a more satisfying idea, would also make sense, as they decoupled the controllers from the rest with the new crossbars. Still wondering what magic they made to scale 512bit-PHYs into 384bit-PHYs die area. at some point they'll want to have faster DRAMs without a complete redesign of the PHYs/controllers.

 

[OpenGL guy]

The interleaving granularity is probably much smaller, could be as small as a cache line size (64 byte) or a a very small integer multiple of it.

Channel stride is 256 bytes. This is documented in the Stream SDK Programmer's Guide.

 

[Gipsel]

(isn't the cache line size on GPUs 128byte? )
I wish there was some ATI guy here to tell what ...

For AMD GPUs it's definitely 64 bytes. It was and is stated several times in architectural presentations and manuals.

 

[Wynix]

Yeah, and while waiting a quarter for pricing to become normal, Nvidia will be ready with their new generation...

You go ahead and wait for the 800 series, if it comes in Q1 it will be on 28nm, 20nm will not be available in the quantities required.

If you don't want to pay the price it's worth, i suggest you look at overclocking a Hawaii Pro.

 

[TKK]

Yeah, and while waiting a quarter for pricing to become normal, Nvidia will be ready with their new generation...

There are many reasons for AMD to listen, and not to learn the hard way.
The process is already quite old, yields are high, AMD itself with this performance are quite late. They simply have to do it in this way. Otherwise they still leave the door open for people to think what to choose...

A lower launch price by AMD would probably just result in accordingly matching price cuts on Nvidia's parts, resulting in lower margins for both. The only winner would be the consumers in that case, but there is no real reason for AMD to wage an aggressive price war.

 

[Davros]

AThe only winner would be the consumers in that case, b

You make that sound like a negative

 

[Bouncing Zabaglione Bros.]

The last few releases we've seen prices go up and stay up for the life of the product. Only when clearing inventory just before being superseded does a card seem to drop in price.

If anything, it seems you're more likely to get a better price with launch bargains and price cuts that are there to encourage pre-orders.

 

[DuckThor Evil]

The last few releases we've seen prices go up and stay up for the life of the product. Only when clearing inventory just before being superseded does a card seem to drop in price.

If anything, it seems you're more likely to get a better price with launch bargains and price cuts that are there to encourage pre-orders.

That doesn't apply in any way to the AMD 7000-series, where the exact opposite happened. Tahiti for example launched at 449/549$, the price soon dropped after the GTX 680 came out and later dropped even more and they added the game bundle as well. The value got a lot better along the way. It's true that inventory clearing drops the prices even further and as we speak there are some pretty amazing Tahiti deals starting to pop up.

edit: Or maybe it applied in the way that prices went up

 

[UniversalTruth]

Do you think two 7970s for 280$ each would perform better than a single R9 290X which can be more expensive than 560$ also?

 

[mczak]

Do you think two 7970s for 280$ each would perform better than a single R9 290X which can be more expensive than 560$ also?

Sure. If AMD fixes the frame pacing problems with CF/Eyefinity, otherwise Hawaii is going to provide a far better experience (but I guess chances are good they will be fixed soon).
They'll also use nearly twice the power...

 

[entity279]

Oh noes.. no product launched an I can already predict pages of pricing speculation and comments.

 

[Esrever]

Im expecting hawaii to have serious supply contraints if they launch at less than $600 with that kind of performance.

 

[Alexko]

Im expecting hawaii to have serious supply contraints if they launch at less than $600 with that kind of performance.

Whatever the price (provided it's reasonable) I don't see why there should be any problems manufacturing enough chips on the now very mature 28nm process.

The RAM used isn't particularly fast, and the PCB should be complex, but nothing unheard of.

 

[Esrever]

Whatever the price (provided it's reasonable) I don't see why there should be any problems manufacturing enough chips on the now very mature 28nm process.

The RAM used isn't particularly fast, and the PCB should be complex, but nothing unheard of.

AMD probably hasn't been producing the chips for a while, its still a pretty large chip. I would also assume a lot of the 28nm process AMD is using is going towards the console APUs and the lower range chips.

 

[mczak]

I doubt there'd be that much of a supply problem.
There's just not too many people buying a graphic card for 600$ no matter how good it is.
Plus if it's really that good they can use up all the 28nm allocation nvidia needed for gk110 previously .

 

[LittleJ]

I doubt there'd be that much of a supply problem.
There's just not too many people buying a graphic card for 600$ no matter how good it is.
Plus if it's really that good they can use up all the 28nm allocation nvidia needed for gk110 previously .

Well we will have to see just how fast Hawaii is before jumping to any conclusions. No doubt NV has the GK110 uber edition that can eclipse the 290x by 5-10% and for them to still claim top dog single gpu. They just wont be able to command rediculous prices for GK110 skus any more. So basically AMD is doing everyone a favor. If you like their cards you are going to be happy with the titan beating 290x at 599-649.99$. If you like NV you might be able to grab a gtx780 for 499-549.99$ or Titan Ultra for 699.99 or thereabouts....

 

[Alexko]

Well we will have to see just how fast Hawaii is before jumping to any conclusions. No doubt NV has the GK110 uber edition that can eclipse the 290x by 5-10% and for them to still claim top dog single gpu. They just wont be able to command rediculous prices for GK110 skus any more. So basically AMD is doing everyone a favor. If you like their cards you are going to be happy with the titan beating 290x at 599-649.99$. If you like NV you might be able to grab a gtx780 for 499-549.99$ or Titan Ultra for 699.99 or thereabouts....

But then AMD could come out with a faster 295X, i.e. Hawaii at ~10% higher clock speed. You know, like in the good old days of the GF 79(5)0s and Radeons X19(5)0s with the successive GT, GTX, GTX, XT, XTX, XXX, GX2, Ultra editions…

I mean, if you think about it, the only reason this stopped is because NVIDIA really opened a gap with G80, and the two companies have never had approximately equally fast top GPUs since then. But it would appear that Hawaii and GK110 more or less fit that bill. Could be fun.

 

[Esrever]

I doubt the performance crown is what AMD is after. Its nice to have it but they are trying to hit better price points than that $1000 crown.