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

^ LOL @ Iceberg.

 

What's the chances of him being able to confirm anything that Dave couldn't already have confirmed? Pretty slim surely.

However no harm in trying for double "confirmed" right?

DAVE WHAT'S YOUR FAVOURITE NUMBER CHOICE, 28 OR 20?

And the great thing is, he can tell us now as it's already been "confirmed" elsewhere.

 

I bet he would say 20 just to mess with us.

 

you certainly right, or give you the age of his children instead. or the birthday of his wife. he dont fall on this type of trap so easely.

 

I think that a better way for AMD to compete would be to compete where Nvidia can't, after years of talk on the matter actually producing "gamer class APU".

 

I think they'd like to do that but no longer have the money for development. We saw a rumour of Kaveri's GDDR5 being cancelled, whether or not that was true is unknown but it seems likely. They are really dependent on the rest of the industry here and it could barely be any slower getting to even DDR4.

They are doing something about it though, GCN and hUMA are surely being developed with lowering bandwidth requirements at the forefront, but that's a long-term solution.

 

Well I do get that they are sadly spread pretty thin at the moment but they have to make the right arbitration.
The sad matter of a fact is AMD doesn't seem able to work simultaneously on server grade CPU, desktop grade CPU (/APUs), and laptop/tablet grade CPU (APUs), in the mean time they are able to quite a couples GPU out a year, I haven't checked, you might know better ( correct if needed ), in ~ a year they got out, Tahiti, Pitcairn, Cap verde, Mars , Bonaire, and a lower end mobile SKU (? not sure and can't remember the name).
They have to make an arbitration, CPU (SoC/APUs) is more vital to their survival than GPU. I guess it could get better when they finally develop all their product on the same process but I think they have to make an arbitration, like now, and looking at their roadmap they have not

 

No matter what you do, there's yet a problem of size, TDP etc for APU.. you cant put a Thaiti or pitcain grade on the same silicon of the CPU.

All the gpu you are citing are all coming from the same architecture familiy, aka GCN. This arch is then declined in different version.

APU is in the low end, middle end market. They will still need dedicated gpu for the high end market, gaming market.. And specially they will still need to developp and do R&D on the GPU front for developp it for their APU ...

Look on global gpu share, they are still largely in front of Nvidia.

 

Well for some reasons unknown to me ( I would guess may be something related to static vs dynamic power use, leakage, etc.) the power consumption of an APU is not the sum of the CPU and a discrete GPU that would match the perfs of the APUs, it is significantly lower.
So clearly they have room in my opinion. They sell for really low price those +300mm^2 4modules FX CPU, which burns an insane amount of power relative to their perfs.
It seems that in a lot of case the CPU perfs prevent games to sustain 60FPS , so they don't need a crazy big GPU to match one of their quad core. Kaveri still fall short, with 8 CUs, they would better go with 14 CUs along with GDDR5 (do salvage part from here).

So speaking of pretty much passing on producing anything in the GPU realm that doesn't involves at least a 192 bit bus with matching overall rendering capability.

Their last financial result were ok, though they are fast at pushing GPU out and too slow at pushing CPU out. I hope that once they only operate on one process, they will increase the rate of release of CPU and lower the number of GPU they release. I do get the CPU need more testing, though they have to acknowledge that not all CPU are server grade.
The CPU are not different from GPU, the have those piledriver cores, they built APUs with 2 modules and the FX line with 8 such cores.

It is to me a better strategy to try to push Nvidia out of a lot of market segment by providing powerful enough APUs vs trying to undercut them by 10 or 20 bucks in the plain discrete GPU realm.
The issue is that if such a move happens, for good, it is likely to create a hole in their revenues for a quarter or two (or more if they messed up) which they may not be in a position to afford.
Though sticking to short term strategy is not good either.

 

Always very important!

 

It is worth noting that there are a number of problems with GDDR5 (capacity, cost) use as main RAM, which make it not necessarily feasible in desktops / laptops, outside of some boutique uses. As we've had ample evidence up to now, AMD doesn't do well as a boutique thing because it appears that they've fallen into a pit of "the cheaper, somewhat worse alternative" brand image (apparently the Tahiti pricing that was supposed to fix that didn't...neither did those misguided 9xxx FXs).

I do believe that they have the technical ability to do it, but the economics are not in their favour (I do secretly hope that someone will roll a Kaveri based laptop with 8 GB of GDDR5). On a separate note, I don't see GCN or the hUMA (sigh, MKT) as being intrinsically meant to lower BW usage (for a wide, high-throughput machine) in any way - quite the opposite in fact.

 

What's in the Playstation 4 then?

 

Would hUMA removing the need to copy data from cpu memory to gpu memory (in an APU) not free up some bandwidth? Or would that just free up cpu cycles?

And yeah, there's no great reason why an APU can't go up to 300W so long as the cooling and power supply supports it. It's pointless under current bandwidth constraints though (and also fairly pointless with AMD's current CPU performance), but in future I can see AMD going with bigger, more powerful APU's once those issues are fixed.

They do still have "good enough" on their side, and APU's are catching up to that point rapidly at the lower resolutions. It's only a matter of time before 1080p APU performance is "good enough" as well - perhaps not for the graphics connoisseur but for everyone else it'll be acceptable.

 

Pitcairn class was a poor choice of an example.
I personally would love to see a 3-4 module + 2048SP 20nm Monster provided they can feed it enough bandwidth.

 

The problem isn't so much one of bandwidth, die size limits or power consumption. AMD can put a Tahiti equivalent with 4 Steamroller cores on a 28nm die. It would be a big die, and it might draw 250~300W (with reasonable clock speeds) but that's something AMD knows how to cool (see HD7990, for instance).

Bandwidth doesn't have to be an issue either. AMD could put a 384-bit GDDR5 bus, or even go crazy with 512 bits, since it has to be shared with the CPU. That's doable, AMD did it with R600, NVIDIA did it with GT200, it's nothing new.

The problem is that you would end up with significant development costs, an expensive platform with a very wide bus (expensive motherboards, lots of memory chips, expensive GDDR5) and all that for a market that's going to be pretty small. There just aren't that many gamers looking for high end hardware, and many of them might prefer to stick with a regular CPU and discrete graphics, for various reasons; such as Intel's higher CPU performance. It's just not worth it.

It's an issue of demand, not of insurmountable technical obstacles.

 

Well that was quick. It didn't take long for a hypocrite affiliated with Nvidia to start crapping on the thread.
http://forums.anandtech.com/showpost.php?p=35413641&postcount=52

 

From Videocardz: "Retail AMD Radeon R9/9000 series arrive mid-October."

This picture is supposedly from an AMD AIB partner's headquarters.

 

Looks like the top model will be a 256-bit 4GB GDDR5 SKU.

 

Silly season for sure!


Sooo is it Diamond Multimedia or Visiontek. I bet its Diamond... being in Cali and all, they are slackers. too funny