AMD Zambezi news, info, fans !

[-Boris-]

Because it is the Cpu that consumes the power budget, not the gpu. The gpu is actually very clean and extremely efficient. (it is a complete marvell... far exceeding the efficieny of SB or any other gpu we know at the moment). They need the high voltages to get yields on the cpu, not the gpu. While it probably would do better without the gpu in the yield department. BD is also suffering issues on the 32nm node. So doubling the llano cores, adding fast l3cache will explode on the current process.... Currently having 50W for 4cores@2,6GHz with proper yields is pushing it for llano.... try double that, add cache and 1,5Ghz and see where that would get you. (most likely to a nuclair generator as power supply..).

I am not talking about the possibilities on a good working process, because that would affect BD also in a positive way.

How do you know that it's the CPU and not the GPU that consumes the power budget? Besides, the silicon and design could be limited by the compatibility with the GPU.

A Phenom II X6 at 32nm would be almost half the size of a Thuban if caches scales as good as cores when shrinking the process. BD must significantly outperform Thuban to justify this change in architecture.

[Manicdan]

How do you know that it's the CPU and not the GPU that consumes the power budget? Besides, the silicon and design could be limited by the compatibility with the GPU.

A Phenom II X6 at 32nm would be almost half the size of a Thuban if caches scales as good as cores when shrinking the process. BD must significantly outperform Thuban to justify this change in architecture.

run a cpu only benchmark
run a gpu only benchmark
run both

the power increase running a gpu only benchmark is like 20w increase, the cpu only is like 60w increase, and both is like 65w increase
these numbers are from memory and not to be considered accurate.

[flyck]

How do you know that it's the CPU and not the GPU that consumes the power budget? Besides, the silicon and design could be limited by the compatibility with the GPU.

First indication:

AMD A8-3850
Radeon HD 6550D
100W
4
2.9GHz
400
600MHz


AMD A8-3800
Radeon HD 6550D
65W
4
2.4/2.7GHz
400
600MHz


TDP significant lower while the only constant is gpu resources and clockspeeds.

second indications:

Reviews A8 3850. They include power figures. it is trading power draw blows with similar clocked 45nm x4 parts.
http://www.anandtech.com/show/4476/amd-a83850-review/9
POwer draw figure cpu only is right where it is for gaming.

A Phenom II X6 at 32nm would be almost half the size of a Thuban if caches scales as good as cores when shrinking the process. BD must significantly outperform Thuban to justify this change in architecture.

Like i mentionned, that is the ideal situation, but the same is true for BD. Ideally on a good process BD would be far better than what they can do now with the current issues. Your hypotetical BD replacement with thuban cores in llano form would simply not do what on this process node what you would expect them to do.

[-Boris-]

First indication:

AMD A8-3850
Radeon HD 6550D
100W
4
2.9GHz
400
600MHz


AMD A8-3800
Radeon HD 6550D
65W
4
2.4/2.7GHz
400
600MHz


TDP significant lower while the only constant is gpu resources and clockspeeds.

You know that you can't read TDPs like that? If one chip gets 66W and one gets 63W, the first will be labeled 100W and the other 65W.

second indications:

Reviews A8 3850. They include power figures. it is trading power draw blows with similar clocked 45nm x4 parts.
http://www.anandtech.com/show/4476/amd-a83850-review/9
POwer draw figure cpu only is right where it is for gaming.

As I read it the GPU intensive load maxes out the TDP more than the CPU intensive load does.

Like i mentionned, that is the ideal situation, but the same is true for BD. Ideally on a good process BD would be far better than what they can do now with the current issues. Your hypotetical BD replacement with thuban cores in llano form would simply not do what on this process node what you would expect them to do.

And you are making bad conclusions. It's reasonable to believe that the modifications in the silicon and transistor types needed for a working GPU isn't optimal for a CPU. GPUs are generally made for low frequencies but higher density on the most power consuming circuits. The design desicions made to fit a working GPU might be bad for a CPU. Llano is not a good example of CPU performance on 32nm. If a Thuban on 45nm isn't to far behind BD then what would happen with a Thuban made on 32nm? It would be almost half the size of Bulldozer, and be capable of higher frequencies than the original Thuban. And still have room for optimizations. And if your theory is correct and 32nm is botched and worse than 45nm, then maybe AMD should stick to 45nm for a while.