AMD Zambezi news, info, fans !

[-Boris-]

Sigh.



AMD A8-3850
Intel Core i3-2105

Idle
43.6W
51.7W

GPU Accelerated Video Transcoding
126W
85W

3D Gaming (Metro 2033)
126W
101W

CPU Load (x264 Encode)
123W
87.6W



Taxing the cpu = 123-43W = +80W!! (gpu is idle in both cases!)
Gaming = 126W-43W = +83W where gpu and cpu are used intensively. (There is no turboboost!, cpu runs at 2.9GHz).
Transcoding = "". high gpu load with load on cpu's (cpu is not idle and is running at 2.9GHz).

And yes you can compare those TDP values. Go read techreport whom tested a 3800 and had similar or lower powerconsumptions than the i3 for cpu load. (and not 36W higher)

Why focus on gaming, both GPU and CPU are loaded in gaming? It doesn't show anything. Compare GPU only(mostly) test with the CPU only(mostly) test. All you can tell is that in this particular case the GPU test loads the processor more than the CPU test. You really can't turn those figures to that it's the CPU that is the power hog. If you have other reasons for those claims post them instead. The figures you have there doesn't cut it to prove your point. And you can never compare power consumption between different reviews at different sites with different boards. So that's just another bad conclusion.

I do believe you guys might have a point, but it seems to that you haven't presented it here, the comparison above just doesn't prove your claims.

-Boris-


1) I was talking about Metro
2)
idle 43.6W
GPU trans-coding 126W the difference is 82W and the same discrete gpu with a bit higher clock as the IGP has a max. TDP of 39W 82-39=43W just for CPU if IGP is working at its TDP limit. So much for your statement how cpu is close to idle.

3) SB consumes more in Metro than in x264 Encode and if you don't know SB has Sandy Bridge Media Engine which is doing video encoding and decoding instead of GPU thats why the consumption is lower than in x264 Encode.

1. Yes, and metro doesn't show as it loads both CPU and GPU. That's why I said you should compare GPU test vs. CPU test.

2. Where do you have these numbers of internal TDP limits for different parts?

3. Still, in this particular test GPU test loads the processor harder than the CPU test.

I'm not even sure what the argument is anymore, all I can say with certainty is:

No because people forget some arguments and focus on other stuff, brings numbers that actually don't say anything thus derailing the discussion from the subject. The argument is, you can't predict Phenom behaviour on 32nm with Llano as a base. There is way to important differences for that to be possible.

- It's the CPU cores that draw the most power by far under 100% load. As mentioned almost all the TDP is eaten up by the CPU.

- If you artificially load up both the CPU and GPU you will far exceed the TDP and the system will overheat and shutdown on a std heatsink (I tried it).
This doesn't happen in real life gaming (even using all cores) because both the CPU and GPU fight for memory BW, rather than maxing out their execution resources.

- The GPU is on the CPU-NB power plane, and it's the CPU's stock Vcore that's pumped through the roof


I have figures from my own testing on my laptop i'll dig them out

You actually bring some points here. And I don't doubt that I might be wrong. You can please dig them out. The other arguments I've got this far haven't showed anything conclusive.

But my main point (which everyone forgets) is that Llano probably isn't made the same way as a Phenom II would be made at all. AMD talks about differences in silicon between different product lines, like Turion and Athlon/Phenom, and between Thuban and Deneb. It's clear that some design choices that might be good for a GPU isn't as good for a CPU. Low power high density trannies might be good for lots of units packed in a GPU at sub 1GHz frequencies. But might need more current to work good in a 3GHz CPU.


One thing more, I don't say that Thuban will be close to BD, it's just an if. If Thuban is close to bd, then I think a 32nm Thuban would be better of.