Remember, the PCH doesn't contain the GPU anymore, it's on the CPU die, yet there we are with about the same apparent sizes.
Just that they are not the big one here is 150 sqmm. Which means by my previous calculation the 320M in 40nm would take almost two thirds of that die in 32 it would be about one thrid not any different than the HD 3000.
In any case looks already pretty cramped with the SFF packageing. A Quad wouldn't even fit an neither would a 32 EU Intel HD 3000.
2 things you still fail to grasp :
I grasp both of your remarks you either deliberately ignore my answers or don't understand them.
The 320M is 40nm and up till now that is what Nvidia had to work with. A 520M is no smaller. It doesn't matter that it is a 2010 GPU it is 40nm and big and that didn't change in the mean time. It will with 28nm but by that time Intel is already at 22nm.
Compare it to a 410M, 520M or anything you want. Those usually just have higher clocks but the shaders and die space didn't change. Also a 520M with 17W TDP is way to hot for an IGP it would need to run much slower
- the Intel 3000HD is built on a 32nm process contrary to the MCP89 which is built on a 40 nm process.
Now where did I fail to grasp that. I did mention it in the calculations more than sufficiently. It doesn't change a thing. Even normalized to the same process node Intel is about on par with Nvidia in die size vs performance.
You claim they are competent because they managed to achieve what the competition did the year prior. Congratulations. Meanwhile, if they hadn't been a bunch of cry babies and litigators, we'd be running 5xx series IGPs from nVidia that would literally crush the Intel 3000HD as has been proven time and again each generation
No I claim they are competent because they can cramp the same kind of performance on the same space. Regardless of the production year.
You claim they are incompetent because they didn't dedicated more space to the GPU.
Check out the specs of a 520M. That thing is no smaller than a 320M has faster clock speed but also too high a TDP and thus would have to run at somewhat reduced clocks. It is almost the very same architecture and the performance it gets from all that additional clock is not exactly crushing. Most games are still unplayable on anyting but low settings.
Compare that one to the AMD Fusion GPU on the A8 Llano much more impressive but also much bigger GPU. That thing is almost half the die but that is also the selling point.
I think when compareing die space used and power efficiency, that in terms of pure gaming power the AMD VLIW5/VLIW4 architecture is the most efficient of them all. No very good for GPGPU but great for gaming.
It sound as if you'd have to give up your religion of Intel hating. Look at the facts and explain why I am wrong about my claim that Intel is very much competent in what they are doing because they fit an about equally capable GPU on an equally big space (even when you normalize the data to the same process node).
And stop ignoreing stuff I have already explained. Either ask if I am unclear or point out actual errors.
Besides whining about the fact that there is no more any Nvidia IGP doesn't help. Intel got the Process node advantage and that more than makes up currently for any not quite as efficient architecture if it was on the same process node. A Nvidia 520M IGP wouldn't really change anything. It would only make the power managment really complicated with CPU and GPU Turbos(and any implementation much slower and less reliable). Also it would require dedicated memory exactly like a real dedicated GPU (which anybody is free to implement) because taking the IMC away from the CPU and move it back would cripple performance a lot. Dedicated IMC would mean addtional space, additional VRAM chips somewhere on the logic board, additional power consumption. What is then really benefit of just putting a real second dedicated GPU on the logic board instead of fusing it with the MCH. The space requirements wouldn't be all that different considering all the IMC stuff. And once you think it worthwhile to add a 520M to MBA it would be a waste to have a 32EUs Intel HD 3000 on the chip that would hardly fit into the SFF package anyway.
Also the good thing about a on die GPU is that it shares the LLCache which kills of the GPGPU bottleneck of data transfer. Probably one of the reasons why QuickSync is so much better than anything comparable on CUDA with so little hardware. This Intel GPU has the potential to be orders of magnitudes more efficient in excuting some OpenCL stuff than any Nvidia/AMD GPU.
You can complain all you want that the GPU is too small for you and you want a Intel HD 3500 which would be a double HD 3000 version. I just think you lack arguments to support Intel's incompetence in that field.
