I hope however typed that transistor count up realized that does not include the memory controllers, nor the shared cache the hd3000 igp has access too. Also none of the power control module (which, to be fair, that is only estimated above ~100million transistors).
You are right. The only thing that annoys me is that you are the only one who seems to even realize such a thing most just. Argue benchmarks and look at a GPU like a blackbox.
It is a bit of an unfair comparison. Yet counting the entire LLC to the GPU wouldn't be fair either. Added IMC, Powermanagement, PCIe controllers as a dedicated GPU would have, one might have to add some 100-150 million transistors. Which just shows how ridiculous it is to cry for a new 520M chipset when the GPU on DIE can save so much.
Yet comparing the 6450 where the 32simds are well over 200 millions transistors just for the simds and stuff they cannot live without like the rops and the stuff that feeds the simds.
Yet the performance difference is small at still 2:1 in transistor count.
I am aware that AMD packs much more tightly and that more than makes up for that seemingly bad number and they run on lower clocks which means it doesn't draw more power.
It still means though that Intel is not that far behind AMD/Nvidia in "what they can do" as some people here insist on being a fact.
Now with the Intel HD 4000 and the 100% increase in power efficiency of the EUs (due to 22nm and design changes) it looks like the Hd 4000 beat anything from AMD/Nvida in W/speed with IB. And the most efficient little GPU that is fast enough is IMO the perfect on DIE GPU.
Intel won't invest in Drivers nearly as much as AMD/Nvidia will. Even if they put somthing that can really rival a 400 core Fusion CPU on the next DIE (at least I doubt they would). It would probably just suck more power than necessary and still make few people really happy. I prefer the small integrated GPU for all the normal stuff and have a big dedicated GPU with much better cared for drivers for actual gaming.
It also not entirely clear how fast a GPU you can put on a CPU DIE before it needs a 3rd 64 bit memory channel. The Fusion GPU sits next to a comparably slow CPU that runs at low clock speeds when the GPU is in use. How much memory bandwitdh remains and at what point would if be just a way too big chip. Also at a certain performance point a GDDR5 equipped dedicated GPU is probably much more efficient.
There are many things one must account for in designing a decent on DIE GPU and great many people in this thread who complain about the preformance don't answer a single one of them.
In theory Intel could put a bigger GPU into the LV Dual Core IB because that has more memory bandwidth to spare. It is a smaller chip than the Quad Core. Quad Core notebooks are most of time quipped with dedicated GPUs and nobody would complain. The only problem is that Intel would want to charge quite a bit more for such a CPU and they want to use one and the same for Core i3 or Core i5 in the very cheap PC Notebooks. I am pretty sure they know all the costs involved and just don't think it is worth it to make say more different variants of chips. They want to bin the LV chips form all the rest and sell that stuff cheap. If they make higher end Dual Cores they probably couldn't match the yields and demand as well.
