At least, that's what I said when I read this article on Tulsa, the upcoming server processor from Intel, and found that it has ONE THOUSAND FOUR HUNDRED MILLION transistors.
Let me repeat that - ONE THOUSAND FOUR HUNDRED MILLION!!!!!!!! That's more bloody transistors than there are people in India!!
*pant, pant, drool, drool* goes the Electrical Engineer in me. Moore's law, may you long hold!
In contrast the Opteron has around 270 million transistors. And check out the photos that you can die for on each of those links. Get it? Die for, processor die, hee hee.
Of course, as I read on /., more and more it looks like Intel is actually a massive manufacturing company, incidentally using its R&D to keep the fabs full. These fabs are among the most technically advanced in the world - few others are shipping 65nm parts in the volume that Intel is!
In this particular case, with Tulsa, Intel is clearly showing the pain it faces with the totally out-dated NetBurst architecture. With its 20+ pipeline stages, the NetBurst architecture requires an extremely low-latency cache to keep bubbles short in its pipeline. In a server architecture, where the code keeps jumping around, making a mockery of Branch Prediction, cache is all the more important. The schmoo-plot in the Tulsa page, and the detailed description, show that the Tulsa can actually run at 3.9GHz! That's a clear victory on processor speed for Intel though.
And that's where the large number of fast transistors bites back - Tulsa runs out of margin on thermal dissipation! The Register had an interesting article quoting an Intel exec that if the power consumption on processors kept increasing at the same rate it was, the heat density would surpass a nuclear reactor!
And what's the biggest cost on the Tulsa? The massive 16MB L3 cache. Necessary not just because of the long pipeline, not just because it's a server processor, but also because the front-side bus is shared between the two cores, unlike the Opteron which has a separate HT link from each processor.
So how're the other guys making it up? Sun's got Niagara, and the upcoming NiagaraII and Rock, AMD is going gung-ho on Opteron and derivatives, IBM has already been doing multi-core, and is looking at Power6 in the near future. These guys have been concious of IPC, power budget and that frequency boosts increase power dissipation quadratically, for quite some time. While Sun is going for 'torrents' of simple processor - 8 processors with 4 threads each in Niagara, AMD and IBM are going for more hefty dual-core and quad-core processors.
Well that was long :-). Let me know if you've got any drool-worthy pics!