Peter Lewis (peter.delete@this.notyahoo.com) on June 2, 2022 6:07 pm wrote:
> What did you mean by Itanium was founded on “wrong assumptions
> about the increasing complexity of processors”?

One of the stated reasons for looking at VLIW processing, of which Itanium is a variant, was the increasing complexity of out of order processors. The idea was to simplify the processor and put the removed smarts into the compiler. The increasing complexity is still happening but better design tools, and other factors, have made it still manageable. And the compilers never lived up to expectations. And arguably still don't.


> > Unbelievably, they also appeared to think you could over take a market from the high end down
>
> Why is this impossible, even if it is different from the history of microprocessors?
> Apple started with high priced iPhones and later introduced low end models.

Not really comparing like to like though. Adding features to new generations of processors to target new markets seems more natural. I wouldn't like to try to strip a processor suitable for a Z series or similar down to one suitable for a laptop. How much work is it to take features away? Do you retain compatibility? If you don't retain compatibility what's the point? Targeting the mass markets gives you the revenue flows to spend on developing higher end specialty products.

One market I can think of that works from the top down is car safety features, often appearing on top of the line Mercedes and migrating to lower tiers of cars as mass adoption drives costs down. But some of that is forced by regulations (seat belts, air bags and crumple zones spring immediately to mind), some by other manufacturers being integrators and trying to move up the value chain (cruise control etc), and some marketing as great safety features become desirable features in other cars (semi automatic braking, higher level automation like lane control). Doesn't really fit in with processors though where excluding ECC memory (which should be near everywhere), there aren't many high end RAS features which would be worth the implementation cost for most systems (think 2 processors working in lock step, fail over features for RAM and hard drives etc).

>
> > and the need for as much as possible to be patentable.
>
> Only one patent is needed to prevent someone from copying a chip. There is no way the failure of Itanium
> has anything to do with a desire for patents. Companies file a lot of patents for bragging rights and
> so they can counter sue anyone that accuses them of patent infringement. When the first Itanium was
> introduced in 2001, it was already impractical to copy a chip. Anyone who introduces a new microprocessor
> instruction set should be happy if some other company wants to make a compatible product.

One patent could potentially be worked around. Each additional one increases the work that must be done to clone the processor. I don't think any company is happy when people copy their products. Intel's actions over the last 40 something years against AMD in particular show they weren't happy at all with them.