Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

Couldn't Intel and AMD implement a "Rosetta2" like strategy? That is, couldn't they ship CPUs that fundamentally are not decoding x86 ops, but some different ISA, and then layer a translation layer on top of it?

The Transmeta Crusoe used to do this, and I think the NVidia Denver cores did too?

I think fundamentally though this analysis highly depends on single-threaded performance being the bottleneck for most apps. A lot of sophisticated workloads are GPU or TPU driven, and so hand-waving away multi-thread performance, and treating single-thread performance as the end-all for client side performance I think is overemphasizing its importance.

Also, the "they don't control the stack" argument is wrong. Effectively, Intel/AMD + Microsoft acted as a duopoly, and if you toss in NVidia, the structure of DirectX and Windows is largely a cooperative effort between these OEMs. If Intel/AMD needed some fundamentally new support for some mechanism to boost Windows performance by 20-30%, Microsoft would work with them to ship it.



The PC world has already switched architectures 32 bit to 64 bit already so its doable.

The big issue I think is if you are Intel / AMD why would you do this? Even if there are performance gains do you really want to undermine one of the things (backwards compatibility) that distinguishes you from the competition? Plus what would the new ISA be?

Plus Intel still has the memory of Itanium.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: