Itanium and compiler changes
Itanium and compiler changes
Posted Feb 14, 2025 19:24 UTC (Fri) by khim (subscriber, #9252)In reply to: Itanium and compiler changes by farnz
Parent article: New leadership for Asahi Linux
Note that Itanium wasn't designed by idiots. Like Transputer they were designing the CPU for the wold where development of single-thread core have “hit the wall” near 100MHz and thus new ways of faster execution were needed.
In that imaginary world of slow CPUs and fast memory access VLIW made perfect sense and was, in fact, one of the most promising designs.
But after Athlon hit 1GHz, at the end of XX century… it became obvious that Itanic just simply make no sense in the world of fast CPUs and slow memory… but Intel had to push it, for marketing reasons, even if it was doomed to fail and it was obvious that it has no future.
Posted Feb 15, 2025 16:37 UTC (Sat)
by farnz (subscriber, #17727)
[Link] (1 responses)
Itanium was a bet that it would be hard to scale clock frequency, but that it would be trivial to go wider (both on external buses and internally). As a bet to take in 1994, that wasn't a bad choice; the failure at Intel was not cancelling Itanium in 1998 when it became clear that Merced would not fit into 250 nm, that process scaling wasn't going to allow it to fit into the next couple of nodes either, and that once you'd trimmed it down to fit the next node, it wasn't going to perform well.
Then, Pentium 4 was a bet that it would be hard to scale logic density, but that clock frequency would scale to at least 10 GHz. Again, wrong with hindsight, but at least this time the early P4 releases were reasonable CPUs; it's just that it didn't scale out as far as intended.
Posted Feb 15, 2025 17:40 UTC (Sat)
by khim (subscriber, #9252)
[Link]
What's notable is that both times bets sounded perfectly reasonable. Cray-1 reached 80Mhz is a year 1975 and PA-7100 and Pentium, POWER2, SuperSPARC… all fastest CPUs for almost two decades topped at around that clock speed! Assuming that trend would continue was natural. Then, suddenly, during Itanium fiasco, 100Mhz barrier was broken and clock speeds skyrocketed… assuming that this trend would continue wasn't too crazy, either! More importantly: when Intel realized that P4 story is bust, too – it quickly turned around and went with P6 descendants… Tejas was cancelled… but for some reason Intel kept Itanium on life support for years.
It's notable that Intel bet twice in a row on technology futures that didn't happen.
Itanium and compiler changes
> It's notable that Intel bet twice in a row on technology futures that didn't happen.
Itanium and compiler changes