logoalt Hacker News

stinkbeetleyesterday at 6:48 AM2 repliesview on HN

AMD did launch amd64 server chips. The first ones were Opteron, first (and only until Nehalem) x86 with ODMC, glueless multiprocessor up to 8-way (though that was a bit flakey, 4-way was decent), and first dual core. It destroyed Intel's CPUs in many measure of performance for server workloads.

Before Intel's decade of humiliation with their 10nm fiasco and before Zen, AMD had a hell of a time getting ISVs and IHVs on board in the server space with producing them, certifying them, etc.,. Which wasn't entirely irrational, Intel was much more mature in a lot of ways in their hardware and software stacks, they were a big company, perceived as less risky, etc.

So it's not too surprising that those companies didn't jump to Opteron. Some did consider it, mind you. Though AMD went through doldrums of their own after Opteron (with Core2/Nehalem-ish era) where they shat the bed with their cores and were pretty handily beaten by Intel for quite a long while until Zen came along -- so that is possibly some vindication for the risk adverse by not going all in on Opteron.


Replies

adrian_byesterday at 9:44 AM

Already at launch, and also during the next few years, until Intel fought back with Core 2 (but in servers the Core 2 descendants became better than Opterons only in 2009, with the Nehalem Xeons), Opteron was immensely better than the UltraSPARC CPUs used in the Sun servers. Only the Fujitsu SPARC CPUs were somewhat competitive, even if still significantly weaker. The Intel Pentium 4 Xeons were also worse, despite higher clock frequencies.

But as you say, the inertia in servers is huge, so Opteron gained market share very slowly, despite being by far the most performant server CPU that was available.

Where I worked at that time, we had big server farms with Sun or Fujitsu servers, which were disliked by design engineers like myself for their slowness. We had to make great efforts to convince the management to buy a few Opteron-based servers. Of course, after we got them they ran circles around the Sun or Fujitsu servers, a single dual-socket Opteron could be much faster than an 8-socket Sun server whose cost was many times greater.

show 1 reply
fulafelyesterday at 7:30 AM

Yep, those vendors could conceivably been onboard and using launching on Opteron (if there was no Itanium and Intel was seen as the common enemy). They might have given AMD some help with the validation, RAS etc too.

Of course in the alternative history Intel might have done something else in place of IA64 that competed with amd64...