UpNDown, sgolds, et all re: being happy
I guess I sort of got my answer to "is anyone happier this week than last". Thanks for the comments. a couple issues I'll address.
1) It seems that many of us agree that it would have been nice for INTC to keep playing the Ostrich, and push the X64s back even further. but, no sense in crying over spilled milk.
2) It's good indeed that INTC validated the architecture.
3) I agree with many of the statements here that, all other things remaining equal, the hammers will often provide a better performance solution.
What I'm not as convinced of is that the platform in toto will get as much favorable funding. Things that make servers reliable - the type that Newisys built in their designs. If server manufacturers are doing this for opteron-based designs, that's great. If this gets the short end of the stick compared to X64s, however, I'd say that brand name and platform robustness could overwhelm the performance delta. But who knows - I've never been in the market for such a system, and personally don't know any people that are, so i'm kinda shooting blind on my thoughts here.
Also, sgolds - you are correct - it doesn't make much sense to compare today's K8s with 2007's Itaniums. However, the infrastructure needs to be built for K8s and X64 systems if they are to be Itanium replacements. As long as Itanium survives in any niche and development continues, it is possible that it will work its way out of a niche. Not assured, of course, but possible. In order for Itanium not to continue, it must be run out of both the HPC and Big Iron space. While hammer is probably suited to do ok in HPC (have we actually seen benchies from an Hammer HPC system?), it would need OEM support to thrive in the Big Iron space, and a general presence which it currently does not have, as I understand it.
Also, I'm wondering if anyone had any thoughts on the managed code aspect of things. aren't all the .Net languages using or moving to managed code, where most of the performance comes by choosing the right compiler for the target architecture, as opposed to hand-tuned assemply or the like? the IA-64 compiler teams would be primarily responsible for the performance increases on the software side, whereas developers wouldn't have to hand tune as much?If so, does this not reduce the progamming cost of developing for the IA-64 platform, if it's truly a "write once, review a little, target many" process? If I'm misunderstanding this, I'd welcome a correction.
