"Snow said the firm should be applauded for "looking at the technology landscape and saying it's time to change things.""

Aye, aye, Cap'n...... High time too and long overdue, which must bode well for the Future and if we consider Change to be XXXXPonentially Rated, I trust you are well strapped in for the Ride.

Proper Planning and Preparation Prevents Piss Poor Performance Permitting a Sane Disregard for Haste.

Doesn't matter if it's better than Intel or not. Intel will still out sell it.

AMD has been ahead of Intel before and not managed to grab a significant share of the market as Intel likes to bend the truth.

AMD will go bust when it can't go down to the next level of 45nm cores or lower. Then everyone will have to pay £1k+ prices for Intel's processors.

Bah Humbug

For over a year I've been debating how and with whom to go on multicore boxes. To date I've built and played with two Athlon boxes. I've committed myself to delving into Linux From Scratch and Xen, as I see the 64 bit boxes best used and protected by virtualization. But when it comes to multicore stuff two big issues jump out. First, with anything above a dual core, there seems to be resource issues as the quad cores are memory starved by hardware bottlenecks. Second, with mulitcores of quad core and above there are serious software issues. A physorg paper, ( sorry I can't cite it as I just gave it a quick read in passing ), suggested intel and all others can't see past problems in software design when trying to implement current practices on multicore platforms. From the reading I've done, over the at least the next 18 months, the best way seems to be to go with dual core, possibly high end gaming mobos.

As a developer of indie, multimedia stuffs ( a recent reincarnation borrowing heavily from a past reincarnation as a wilderness photographer ) I see virtualization on multicores using a stripped down version of, say, Debian and Xen dedicated to a small set of programmes to be the way to go to get the most from mulitcores. The best buys today would seem to be Asus's mobo for gamers like the Striker (intel) and Crosshair (amd). I'm open to being bitch slapped around and pointed in the right direction. :)

The memory bottlenecks might be (alright, probably are) an issue for Intel, but AMD's quad cores are supposed to sport 2 dual channel memory controllers on the CPU, which should assure that the quads scale up just as well as the dual cores did (in NUMA configurations, that is.)

I've been planning my next 8-core (two CPU) workstation, which among other things, needs to crunch large PDE systems. I've been waiting for the Barcelona for several reasons, and it looks like it is delivering what I expected:

1. Twice the memory bandwidth of Intel (in NUMA configuration)

2. DDR2 instead of those heinous FBDIMMs that Intel uses with the Xeons

3. Faster SSE floating point performance

Now I just have to wait a few more months for the 2.5 GHz parts.

My next big decision is WIndows XP x64 (which has worked perfectly for me) or chance installing 64-bit Vista?

Thanks for posting. I'm aware of the NUMA configurations and the AMD performance from recent postings on, IIRC, a Beowulf mailing list. Candidly my conflict arises from a need to develop on a platform that is accessible to others, who for lack of a better word, might be described as multimedia 'enthusiasts', who, I see, as much akin to gamers. Even if I were to look at a workstation box I still don't like what I see on the market today. Lastly, having come up from 286 boxes, I'm gun shy of implementing relatively, expensive hardware solutions to bottlenecks such as those exemplified by the history of the PC bus, the more so when there's no apparent, widely held consensus. I guess this is where the 'bleeding' part of being on the 'bleeding edge' comes in. Thanks again.

I do hope you are referring to the x86 platform here? The big boys have been addressing this issue for a LONG time. Probably the best known example is Sun's Niagara chip, an 8 core, 32 thread SPARC proc with 4 memory controllers. This chip rocked on data shoving, multi-threaded applications such as web serving. A significant reason for this is the large memory bandwidth.

Anyway, my point isn't to start a SPARC vs. x86 war - they are two different worlds. I just wanted to point out that this problem really has been dealt with a long time ago, with a pretty good degree of success.

-daniel

I am confused about the whole fbdimm thing they supposedly have some benefit of use but from the comments I have read almost everywhere they seem to be hot and expensive there is no way in hell either of those is good for a server they are price sensitive as hell and thermal loading is a big problem maybe AMD is not such a long shot.