Intel IDF Day 1 - Quad Core, Santa Rosa And More

MojoKid writes "From demos of the new Alan Wake game engine on a 3.73GHz overclocked Quad-Core QX6700 to design showcases with a wafer of 80-core teraflop capable chips, Intel's IDF opening day was brimming with tech-wonder from the company affectionately known as Chipzilla. Paul Otellini also showed pics of upcoming fab facilities in Arizona (Fab 32) and Israel (Fab 28). In total, Intel will have three 45nm fabs by the end of next year at an investment of about $9B, all targeted 45nm manufacturing processes. Finally, a bevy of Quad-Core Kentsfield-based systems are shown here, with Dell and Voodoo's offering looking especially swank."

Comments

Moving fast now, eh?

[tygerstripes]

It seems the beast has awoken. They were so far behind in the chip-war with AMD for so damn long, it seemed their market share was getting nibbled to death.

Now they finally seem to have woken up and, by god, they are really moving now, aren't they. $9bn in 45nm fabs? A wafer of 80-core chips already? Speaking as a one-time AMD Fanboi, I have to say - the daddy is back.

(Let the flaming commence)

Re:Seems 80 cores isn't enough to resist the /. ef

[TheRaven64]

Sure, if you pick up Process Explorer, you'd think there's hundreds of threads running, but the truth is, most of those are idling.

Pretty much all desktop apps can be split into two categories:

1. The ones that contribute to the 5-20% load that your CPU generally sits under. (Web browsers, mail clients, music players, etc).
2. The ones that cause the load to spike at 100% for extended periods. (Audio/Video encoders, compilers, typesetting engines, etc).

Applications in category 1 will not see any benefit from a CPU that's twice as fast. That 5-20% load may drop to 2.5-10%, but no one cares. Those on category 2 will complete in half the time (assuming that they are CPU-limited and linearly scalable). As CPUs get faster, more and more things fall into category 1. Once you run out of things in category 2, stop upgrading. This happened for a lot of people about five years ago.

I recently found out about an interesting experiment Intel did a few years back. They have a full-system simulator that allows them to test various things easily. They modified it so that all CPU operations took zero (simulated) time to complete. This gave about a 2.5x speed improvement for most tasks, i.e. an infinitely fast CPU only gave a 2.5x speed boost to most tasks. It doesn't take a huge speed increase before you run out of CPU-limited things and start hitting memory, disk, and network bottlenecks.