There was an immeasurable distance between the quick and the dead: they did not seem to belong to the same species; and it was strange to think that but a little while before they had spoken and moved and eaten and laughed - W. Somerset Maugham
Himbeerkuchen: 25.6 GByte/s is correct, this is the max aggregate theoretical bandwidth of a full-width (2 quadrants) QuickPath link:

6.4 (GTransfer/s) * 20 (bits) * 0.8 (accounts for 8b/10b encoding) / 8 (bits/bytes) * 2 (directions) = 25.6 GB/s (per link)

Or 12.8 GB/s per direction because QuickPath links are dual-simplex. FYI full-width (32x32 bits) HyperTransport 3.0 links have an even higher bandwidth of 41.6 GB/s per link or 20.8 GB/s per direction.
From todays' opteron & Xeon processors, these produce 4X higher server scores, Remember there are 3 channels of faster memory, plus HT plus picture shows 2up.It creams AMD by full 1/3 Faster, if AMD 4 core even works by next year.
Thomas Drashek
Nope, he's right about the 25,6 GB/s number.
Remember, we're talking about very fast connections between a CPU and its North bridge here.

Currently, I believe the 12,8 GB/s is the norm with two DDR2 PC6400 800MHz modules in dual-channel mode.

So it's a nice x2 jump.
Why is the post "Intel puts flesh on Larrabee, Nehalem and Dunnington"?
I didnt see any mention about flash on these cores. That would really be something and a big surprise but the inq is posting fud!! Only mention of 16Mbs of cash which is alot bit its not flash!!! I was so excited to read this article but alas its just crap!!! Imagine 8core 16threads, 16megs of cash L1, L2 ,L3 cash 1gig of flash on a processor!!!!!
25.6 GBps (B reads "byte") means 25.6 x 8 = 204.8 Gbit/s. That would really be a lot of data, in fact about 20 times what a typical backbone optical fiber is carrying. Shurely you meant 25.6 Gbit/s in the first place?

AT adds: Intel says GBps...
Nice design. I'm sure the good
'people' at M$oft are busily working
to make sure their newest OS will drag these
processors down to pentium 90mhz
performance while at idle. Maybe they will include two onscreen clocks in the widgets,
and these will each use 2000Mhz to make sure they never show the same time as the one on the taskbar.
Himbeerkuchen: 25.6 GByte/s is correct, this is the max aggregate theoretical bandwidth of a full-width (2 quadrants) QuickPath link:

6.4 (GTransfer/s) * 20 (bits) * 0.8 (accounts for 8b/10b encoding) / 8 (bits/bytes) * 2 (directions) = 25.6 GB/s (per link)

Or 12.8 GB/s per direction because QuickPath links are dual-simplex. FYI full-width (32x32 bits) HyperTransport 3.0 links have an even higher bandwidth of 41.6 GB/s per link or 20.8 GB/s per direction.
The Intel powerpoint included in the article CLEARLY says

"25.6 Gb/sec max bandwidth"

NOT 25.6GB as stated in the article.
From todays' opteron & Xeon processors, these produce 4X higher server scores, Remember there are 3 channels of faster memory, plus HT plus picture shows 2up.It creams AMD by full 1/3 Faster, if AMD 4 core even works by next year.
Thomas Drashek
Nope, he's right about the 25,6 GB/s number.
Remember, we're talking about very fast connections between a CPU and its North bridge here.

Currently, I believe the 12,8 GB/s is the norm with two DDR2 PC6400 800MHz modules in dual-channel mode.

So it's a nice x2 jump.
I didnt see any mention about flash on these cores. That would really be something and a big surprise but the inq is posting fud!! Only mention of 16Mbs of cash which is alot bit its not flash!!! I was so excited to read this article but alas its just crap!!! Imagine 8core 16threads, 16megs of cash L1, L2 ,L3 cash 1gig of flash on a processor!!!!!
"DDR-3, 1066MHz and 133MHz"

Thats a bit of a jump ;-)
25.6 GBps (B reads "byte") means 25.6 x 8 = 204.8 Gbit/s. That would really be a lot of data, in fact about 20 times what a typical backbone optical fiber is carrying. Shurely you meant 25.6 Gbit/s in the first place?

AT adds: Intel says GBps...
Nice design. I'm sure the good
'people' at M$oft are busily working
to make sure their newest OS will drag these
processors down to pentium 90mhz
performance while at idle. Maybe they will include two onscreen clocks in the widgets,
and these will each use 2000Mhz to make sure they never show the same time as the one on the taskbar.