AMD 64 fsb?

Come on... www.amd.com

Her står der en masse godt

Citat [START]

The 3200+ uses an fsb to go from the cpu to the memory controller. It runs at 400mhz and uses 64 distinct metal paths on the mobo, so that 64 bits can be sent per clock tick. Since there are 8 bytes in 64 bits, this produces a bandwidth of 3.2GByte/second. These 64 paths can be used to send and receive data, so like a one-lane bridge, traffic can only go in one direction at a time, which produces delays known as latency. 3.2GB/s is a theoretical limit that never gets approached in reality.
The amd64 does not have an fsb between the cpu and the memory controller. The path between them is called a hypertransport path. Instead of 64 paths that can either send or receive, it has 16 paths for sending, and 16 more for receiving. Instead of a 400mhz clock it uses an 800mhz clock. The amd64fx doubles it to 1600mhz

So the 3200+ has a max bandwidth of 400mhz x 64-bits = 3.2GB/s, and the amd64 has a max bandwidth of 800mhz x 16 bits x 2 directions = 3.2GB/s also. That's for the amd64. The amd64fx gets you to 6.4GB/s.

The lower latency of the new method is more important than the theoretical max throughput. How often did you need to send 3.2gb of data in one second with no need to change directions. Never. The new hypertranposrt implementation is far superior to the old fsb method because of its dedicated directional paths and reduced delays.

For now, all this really means is that your memory throughout is limited just by the memory speed, which for many systems was already the case.

But for amd users who heretofore used a single channel memory since dual channel provided no benefit, well now that dual channel solution is at least twice as fast as the old single channel memory.

The bottom line is that cpu advances come from increasing ghz of the cpu, reducing the amount of time its waiting on the rest of the system, or increasing what the cpu can get done in one clock tick. Bigger cache means the cpu waits less. Hyperthreading (for intel) means it waits less. Hypertransport (for amd) means it waits less. Quantispeed architecture (began with the amd xp) and 64-bit registers (begins now) means it can get more done per clock tick.

Citat [SLUT]

Taget fra

http://www.computing.net/cpus/wwwboard/forum/7690.html

Håber du kan bruge det :-)

<\\izemate