Can someone explain quad-pumped??

[PremiumG]

How can Intel quadruple the bus speed?? Is it like DDR2 by sending 2 bits on rise and fall of clock cycle?

Also, what's hyperthreading? Its tricking the programs to think they're two processors right? It seems that would slow down things rather than speeding up because of more "book keeping". Its really only one processor but through "book keeping" and additional overhead, it can be made to seem like two. So wouldn't it slow things down?

 

[PsharkJF]

1st : yes, much like DDR2.
2nd: HT (synchronous multithreading, if I remember the other acronym for it) is just the software/cpu/bios' effort of making sure all parts of the processor are in use in each clock cycle. HT speeds up multithreaded applications but can slow down extremely serial (sequential) programs.

 

[AnnoyedGrunt]

Just a bit more detail on hyper threading:
Basically, the newer P4 architectures have a long pipeline. That means it takes many small steps to complete one operation. If your cpu was an assembly line, then the P4 would have many stations, and each station would do a very simple task. The AMD chips on the other hand, would have far fewer stations, but each task would be more complicated. Therefore, the P4 assembly line would be able to pass off the work from one station to the next at a much faster rate, because each station does very simple operations. This is one way the P4 can run @ such a high speed.

However, often the CPU must make a guess as to what each station can do, beacuse the results from a previous command are still on the assembly line somewhere but they are needed for the next computation. The CPU will guess what the result should be, and begin executing down that path. Occasionally it will guess wrong, so it will throw away all that work at start over. At that point, you have all these little stations that are waiting for their data (this is called a pipeline stall).

What hyper threading does, is track multiple processes at once, so that if one of the threads stalls, the CPU can toss in other threads that are ready to be worked on. This keeps the pipline filled and reduces the overall impact of pipeline stalls.

So, hyper threading is really a way of masking one of the problems with the P4's long pipeline. That is one reason why the Pentium M and the A64 won't benefit from hyperthreading as much as the P4 will (both of those processors have a much shorter pipeline). There will still be some benefit to be sure, but not as much as the P4 sees. There is additional overhead in some cases, but most of the tracking is done in the CPU, and the hit is therefore overshadowed by the benefits when mulitasking or when running a hyperthreaded enabled application.

I think Arstechnica had a very detailed description of all this when they talked about the successes and failures of the p4 design. You might check out their article for the "real" story.

EDIT: Here's the article I was thinking of, with links to many other related ones:
http://arstechnica.com/cpu/004...future/prescott-1.html

-D'oh!