Shanghai / Istanbul vs Nehalem?

[IntelUser2000]

Did anyone know Nehalem has a 16 stage pipeline?? Well... it does. I assume the 2 extra stages are to enable multi-threading as they did with Atom.

 

[Idontcare]

Originally posted by: IntelUser2000
Did anyone know Nehalem has a 16 stage pipeline?? Well... it does. I assume the 2 extra stages are to enable multi-threading as they did with Atom.

Why would multi-threading necessitate an increase in the number of stages? Hyper-threading is a means of running things in parallel, i.e. duplication of hardware internal to the core.

If the stages changed it is no doubt because Nehalem is an entirely new architecture and not necessarily due to any one specific architecture feature.

I'm sure Dmens could give us some insight into why, hypothetically of course, Intel would want to increase Nehalem's stages relative to Penryn.

 

[jones377]

Originally posted by: Idontcare

Originally posted by: IntelUser2000
Did anyone know Nehalem has a 16 stage pipeline?? Well... it does. I assume the 2 extra stages are to enable multi-threading as they did with Atom.

Why would multi-threading necessitate an increase in the number of stages? Hyper-threading is a means of running things in parallel, i.e. duplication of hardware internal to the core.

If the stages changed it is no doubt because Nehalem is an entirely new architecture and not necessarily due to any one specific architecture feature.

I'm sure Dmens could give us some insight into why, hypothetically of course, Intel would want to increase Nehalem's stages relative to Penryn.

If you try to do more with the same number of pipeline stages you end up running at a lower frequency. I'm guessing Intel has rebalanced the pipeline after including SMT (and other improvements) and found that adding 2 more stages was the best trade off. BTW, doesn't the 1 extra cycle latency of the L1 cache account for one of those stages?

 

[zsdersw]

Originally posted by: Idontcare
I'm happy to see multi-threaded apps getting faster and faster, but I would have liked to have seen some xtor budgeted for giving us a 10% improvement in single-threaded IPC as well given that Nehalem is a tock and all.

4-door sedans are pretty much a box in front for the engine, a bigger box in the middle for passengers, and a smaller box in back for the trunk. You can't stray too far from that formula and call it a 4-door sedan. So too is it, I suspect, with IPC; it's pretty much as good as it's going to get with the all of the other considerations for CPU design taken into account.

 

[Idontcare]

Originally posted by: zsdersw

Originally posted by: Idontcare
I'm happy to see multi-threaded apps getting faster and faster, but I would have liked to have seen some xtor budgeted for giving us a 10% improvement in single-threaded IPC as well given that Nehalem is a tock and all.

4-door sedans are pretty much a box in front for the engine, a bigger box in the middle for passengers, and a smaller box in back for the trunk. You can't stray too far from that formula and call it a 4-door sedan. So too is it, I suspect, with IPC; it's pretty much as good as it's going to get with the all of the other considerations for CPU design taken into account.

Apropos analogy given today's focus on getting those cars off the power-consumption trend of the 20th century and into more fuel-efficient drive-trains while not really worrying about improving acceleration or top-speed.

I still want my flying car, with 300mpg if possible too please.