nMP Turbo Boost

[VirtualRain]

For those interested, here are some results for the 12 core:

1 thread: around 3.3Ghz peaking at 3.37Ghz
2 threads: around 3.2Ghz peaking at 3.26Ghz
3 threads: around 3.1Ghz peaking at 3.15Ghz
4 threads: around 3.0GHz peaking at 3.08Ghz
5-12 threads: solid 3.0Ghz
24 threads: 2.97Ghz

Thanks. It's consistent in that it also cannot hit top turbo speed.

To characterize the results anecdotally, I would say that the 12-core is definitely a trade-off in clock speed for cores whereas the 8-core offers a great balance of both while the 6-core is the bang-for-the-buck leader (all IMHO of course).

 

[Scissors]

I'd be really interested to know whether the different approach to turbo frequencies between the 4,6, 8 and 12 core processors is the result of technical limitations, or whether instead it's a product of Intel's long term development/marketing strategy. This isn't my area of expertise at all, but my instinct tells me that processors with 4,6, 8 or 12 identical cores running 4x 100% threads would all draw and radiate the same overall amount of energy. Accordingly, why can't they all run at 3.9GHz?

 

[VirtualRain]

I'd be really interested to know whether the different approach to turbo frequencies between the 4,6, 8 and 12 core processors is the result of technical limitations, or whether instead it's a product of Intel's long term development/marketing strategy. This isn't my area of expertise at all, but my instinct tells me that processors with 4,6, 8 or 12 identical cores running 4x 100% threads would all draw and radiate the same overall amount of energy. Accordingly, why can't they all run at 3.9GHz?

Well according to the specs they could all run a single core at 3.9... but not four cores. However, to do that, the CPU actually needs to shut down power to the unused cores since an idle core still consumes power. These are the C3/C6 sleep states and there's overhead in sleeping/waking cores, which incurs a performance penalty likely negating any benefits of running a single core at 3.9GHz unless it's for an extended duration. Apple does not appear to have implemented these states, which is probably a good thing... I think we're all better off with OS X spreading all it's threads across 6, 8, or 12 cores running at 3.xGHz rather than trying to force everything through one core running at 3.9GHz.

The bottom line is, that the top rated turbo boost speed, is pure fantasy. People shouldn't look at that when evaluating CPU choices. In any real system (Mac or Windows) you'll only ever see the second top turbo speed. Although I say that without having seen what the 4-core does... maybe it can hit 3.9Ghz?

 

[Scissors]

Thanks for that VirtualRain.

I realise my question was badly put - I should have said "a single core" at 3.9GHz.

What you say about Apple not fully sleeping cores makes sense.

 

[sirio76]

The bottom line is, that the top rated turbo boost speed, is pure fantasy. People shouldn't look at that when evaluating CPU choices.

I can't speak for the other models, but the 8core can boost to 3.77ghz(or 97% of 3.9ghz). I don't call that "pure fantasy", to me 3% look more like a rounding error
It would be interesting to see if those CPU consistently reach the declared turbo under windows or Linux. On the nodes I'm using for distributed rendering running Windows7, I've set all the cores to run @4.125ghz, but the cores do not run exactly at the same speed, they are more like c1=4.121ghz, c2=4.123ghz, c3=4.120ghz, ecc.

 

[VirtualRain]

I can't speak for the other models, but the 8core can boost to 3.77ghz(or 97% of 3.9ghz). I don't call that "pure fantasy", to me 3% look more like a rounding error

Well, you can view it that way if you like, but this isn't horseshoes or hand-grenades. The fact is, none of these CPUs are hitting their top advertised turbo speed.

In the case of the 8-core, if you look at the Intel Power Gadget log, you'll see that 3.77 is not the actual CPU clock speed, it's simply an average. Actual CPU clock speed is a multiple of 100MHz and in the lightly threaded workload you're referring to, the multiplier is switching back and forth between 37 and 38. If it was hitting 3.9GHz, the log file would show it.

So it may be close, but I stand by my remark that top turbo speeds are pure fantasy (until someone shows me an Intel Power Gadget log that says otherwise).

It would be interesting to see if those CPU consistently reach the declared turbo under windows or Linux. On the nodes I'm using for distributed rendering running Windows7, I've set all the cores to run @4.125ghz, but the cores do not run exactly at the same speed, they are more like c1=4.121ghz, c2=4.123ghz, c3=4.120ghz, ecc.

Indeed... I did some research on this in a few Windows enthusiast forums (like Xtremesystems.org) where I use to hang out and the same situation appears to hold true for Windows systems. To achieve top turbo boost speeds, you need a motherboard and bios that supports C3/C6 sleep states and you actually need to manually force sleep on multiple cores for the CPU to trigger the top turbo state. So it can be done, but not under normal operating conditions on Windows or Macs.