Hi guys,

Are these Intel dual and quad core chips really 2 and 4 individual processors or is it smoke and mirrors? If you get one of these chips can you run a true SMP OS or is it just that everything is transparent and you get approx. 2 or 4 times the power of one processor?

Thanks.

The processors have two computing cores. You need an SMP os to use both cores. Note that, generally speaking, this is worse than having a single core that is 2-4 times as fast.

I understand there is management overhead which makes 2 actual processors not as fast as 1 processor at twice the clock speed.

Are you saying these "cores" are like 2 completely separate functional processors?

Thank you.

Yes.

They are better than having the same number of separate CPUs (of the same specs) because the memory sharing and cache coherency management are tighter.

They also cost less than the same number in separate single core packages, so the idea of having multiple single core CPUs is obsolete. (There are packaging and thermal limits, so two four core packages still makes sense vs. one eight core package).

Yes.

Rarely even close. Likely much of the software you run won't be multi threaded well enough to make good use of more than one processing unit. Even if it were, a single CPU with double the speed (and double the cache size) would get much more work done. If they could make one CPU twice as fast at a comparable cost to a dual core they certainly would do that instead. (One cache, twice as large, could be made for slightly less than the cost of the two caching systems in a dual core package).

Summary: Dual core is slightly better than two single core packages of the same specs, but much worse than one single core package with double the specs (if that were available).

"management overhead" is a severe understatement for that issue.

What should "completely separate" mean? What should "processor" mean?

Excluding some obscure thermal and fault tolerance issues, any difference between dual core and completely separate is an advantage for dual core.

Thanks John for the great explanation. Just trying to learn enough about these chips to buy a new box!

The right answer is almost certainly dual core.

Even looking at the fraction of just CPU cost, the difference between single core and dual core at the same spec level is small. Looking at that difference as a fraction of the entire system cost may make more sense and then it looks even smaller.

For an ordinary home user, dual vs. single core won't approach a factor of two performance difference, but it will be a significant performance difference and worth the price difference.

The cost difference between dual core and quad core at the same spec level is much higher (as a fraction of total system cost). Also, comparing at the same spec level means you're ignoring the higher spec options that are available in dual core and not quad.

The performance difference between dual and quad at the same spec level for an ordinary home user is tiny, no where near as large as the performance difference between single and dual.

So unless you have specialized needs that make real use of quad core, it doesn't make sense.

By "spec" in all the above, I primarily mean L2 cache size and CPU clock rate. Which of those two matters more depends on what software you are running, but both are probably important.

Thanks again for this info. I will check in with you if I have more specific questions. I hope you stay on the forums!

Well, its not smoke and mirrors. And AMD have been, indirectly, making the slightly odd claim that an issue with the Intel quad core parts is that they aren't four CPUs on one die, which is slightly difficult to understand... (You might as well ignore this point! AMD is just trying to score meaningless marketing points off Intel here.)

I'm not sure how this is a useful question. You can run an SMP OS on a single core processor. In fact, these days its often done in Linux. What you don't get, is any advantage out of doing it.

I'd like to answer that question in three ways: Yes, No and It Depends.

Yes, under certain very particular circumstances, you can get a throughput approximately equal to double that of a single core at a similar frequency and with a similar amount of cache per core (and always assuming other things like the disk drives aren't a bottleneck). This may get close to occuring in, say, a server, where there are a myriad of little tasks queued for execution. It can also happen with certain specifically written apps (say applying transformations in a specifically optimised photo-processing app).

Also, No, it doesn't. This is also a correct answer. For certain apps, say one monolithic (single threaded) app, like many games (unless particularly re-written to take advantage of the cores) you get a throughput almost the same as a single core (with the usual conditions). Although, you could, maybe, be burning a CD at the same time, where, on a single core, your game might grind to a halt if you tried this kind of stupidity.

So, the real answer is probably, maybe. Maybe even definitively maybe. Somewhere in between in most cases. And if you only expect a small speed boost, then you are not going to be dissappointed. And that's probably what you'll see most of the time until apps are re-written. But the cost is small, and more and more apps will be re-written over time. And in a situation like a Linux box, there may be quite a few little processes that can be executed on the 'other' core from the one executing the 'big' application.

Well, in most cases 'management overhead' wouldn't be quite the right description, but its the right general idea: you certainly don't simply get 2x or 4x performance.

Yes. (Really, very nearly!) They are functionally separate processors. Depending on supplier (Intel and AMD) they have different arrangements with cache, and that cache sharing causes a small deviation from 2x or 4x, but that isn't the biggest problem (so you can ignore it, if you like, but you did ask).

Depends somewhat on the definition of 'better' that you choose.

They could make one cpu with twice the clock speed, if that's what you mean, but the power issues would be a bit troublesome. Two cpu cores at half the clockspeed does not give you the same power consuption once we get into this regime in which power consumption climbs non-linearily with clockspeed.

Now 'one core with twice the clockspeed' may be a piece of silicon that could be churned out economically, but then there are many other costs that go into the system (bigger power supply, oh lets have watercooling or maybe even N2, fans). And then there is all the criticism that Intel got over the rather warm, and therefore noisy, Netburst Pentium D parts. You think Intel wants to go back there again? (Although avoiding a re-run of that particular debacle may not be their main motivation.) And, in particular, with all organisations claiming 'greenness', this isn't where Intel wants to be.

If your definition of 'better' leans heavily on power efficiency, then this is important. If you just want to frag the bad guys and don't care if you need your own electricity substation to do it, then that's rather different.

I think you misunderstand what I was comparing in the place where you quoted my use of 'better'.

Given a core with specs X (including speed and cache, etc.)

Compare one package containing two of those cores (AKA dual core) vs. two packages, each containing one of those cores.

The former has better cache coherency management and better memory sharing, each of which result in a tiny improvement in system performance, so to the small extent that there is a performance difference, the dual core is "better".

There are still plenty of motherboards for two CPU packages. My point was that up to quad core, if you want more cores per motherboard, the best way to get that is more cores per package, not more packages per motherboard:
One package of 2 cores is much better price performance than two packages of 1 core.
One package of 4 cores is slightly better price performance than two packages of 2 cores each.
One package of 8 cores is not quite practical yet, so two packages of 4 cores each is the way to get 8 total.

Regarding the comparison of dual core with specs X per core vs. single core with spec 2X per core, you might need to define "2X".

The CPU has a speed and a cache size. But the cache also has a speed. Is it fair to say "2X" means the CPU, including the cache is twice as fast AND the cache is twice as big? On that definition of "2X" you couldn't find any real world workload for which Dual core at spec X could beat single core "2X".

Doubling CPU speed without doubling L1 cache speed is unsound. A fair "2X" might mean the CPU and L1 cache speed are doubled and L2 cache size is doubled, but L1 size and L2 speed aren't doubled. Then there would be some real world workloads for which Dual core X would beat Single core 2X. But still not very many. For a home computer user, VERY few.

Even with that more conservative definition of "2x" getting 2x beyond the spec of a moderate price dual core CPU is very prohibitive.

Cooling is not just a matter of conducting the heat from the surface of the CPU package to the outside world (solved by your water cooling suggestion). You must also conduct the heat from the gates to the surface of the CPU package. The more heat you must conduct, the greater the required temperature difference (package surface temperature below operating temperature of the gates). That 2X spec probably blows right past merely bringing the package surface down close to room temperature (all that water cooling does). You probably need to go below room temperature (At least a refrigeration unit and coolant).