Originally Posted by DJRumpy
And I will ask again, is there anything, other than your belief (as in links to real benchmarks) showing an improved scalability for quad or dual-quad cores under SL other than increases that one would expect from additional cores?
Phoronix has been benchmarking SL against Linux based machines. Haven't been to the sight in awhile so I don't know the status of the testing which has been on going. However some codes are showing very good behaviour under SL. Unfortunately testing to date has been on a dual core machine.
At the most basic level and assuming GDC is 100% efficient, a dual-core could process a task in half the time that a single core could.
Yes a highly parallizable task. This is actually a key concern the app or sections of it need to be amendable to parallel processing in the first place. I can assure you though that the apps that have parallizable code will benefit from the additional cores.
Again at 100% efficiency, a quad core could process the same task in 25% of the time that a single core could.
We both know however that 100% seldom happens. In any event why deny the utility of more cores? Nobody that I know of denies that GPU processing works for an assortment of apps rather well, so why deny the value of four cores over one or two?
You're implying that simply adding more cores will somehow make GDC more efficient.
Efficient is your word. That isn't what I'm really saying at all. What I'm saying is that there is a thread pool managed by the OS that can make use of all cores available to it. That is the huge concept.
Now combine this with the fact that the programmer doesn't need to know about the hardware anymore to leverage those cores then you have a bigger win. With GCD the system figures out how many cores to run the parallel threads on. This means that additional cores in many cases automatically speed up the programmers code.
The GCD will simply work with what it has, be that 2 cores, 4 cores, or 8 cores, or whatever. It won't suddenly get more efficient with more cores,
Exactly, GCD works with the cores available to it. This can speed up an apps highly parallel code significantly as more cores are added. The obvious point here is that for many users four cores will be far better than two.
It will work within that efficiency regardless of the number of cores, and we know that it isn't 100% efficient (it will never be able to dole out all work perfectly across X number of cores in any case).
Yes it is true that there is no such thing as 100%. Depending on the code you can come awfully close though. The converse is also true in that some code is very much bound to a single prcessor or a small number of threads.
There are other benchmarks floating around the net that highlight GCD and SL but I don't have links at the moment. You could try looking at www.macresearch.org
?? And see what they have to read up on.