Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.

Still no info here on Turbo Speed vs Cores used. The Max speed (always quoted everywhere) is always for just one core used. I have followed the link for Datasheet from the W-2155 page on ARK and it goes to a tableof XEON data sheets that does not include the W series as far as I can tell. I have seen people post these kinds of things - where do they find them?

https://ark.intel.com/products/125042/Intel-Xeon--W-2155-Processor-13_75M-Cache-3_30-GHz
View Datasheet goes here
https://www.intel.com/content/www/us/en/processors/xeon/xeon-technical-resources.html
 
Last edited:
Its the engineer in me :)
Where can I find the data sheet for the new iMac Pro processors that gives the Turbo Freq vs Cores used numbers?

AFAIK the chips used are a downclocked version of the Intel W Xeons:
https://www.intel.com/content/www/us/en/products/processors/xeon/w-processors.html

No link on these gets to a "real" chip level datasheet that I can find.

Thanks!

I've been playing with the Intel Power Gadget trying to understand just what the Frequency display is referring to.

I experimented on my 3.5 GHz MP6,1 6-core that has a max Turbo Boost of 3.9 GHz.

Using Instruments (/Applications/Xcode.app/Contents/Applications/Instruments.app) to control the number of active cores and no Hyperthreading I can get Power Gadget to display various Frequencies while a single process runs using 100% CPU.

6 cores - Varies between 3.60 to 3.68 GHz
5 cores - Varies between 3.59 to 3.66 GHz
4 cores - Steady at 3.7 GHz
3 cores - Steady at 3.7 GHz
2 cores - Steady at 3.7 GHz
1 core - Steady at 3.9 GHz (max Turbo boost)
 
  • Like
Reactions: kohlson
^^^ This makes some sense to me and I would approximate that
1 Core = max 3.9
2-4 = 3.1
5-6 = 3.6 something
 
I played some more....

Using Instruments (/Applications/Xcode.app/Contents/Applications/Instruments.app) to control the number of active cores and no Hyperthreading I can get Power Gadget to display various Frequencies while a various number of processes running using 100% CPU.

With single process using 100% CPU

6 cores - Varies between 3.65 to 3.68 GHz
5 cores - Varies between 3.65 to 3.68 GHz
4 cores - Steady at 3.70 GHz
3 cores - Steady at 3.70 GHz
2 cores - Steady at 3.70 GHz
1 core - Steady at 3.90 GHz (and will sometimes will fall off to 3.70 GHz)

With two processes using 100% CPU

6 cores - Varies between 3.60 to 3.62 GHz with occasional spike to 3.40 GHz
5 cores - Varies between 3.60 to 3.62 GHz with occasional spike to 3.40 GHz
4 cores - Steady at 3.70 GHz
3 cores - Steady at 3.70 GHz
2 cores - Steady at 3.70 GHz
1 core - Steady at 3.70 GHz

With three processes using 100% CPU

6 cores - Steady at 3.60 GHz
5 cores - Steady at 3.60 GHz
4 cores - Steady at 3.70 GHz
3 cores - Steady at 3.70 GHz
2 cores - Steady at 3.70 GHz
1 core - Steady at 3.70 GHz

With four processes using 100% CPU

6 cores - Steady at 3.60 GHz
5 cores - Steady at 3.60 GHz
4 cores - Steady at 3.70 GHz
3 cores - Steady at 3.70 GHz
2 cores - Steady at 3.70 GHz
1 core - Steady at 3.70 GHz

With five processes using 100% CPU

6 cores - Steady at 3.60 GHz
5 cores - Steady at 3.60 GHz
4 cores - Steady at 3.60 GHz
3 cores - Steady at 3.70 GHz
2 cores - Steady at 3.70 GHz
1 core - Steady at 3.70 GHz

With six processes using 100% CPU

6 cores - Steady at 3.60 GHz
5 cores - Steady at 3.60 GHz
4 cores - Steady at 3.70 GHz
3 cores - Steady at 3.70 GHz
2 cores - Steady at 3.70 GHz
1 core - Steady at 3.70 GHz
 
  • Like
Reactions: kohlson
^^^^^
Cool - This looks pretty right to me (I had the same machine as well)
now if only you had the iMac Pro too :)
 
  • Like
Reactions: KalanHowse
8-core iMac Pro, using stockfish (chess engine) to selectively utilise 1-16 cores. Using a chess engine allows one to see how much output one is getting in terms of problem solving, as this does not necessarily scale linearly with the clock speed (or indeed the number of cores the engine is using).

First of all, using only 10-20 second short tests where CPU temperature does not get a chance to rise above 80C. Using Instruments.app to alter processor cores, and Intel Power Gadget.app to measure CPU speeds and temps.

(a) if on default out-of-the-box 16 active processor cores with hyperthreading on, then it makes no difference whether Stockfish runs on 1, 2, 3 ... all the way up to 16 cores, the clock speed is 3.90GHz in all instances.

(b) If I ask Stockfish to use just 1 core ...
(i) and turn down the CPU to 1 active core without hyperthreading, then the clock speed goes up to 4.2GHz.
(ii) CPU is on 2 cores, still 4.2GHz
(iii) CPU 3 cores, 4.0GHz
(iii) CPU 4 cores, averaging 3.98GHz
(iv) CPU 5 cores, 3.92GHz
(v) CPU 6-8 cores, 3.90GHz

Please note that Stockfish actually runs at least 10% FASTER as single core on 16-core CPU at 3.9GHz, than it does on 1-core CPU at 4.2GHz, presumably due to interruption by other processes. "Efficient" execution of single-core Stockfish, i.e. where delivers chess as fast as on an out-of-the-box iMP, is only reached when 4 or more CPU cores are active.

Now at longer execution times on out-of-the-box iMP,
(a) using 16 Stockfish cores, when CPU temp hits 91-93C (which is fairly soon, within a minute or two), then the CPU starts to throttle and speed jumps to and fro between 3.63 and 3.90GHz
(b) using 8 Stockfish cores, even though CPU temp VERY gradually rises to 93C NO throttling occurs, and it continues to run at 3.90GHz. I watched this for about 15 minutes and it was flat-lining the curves at that point, steady state.

Note that Stockfish runs FASTER at 8 cores than it does at 16 cores even before throttling sets in: hyperthreading 16 cores is less efficient than running 8 unhyperthreaded cores.
 
  • Like
Reactions: kohlson and bxs
8-core iMac Pro, using stockfish (chess engine) to selectively utilise 1-16 cores. Using a chess engine allows one to see how much output one is getting in terms of problem solving, as this does not necessarily scale linearly with the clock speed (or indeed the number of cores the engine is using).

First of all, using only 10-20 second short tests where CPU temperature does not get a chance to rise above 80C. Using Instruments.app to alter processor cores, and Intel Power Gadget.app to measure CPU speeds and temps.

(a) if on default out-of-the-box 16 active processor cores with hyperthreading on, then it makes no difference whether Stockfish runs on 1, 2, 3 ... all the way up to 16 cores, the clock speed is 3.90GHz in all instances.

(b) If I ask Stockfish to use just 1 core ...
(i) and turn down the CPU to 1 active core without hyperthreading, then the clock speed goes up to 4.2GHz.
(ii) CPU is on 2 cores, still 4.2GHz
(iii) CPU 3 cores, 4.0GHz
(iii) CPU 4 cores, averaging 3.98GHz
(iv) CPU 5 cores, 3.92GHz
(v) CPU 6-8 cores, 3.90GHz

Please note that Stockfish actually runs at least 10% FASTER as single core on 16-core CPU at 3.9GHz, than it does on 1-core CPU at 4.2GHz, presumably due to interruption by other processes. "Efficient" execution of single-core Stockfish, i.e. where delivers chess as fast as on an out-of-the-box iMP, is only reached when 4 or more CPU cores are active.

Now at longer execution times on out-of-the-box iMP,
(a) using 16 Stockfish cores, when CPU temp hits 91-93C (which is fairly soon, within a minute or two), then the CPU starts to throttle and speed jumps to and fro between 3.63 and 3.90GHz
(b) using 8 Stockfish cores, even though CPU temp VERY gradually rises to 93C NO throttling occurs, and it continues to run at 3.90GHz. I watched this for about 15 minutes and it was flat-lining the curves at that point, steady state.

Note that Stockfish runs FASTER at 8 cores than it does at 16 cores even before throttling sets in: hyperthreading 16 cores is less efficient than running 8 unhyperthreaded cores.

redshift27: Thanks for that posting. Very interesting and helpful in understanding how the core frequency varies with workload.
 
redshift27: Thanks for that posting. Very interesting and helpful in understanding how the core frequency varies with workload.

Yes - excellent and insightful post! Many times in my Audio work (though in the last couple of years this is less of a factor) I have found a bit better low latency recording performance by turning HT off.
 
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.