Core Parking

[n0cturnal]

This is very true. Very few games ever make full use of more then 2 cores, or the 64-bit technology.

It doesn't matter if they have full use of more than 2 cores, even a game as badly optimized as wow can utilize more than 2 cores, yes it puts most of the load on 2 cores but it still has a few minor thread that it can put on other cores as well, also lately there have been a large increase in games that can utilize even more than 4 cores. Stop looking back and plan for the future instead.

[TaintedOne]

-snip-

Well I appreciated the effort Cyrops

[Cyanotical]

Think you're confusing core parking with dynamic core assignment.

Core parking would be disabling one or more cores if they are no tasks requiring them running.
What you are talking about is dynamic core time allocation or w/e the name is, anyways it's about OS executing task X on core Y then moving it to core Z and back to Y and so on and so forth, why is it done? I'm not sure.

certainly possible, it's not something i've even looked into in probably over 6 years, i was under the impression that core parking was related to locking threads to certain cores and then c6ing the unused ones to save power

[Cyrops]

And the results are in

Testing rig:
i7-4771, 1600:9-9-9-24 16gigs (4x4) RAM, intel hd 4600 graphics.
Testing:
Flight path from Shrine to Binan (just north of shrine, short flight).
Used /timetest it disables drawing units on ground during flight and fixes camera at the 'ground' level looking straight into your chars back, does not adjust zoom.
Ran flight path 6 times to Binan and back to Shrine, so total of 12 flights.
When second wow was launched I had the char afking in a building in Mist island (panda starting zone).
Software:
Built in WoW /timetest and Fraps benchmark logs.


My early results hold true!

single wow average fps: 48,591 (4 parked) 52,012 (unparked)
double wow average fps: 29,356 (4 parked) 31,703 (unparked)

single wow fps increase by having cores unparked: 7%
double wow fps increase by having cores unparked: 8%

Conclusiong:
Having parking disabled increases multithread performance.

For raw data you can get this and check my numbers: astroverse.eu/storage/cpu_core_parking_data.ods

Notes:
Interestingly, when 4 cores were parked, max usage per core was 57 43 48 48 %(5-11 parked cores)
When parking was disabled max usage per core was 97,41,54,23,60,63,55,37% (3 cores had higher max usage than when parking was enabled).

Edit:
This just hit me, does this test debunk that WoW doesn't need more than 2 cores to run great?
I have this urge to park 6 cores and run the test again >.>

[Cyanotical]

This just hit me, does this test debunk that WoW doesn't need more than 2 cores to run great?

that was debunked a long time ago, but you've added further evidence to prove it

[Cyrops]

that was debunked a long time ago, but you've added further evidence to prove it

Yea gonna save the link to this thread for future X)

[yurano]

I performed a similar test to Cyrops, fixing issues I found with his test setup:

• i7-4771 has varying turbo speeds based on num cores active
• Software core parking questionable
• iGPU results aren't ideal
• Stock cooler likely used, leading to potential issues with thermal throttling

Test system

• i7-4770K @ 4.2 Ghz - same turbo ratio regardless of num cores active
• H100
• 4x8GB 1600 Mhz C8
• Gigabyte 760 (stock)
• Windows 8.1
• WoW PTR 5.4.2 17658 (Release x64) Dec 6 2013

Methods

• Settings - Max GPU load 85%
• Core parking was achieved in BIOS. Settings varies based on number of real cores active and whether HT was enabled.
• 8 /timetest flights were measured for each test setup. The flight was between Gao-Ran Battlefront, Townlong Steppes and The Sunset Brewgarden, Dread Wastes.
• Switching between test setups required a restart so data from the first two flights was discarded for each test setup.
• Dual core results had an oddly erroneous test flight 4. Flight 4 was discarded and two additional test flights were performed.
• Average FPS reported below. STD refers to the standard deviation of AVG FPS for the flights.

Results
Spreadsheet

Gao-Ran to Sunset

	AVG	STD	% of i7
Dual Core	104.023	0.611	98.2%
i3	97.259	0.386	91.8%
Tri Core	106.113	0.622	100%
i5	103.471	0.791	97.7%
i7	105.909	0.739	100%

Sunset to Gao-Ran

	AVG	STD	% of i7
Dual Core	102.553	0.918	94.8%
i3	99.441	0.652	92.0%
Tri Core	108.196	0.337	100%
i5	108.290	0.892	100%
i7	108.090	0.499	100%

Conclusion
The minor increase in performance between Dual and Tri core results show that WoW has two primary threads with miscellaneous thread(s) sitting on a third core, if available. The 2-5% performance penalty when going from Tri to Dual core represents the performance demand of these miscellaneous thread(s).

No performance difference between Tri core and i5 results show that the miscellaneous thread(s) don't require more performance than a real Haswell core.

Interestingly, i3 performance is lower than Dual core performance. Miscellaneous thread(s) on HT interfere with real core performance, affecting the bottleneck primary threads. HT interference is greater than the penalty from accommodating misc and primary threads on the same real core.

Minimal difference (within margin of random error) between i5 and i7 results show that the HT interference doesn't affect WoW on i7s.

[Cyrops]

I performed a similar test to Cyrops, fixing issues I found with his test setup:

• i7-4771 has varying turbo speeds based on num cores active
• Software core parking questionable
• iGPU results aren't ideal
• Stock cooler likely used, leading to potential issues with thermal throttling

The table only proves that turbo boost increases speed most when you have several cores parked vs when parking is disabled.
What do you mean "Software core parking questionable"? It's an option in Windows, not sure what is questionable about it?
iGPU results are relevant, results don't get skewed by GPU.
Stock cooler? really? Evo212 cooled by double Enermax Case Fan Magma 120mm (CPU never goes past 55C on full load).
And the fact you ran win8.1 and I ran win7 might affect results.

Also you did less tests than me. Besides I had Fraps do the test as well.

Can't copy paste your results as it doesn't copy the decimals -_-
and when you have HT enabled it reduces the avg FPS?


Nice test with SLI @evn

[yurano]

The table only proves that turbo boost increases speed most when you have several cores parked vs when parking is disabled.

Its already widely known that turbo boost affects performance. Your test only proves that turbo boost on an i7-4771 works as expected; higher turbo clock speeds from parked cores improves WoW performance.

Whats unknown is whether core parking affects performance if clock speeds are held constant across test cases.

In my opinion core parking is nice feature for single threaded applications, but due to some 'bug' or oversight, the cores don't get unparked efficiently.

Your conclusion is that core parking improves single threaded performance. I'm saying that core parking improves single threaded performance because it activates turbo.

In a situation where there is no turbo, core parking does not improve single threaded performance.

What do you mean "Software core parking questionable"? It's an option in Windows, not sure what is questionable about it?

Software core parking doesn't fully park all cores because there's some residual activity on the parked cores. Check Task Manager if you don't believe me.

BIOS core parking ensures that a core is truly parked.

iGPU results are relevant, results don't get skewed by GPU.

iGPU usage heats up the CPU die and can affect turbo.

Stock cooler? really? Evo212 cooled by double Enermax Case Fan Magma 120mm (CPU never goes past 55C on full load).

Except this was never stated.

And the fact you ran win8.1 and I ran win7 might affect results.

Again, you didn't state this in your post.

Windows 8.1 is the correct choice because it is the most up to date.

Also you did less tests than me.

The relative SD from my tests are small enough to demonstrate a significant difference between the test cases. Choosing an appropriate sample size is the first rule of running proper statistics.

and when you have HT enabled it reduces the avg FPS?

Only when 2 physical cores are active. HT doesn't affect results when there are 4 physical cores.

[Cyrops]

Its already widely known that turbo boost affects performance.

While that may be true, my test in no way confirms that. If anything my test shows that when you have no cores parked the turbo boost works more efficiently.

Your test only proves that turbo boost on an i7-4771 works as expected; higher turbo clock speeds from parked cores improves WoW performance

If what you said in bolded was true, wow would have higher peformance with core parking enabled, which it didn't.

Only when 2 physical cores are active. HT doesn't affect results when there are 4 physical cores.

Your test results are iffy. For example comparing i5 vs i7: 103.471 vs 105.909 (G to S) that's almost 2.35% increase in avg FPS.
But then 108.290 vs 108.090 (S to G) you get a Decrease in FPS?
And then your dual core outperforms quad core, while best performance is from tri core. Usually dual core is better than quad core if the dual core has higher clock speed, but since you keep speeds at a constant I am really confused by the results. They just seem to contradict everything.
And to quote yourself, rewording a bit "and you used an overclocked CPU, leading to potential issues with thermal throttling"

Regardless our tests are too different in configuration.