nMP Turbo Boost

[Tutor]

Poking Up The nMP’s Performance Potential By Raising Its Turbo Boost Potential

The potential for Turbo Boost (Turbo) frequency [and the stage of that boost] is based on power, temperature and current. The more current being drawn by the CPU, the less likely it is that the highest Turbo Stage will be reached. The hotter the CPU, the less likely it is that the highest Turbo Stage will be reached. The more watts being used by the CPU, the less likely it is that the highest Turbo Stage will be reached. So if you want to maximize the nMP’s Turbo Boost potential you have to assess what you can realistically do to control each of the three factors. On the Mac platform, two of the factors are beyond your reach, as you should use the platform to suit your needs and not the other way around, i.e., I wouldn’t recommend modifying what applications you run, what other applications you run simultaneously, or when you run them just to get a desired performance metric. That leaves you with the temperature factor, which is something that you can affect. Also, don't forget that GPUs like it cool too. How you might affect/control the temperature factor is up to you.

Have you ever heard or use the old school phrases, “Put a cap on it” or “Chill.” Those phrases might still have some currency here. From the low hanging fruit to that at the tree’s pinnacle, here’s three of my suggestions.

1) Keep the area where you set your nMP, cool and clean. Cleaniness is important because you do not want to have dust/air flow blockage issues which could raise the operating temperature of your system. Keeping your nMP cooler by setting your heating/air conditioning system(s) reasonably can help improve system Turbo potential.

2) I’ve begun looking into the potential of the nMP for improving my audio/ visual workflow. If I decide to get one, it’ll be a refurb 4-core with a couple of beefy GPUs. The first things that I’d do after getting it would be (a) swapping in a used Sandy Bridge 8-core, (b) setting my heating/air conditioning system reasonably, with the performance of my computer systems in mind but not at the expense of my personal comfort, (c) cleaning my workspaces and keeping them clean, and (d) putting a cap on my nMP. But it wouldn’t be a cap like many might first imagine - it’ll be a stocking cap made from a discard. The leg of the stocking will be allowed to extend to the base of the nMP so that the nylon will help act as a dust filter. That’ll reduce the need to blow my nMP to keep it clean internally. Keeping it cleaner internally should help to improve it’s performance by reducing dust buildup related thermal increases. But here’s the kicker, in the foot portion of the stocking that would be at the top of the nMP case, I intend to install a quiet helper fan - something along the lines of this [ http://www.directron.com/tfd9515m12zp.html ] but that fits snuggly, to help maximize my CPU and GPU related performance.

3) If you’re really OCD about maximizing Turbo Boost potential, a somewhat extreme alternative is chill your nMP with, e.g., a DRE-030 Desktop case with a 800 or 1500 BTU Solid state air conditioner - http://www.eicsolutions.com/workstation-electronic-enclosure.php. It'd give much more environmental control and appears to be large enough to house some some other peripherals that might also benefit from that environment. It'd also act as a den to hide some snakes. Don't like blue, there's Krylon.

 

[sirio76]

Unfortunately, the forum pretty printer for that link clips the full name. The last part of that is "/estimating-mac-pro-performance/". That isn't a benchmark. A benchmark is a measurement, not an estimate.

Not really an estimate, this are results from actual benchmark scores:

http://browser.primatelabs.com/geekbench3/86294

http://browser.primatelabs.com/geekbench3/179463

And looking at your link they seems consistent with other geekbench results from similar nMP:

 

[Gav Mack]

Looks similar to when a chip isn't getting enough offset voltage thrown at it while stepping up or down.

There is one other factor after tearing down lots of Apple kit even under the warranty period - poor thermal paste application and quality of paste itself. If the paste is substandard over the turbo boost core on the die it will hit temperature too quickly and not clock high enough regardless of whatever you try to max the boost out.

It will be interesting to see the hardware service manual when it's released - I haven't seen a change in thermal paste policy to engineers from 'glop this much out of the syringe on is good' so far!

 

[VirtualRain]

Lots of good discussion here... before I get into the details, I've been doing a bit of reading, and to some extent, what deconstruct said earlier about forcing cores to be disabled in order to achieve the highest turbo speeds has merit.

From what I've read, in order to achieve the highest turbo modes, you not only need ample headroom, but the processor has to put one or more cores into the C3/C6 sleep state. This obviously comes with some compromise in terms of overhead in putting cores to sleep and waking them. And given all the background tasks OS X has going on, this would be of dubious benefit.

At any rate, it seems like Apple is not even taking advantage of this feature. Which would explain why the top Turbo states I'm seeing are limited to those were cores do not need to be put into C3/C6. Some DIY motherboards allow you to disable cores, and force turbo modes, these are typically the only reliable way I've read to see those top speeds.

Looking at this benchmark: http://www.primatelabs.com/blog/2013/11/estimating-mac-pro-performance/
seems that the 8core beat both the 4 and 6core in single threat performance, just like the turbo was working better.
That's the reason why I go for the 8core version, I do not necessarily need many core on my main workstation(I've a small farm for multithreaded software that offer much better price/performance than every dual Xeon ws), but the single core performance is very important while working.

When you (or anyone else) get's there 8-core or 4-core, I'd really like to do some comparisons to see if what you're suggesting is true.

I have an odd way of viewing increase. That's why I suggested that you load up all of the cores with multiple tasks. To me the sweetest point is Turbo stage 1. Turbo stage 2's sweetness is questionable/dubious. The least sweet turbo point is Turbo stage 3. And that's the way I've always viewed Turbo. As the core counts increase, the sweet point moves closer towards no-turbo speed, unless there is hardly any drop off in the number of cores that participate at each higher stage and the stage level differences are significant. For example, if you have a 6-core running at 3.5 GHz at base and Turbo stage 1 is 3.6 GHz and everyone of those cores participates at stage 1, to me you've gained 6x100 MHz or gained 1.029x per core over their base. If the top level (Turbo stage 3) is 3.9 GHz and only one core is participating at a time at that level, then all you've gained is 4x100 MHz and for that one core's bounty of 1.114x over it's base, the other cores take breathers in-between each of their potential substitutions for that one. To me you could be paying a big price in overall performance for the 3.9 GHz bragging right. So having gotten into this crevice of my mind, it probably now comes as no surprise to you why I'd consider stage 2's sweetness dubious. You could get 4-core participation at 3.7 (which would be 4x200 MHz, but 4-cores working simultaneously is less than 6-cores working simultaneously) or you could get 2-core participation at 3.7 (which would be 2x200 MHz, but 2-cores working simultaneously is less than 6-cores working simultaneously). So if, like me, you need the power to do multiple taxing things simultaneously, then are the winner even though I know that you wanted to, but didn't, see 3.9 GHz. Check out this article when you get some free time [ http://www.computerwoche.de/fileserver/idgwpcw/files/1595.pdf ]. Among other things, it confirms what I've experienced and read elsewhere - " ... the Turbo frequency [stage reached is] based on power, temperature and current. Thus, to the extent you keep your cores cool (and if you could underclock them to lower their current draw and help keep their wattage below TDP) then you can increase the chances of your getting to those higher stages if that is where you truly want to go.

That's a good way of looking at it... and you're right, given the number of threads and processes executing nearly continually on my system, I'd much rather have 6x100 and occasionally 6x200.

And I agree with you, that turbo boost depends on power, temperature, current, and load, but it also requires support in firmware, particularly for C3/C6 low-power states. From what I've read, without that, the top turbo modes will never be realized.

I'd say the best way to compare would be to take a 4 core and a 6 core, throw at them the exact same single threaded task and see which finishes first. If the 4 core finishes considerably faster that means the turbo on the 6 core isn't really going as high as the turbo on the 4 core.

Yes, you could do this, but that's only necessary if you feel you can't trust Intel's own monitoring tool. I have no reason to believe that Intel's Power Gadget is not reporting the correct CPU speed. Simply comparing IPG log files should be sufficient to see which machine is performing at a higher clock speed.

I think to see the full single core turbo-boost you need to make sure the total running processes won't add up to more than 100% CPU load, otherwise processes will need to run on more cores. So you'd be more likely to see the maximum turbo boost when your workload is limited to say, 95% load, for example.

There are also other factors. For example, while pushing all processes onto a single, faster running core might seem like a good idea in theory, it isn't necessarily the best thing to do, as the cost of swapping different processes in and out is still fairly high in order to give them each some CPU time; however, keeping them on separate cores can be better overall as you need less such switching per core, which means less wasted time. OS X can actually have a pretty huge number of threads even when idle, each of which will need to get some CPU time now and then, as even sleeping threads may wake up periodically as well, the end result is that it can be very difficult to setup a situation where you have few enough processes that running them on a single, faster core is better than spreading them out to reduce wasted time switching processes.

Of course we also don't know enough about how OS X decides how to allocate processes to cores.


So yeah, not sure you're likely to really see the "fastest" turbo boost setting. Maybe in future the technology will get better at running one core at full speed, with the others still available for idle/background processes, and OS X will take full advantage of it, but it's just one of those things that in reality you won't see a huge benefit from.

It'd be interesting to see the same tests on a 12-core CPU, as while it's maximum turbo boost speed is lower, it may still spend most of its time at the same speeds. You'd hope so anyway, given how much it costs

I think Apple sees it the same way you do and perhaps this is why they haven't implemented C3/C6 states.

So two things to get to max Turbo. Thermal headroom and also workload that needs to get done. If don't have both there is little good reason to ramp up the clock to max.

I think I've demonstrated that there is plenty of thermal headroom and I've thrown a variety of different workloads at the CPU, but I think it comes down to a third thing that's missing... proper firmware support for EIST, C-states, etc.

Have you ever heard or use the old school phrases, “Put a cap on it” or “Chill.” Those phrases might still have some currency here.

I'm tempted to setup my system temporarily on the porch. While it's not as cold as Chicago it would buy me another 15-deg C lower ambient which should certainly confirm if temperatures and Apples cooling system are in any way responsible for lack of turbo. I seriously doubt it though.

 

[sirio76]

When you (or anyone else) get's there 8-core or 4-core, I'd really like to do some comparisons to see if what you're suggesting is true.

I'll check that for you.
Anyway, going down from 3.9 to 3.7ghz is just a 5% decrease in CPU performance, I think you can live with that

 

[deconstruct60]

Not really an estimate, this are results from actual benchmark scores:

For two but even article overtly says are filling in the blanks from other system. If trying to measure the scale of turbo across models you'd need real results under real conditions.

And looking at your link they seems consistent with other geekbench results from similar nMP:

Given using Xeon E5 measurements not going to have huge variation.

The quirkness of the Geekbench and single thread results still stands when there is variation in that there really is no single thread unless firmware disable all the other cores. The shifting numbers are in part indicative of how much control or not the users invoke before running the test on their host systems.

----------

.....
I think I've demonstrated that there is plenty of thermal headroom and I've thrown a variety of different workloads at the CPU, but I think it comes down to a third thing that's missing... proper firmware support for EIST, C-states, etc.

Actually not. You have narrowed the context down to the CPU but have left out what the CPU is coupled to. The dynamic shifting of all the three of major heat sources can have a feedback failure mode that you really haven't touched on at all.

 

[VirtualRain]

Actually not. You have narrowed the context down to the CPU but have left out what the CPU is coupled to. The dynamic shifting of all the three of major heat sources can have a feedback failure mode that you really haven't touched on at all.

I'm not really sure what you're getting at here.

 

[iBug2]

Btw I have been playing with the power gadget on my rMBP and I have never seen it 3.7 Ghz either, at most I've seen 3.5 Ghz.

 

[Derpage]

I'm not really sure what you're getting at here.

I believe he's pointing out that there are more thermal factors than just the CPU's innate qualities when you attach multiple devices to a single heat sink. The throttling may occur due to temperatures increasing in a region that isn't typically associated with the CPU now that multiple items all share the same heat sink.

 

[VirtualRain]

Btw I have been playing with the power gadget on my rMBP and I have never seen it 3.7 Ghz either, at most I've seen 3.5 Ghz.

Yeah same here... if I run CinebenchR15, I can peg my i7-3540M at 3.5GHz and it apparently has a max Turbo of 3.7. Even if I limit Cinebench to 1 thread, the CPU still maxes out at 3.5GHz.

I think it simply comes down to Apple not implementing C-states... which is actually fine, since there's overhead in sleeping and waking cores.

However, one thing I learned from this, is that when it comes to evaluating the max turbo clock I'm likely to see, I need to chop a couple multipliers off the top.

 

[deconstruct60]

I'm not really sure what you're getting at here.

The firmware settings of what Apple does/doesn't do need to take into account the whole system. You can't design a Mac Pro solely around the CPU and its TDP. How the whole system runs ( and the feedback connections between the components ) ways in on much actual leeway there is on thermal headroom.

If all three major heat sources try to go "max possible TDP" at the same time may not end up with a good outcome in terms of conflicting feedback that will flow into the three diferent control mechanisms.

Similar issues to different degrees in other system designs. None of these subsystems are 100% decoupled from one another in other designs either. Mac Pro is higher but it is nothing new to have coupling.

 

[VirtualRain]

The firmware settings of what Apple does/doesn't do need to take into account the whole system. You can't design a Mac Pro solely around the CPU and its TDP. How the whole system runs ( and the feedback connections between the components ) ways in on much actual leeway there is on thermal headroom.

If all three major heat sources try to go "max possible TDP" at the same time may not end up with a good outcome in terms of conflicting feedback that will flow into the three diferent control mechanisms.

Similar issues to different degrees in other system designs. None of these subsystems are 100% decoupled from one another in other designs either. Mac Pro is higher but it is nothing new to have coupling.

Yes, I think this is stating the obvious... No where in my testing, did I try to invoke CPU Turbo Boost with all three major heat sources at max possible TDP. In fact as you can see from my testing, I tried to invoke Turbo Boost under the most benign conditions: all heat sources at idle steady-state with a single-threaded CPU task.

 

[deconstruct60]

... In fact as you can see from my testing, I tried to invoke Turbo Boost under the most benign conditions: all heat sources at idle steady-state with a single-threaded CPU task.

And which part of the firmware knows this is the case? Your whole point is that you are testing under a narrow subset of conditions. The whole range of conditions actually matter in design; including being statefully aware.


That is in part what I was wondering if Apple put any effort into. If increase the coupling then if don't increase how "smart" the thermal management system is then the design will take on limitations. Can't increase complexity and keep just a "dumb" management.

 

[VirtualRain]

Now that a lot more people have their nMPs in hand, I'm wondering if anyone else can report the turbo boost clocks they're seeing. Especially Quad and 8-Core CPU variants.

I'd recommend the Intel Power Gadget (link) to monitor CPU clock speed and Cinebench R15 which allows you to load a number of CPU cores of your choosing. Note your clocks at idle (browsing), with one R15 thread, two threads, and max multi-threading.

Let's compare notes.

 

[Bad Leopard]

Now that a lot more people have their nMPs in hand, I'm wondering if anyone else can report the turbo boost clocks they're seeing. Especially Quad and 8-Core CPU variants.

I'd recommend the Intel Power Gadget (link) to monitor CPU clock speed and Cinebench R15 which allows you to load a number of CPU cores of your choosing. Note your clocks at idle (browsing), with one R15 thread, two threads, and max multi-threading.

Let's compare notes.

Total noob question here ...I know I know, but, how do you open the power gadget after double clicking the installer? I am not seeing it anywhere.

 

[VirtualRain]

Total noob question here ...I know I know, but, how do you open the power gadget after double clicking the installer? I am not seeing it anywhere.

Hmm... There should be a folder called "Intel Power Gadget" in your Applications folder... perhaps try using Spotlight?

Anyone else want to try some turbo boost testing between SuperBowl commercials?

 

[AidenShaw]

Hmm... There should be a folder called "Intel Power Gadget" in your Applications folder... perhaps try using Spotlight?

Anyone else want to try some turbo boost testing between SuperBowl commercials?

I don't have a "SuperBowl Commercials" folder on my system. Do I need to tell Spotlight to cripple my system for a few days looking for it?

 

[Bad Leopard]

Hmm... There should be a folder called "Intel Power Gadget" in your Applications folder... perhaps try using Spotlight?

Anyone else want to try some turbo boost testing between SuperBowl commercials?

Yep, I'm an idiot. I'm new to Macs, and the other day I un-ticked the option to show external drives on the desktop so that I wouldn't always see my external ssd Photoshop scratch drive....well I didn't realize that it would also hide the temp drives when installing a program, so it turns out I never really installed the gadget! LOL

Anyways, got it. Thanks!

 

[ha1o2surfer]

Another thing I don't see mentioned here much is something called Nominal TDP values. There are upper and lower "versions" of TDP that can have an impact on CPU Turbo Boost.

In the photo below, the upper TDP limit of my 3840QM is set to 72 watts
http://i.imgur.com/8alfXsm.png

This allows my CPU to turbo beyond 3.8Ghz with 4 cores active. If I set it to 45 watts the turbo is locked to 3.6. If Apple doesn't have configurable TDP enabled or has the upper limit locked to 80 watts, the CPU will never have the wattage headroom to turbo. This is also seen in their mac books as well to keep the temp down. Apple isn't dumb for doing this because setting the TDP to 72 watts yields a HUGE increase in power consumption (as seen in the picture) and a very low increase in performance. In other words, huge increase in power consumption for a small increase in performance doesn't make sense

 

[VirtualRain]

Another thing I don't see mentioned here much is something called Nominal TDP values. There are upper and lower "versions" of TDP that can have an impact on CPU Turbo Boost.

In the photo below, the upper TDP limit of my 3840QM is set to 72 watts
http://i.imgur.com/8alfXsm.png

This allows my CPU to turbo beyond 3.8Ghz with 4 cores active. If I set it to 45 watts the turbo is locked to 3.6. If Apple doesn't have configurable TDP enabled or has the upper limit locked to 80 watts, the CPU will never have the wattage headroom to turbo. This is also seen in their mac books as well to keep the temp down. Apple isn't dumb for doing this because setting the TDP to 72 watts yields a HUGE increase in power consumption (as seen in the picture) and a very low increase in performance. In other words, huge increase in power consumption for a small increase in performance doesn't make sense

Yeah, makes sense.

I think we're now at the point where we have to accept that the max turbo speeds advertised are like Unobtanium... However, I think it's still relevant for buyers to know, just what turbo speeds are obtainable.

I've demonstrated that the 6-core will operate under full load at 3.6GHz with an occasional jump to 3.7GHz under lightly threaded loads.

It would be useful to know what happens with the Quad and Octo processors as well.

 

[ha1o2surfer]

Yeah, makes sense.

I think we're now at the point where we have to accept that the max turbo speeds advertised are like Unobtanium... However, I think it's still relevant for buyers to know, just what turbo speeds are obtainable.

I've demonstrated that the 6-core will operate under full load at 3.6GHz with an occasional jump to 3.7GHz under lightly threaded loads.

It would be useful to know what happens with the Quad and Octo processors as well.

haha unobtanium, good one! Apple seems to advertise a CPU's nominal clock speed mostly because of this. Macbooks are often referred to as 2.6/15/512 blah blah. I can't help to wonder why

If you are willing to try, I would try using a program called Throttlestop under windows to see if you can force the CPU to turbo higher.

 

[Scissors]

I receive my 8-core nMP on Friday so I thought I'd chime in on turbo performance. I don't have Cinebench, but I do know that the terminal command "yes > /dev/null &" is a very easy way of creating a single thread which occupies 100% CPU time, so I used that instead.

The list below is number of "yes" threads and the CPU speed most consistently observed by the Intel Power Gadget. Admittedly, the test was over a relatively short period of time (a couple of minutes for each step up), but it still reveals an interesting result.

1: 3.77
2: 3.67
3: 3.49
4 to 16: 3.4

On the downside, the turbo doesn't crank up to quite as high as it is billed for on 1 to 4 cores (9/8/7/5). On the upside, the turbo does reach its full potential on 5 to 8 cores, matching the advertised 4/4/4/4. On balance, I'm quite impressed with this.

When I have time, I'll measure this over a longer period and report back.

Hope this is of interest.

Mark

Edit: In case anyone tries to use the "yes > /dev/null &" command, to kill all "yes" processes, just type in "killall yes".

 

[VirtualRain]

I receive my 8-core nMP on Friday so I thought I'd chime in on turbo performance. I don't have Cinebench, but I do know that the terminal command "yes > /dev/null &" is a very easy way of creating a single thread which occupies 100% CPU time, so I used that instead.

The list below is number of "yes" threads and the CPU speed most consistently observed by the Intel Power Gadget. Admittedly, the test was over a relatively short period of time (a couple of minutes for each step up), but it still reveals an interesting result.

1: 3.77
2: 3.67
3: 3.49
4 to 16: 3.4

On the downside, the turbo doesn't crank up to quite as high as it is billed for on 1 to 4 cores (9/8/7/5). On the upside, the turbo does reach it's full potential on 5 to 8 cores, matching the advertised 4/4/4/4. On balance, I'm quite impressed with this.

When I have time, I'll measure this over a longer period and report back.

Hope this is of interest.

Mark

Edit: In case anyone tries to use the "yes > /dev/null &" command, to kill all "yes" processes, just type in "killall yes".

Thanks. This appears consistent with the 6-core... you're able to hit Turbo Speeds one notch below maximum occasionally (3.8GHz in your case vs 3.7GHz in my case) on lightly threaded workloads and on most multi-threaded workloads, you hit minimum rated turbo speeds (3.4GHz in your case vs 3.6GHz in my case).

This means that in practise, the 8-core is slightly faster at single core operations than the 6-core, slightly slower at multi-threaded applications (3-6 cores), and then obviously better with 7-8 cores. I assume during browsing and other light tasks, your CPU is at 3.0GHz?

It will be interesting to see results from the Quad core... I wonder if it can hit 3.9GHz under any scenario?

 

[Scissors]

I assume during browsing and other light tasks, your CPU is at 3.0GHz?

Yup, it idles at 3.0GHz, but as I'm sure is the case with yours, doesn't take much effort at all to cause that to rise.

 

[peabo]

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