2018 MBP i9: Testing w/ Volta & Fan Control

[Aea]

TLDR: Power limit to 45W and you'll see a 15% performance jump.

I ran ten Cinebench CPU Benchmarks in a row, giving the machine adequate time to cool down to idle between each run. No software changes were made and the CPU was idle otherwise (only running the software below & some small menu bar apps)

Software Used:
* Cinebench https://www.maxon.net/en/products/cinebench/
* Macs Fan Control https://www.crystalidea.com/macs-fan-control
* Volta http://volta.garymathews.com/

 

[mango316]

very nice informative analyses, thx!

 

[Standard]

Thanks for this information and taking the time to do this. Nice pup you have there !

 

[winterny]

The throttling is not thermal as far as the CPU overheating ... It's power throttling because the VRM's that are powering the CPU are being overloaded, and overheating.

When the VRM reaches it's maximum temperature, the system sends a signal to the CPU to drop it's multiplier to 1, which drops the clock speed to 800Mhz. VRM cools down, and it will then turboboost for a short time, overload the VRM again, and throttle again.

The throttling back and forth is less efficient than just holding at a power draw rate which the VRM can handle.

 

[Aea]

The CPU is overheating. The CPU VRMs are overhearing.

Those seem like splitting hairs. I don’t see how the latter isn’t a thermal throttle by another name.

 

[Feenician]

Great analysis @Aea. Thanks a lot. I really wish Volta could be set with more granularity so we could see where the limits lie.

 

[winterny]

The CPU is overheating. The CPU VRMs are overhearing.

Those seem like splitting hairs. I don’t see how the latter isn’t a thermal throttle by another name.

I make the distinction because even if you modify a 2018 i9 MBP by putting better Thermal Paste, it won't improve your performance because the VRM is being maxed out before the heatsink/fan/TIM setup.

Also, if the CPU were being thermally throttled, it would hold a relatively steady clock speed rather than jumping from 800Mhz to full Turbo and back every couple of seconds, and the overall performance would be better if the CPU *was* thermally throttled.

 

[content]

Please correct me if I am wrong, but according to notebookcheck, there is a 13% gap between the mid model 6-core 2018, and the base model 4-core 2017. And by limiting power consumption to the processor on the 2018, this gap widens. The 2018 now sees 20% more performance. However, at the cost of closing the gap between the two when doing every day work that does not tax the processor. Further it appears, they would need to limit power consumption to the 2017 as well, to make the test completely fair. I mean, the gap may get close to 13% all over again, if they did this. If they applied the same tweak to the 2017 as well.


And you have just shown that even when limiting the power consumption to the i9, it still has a worse score than the mid model 6-core 2018 with the same tweak. Ultimately making the i9 even closer in performance to the base model 4-core 2017. All without the same tweak applied to the base model 4-core 2017... Which may now make all of these tweaks meaningless to an extent?


I am referring to:


First article:


https://www.notebookcheck.net/The-n...nd-clearly-beaten-by-the-XPS-15.317264.0.html


Second article:


https://www.notebookcheck.net/Apple...e-performance-with-a-few-clicks.317552.0.html

 

[Aea]

An apples to apples comparison would be running the machines at their best configurations.

If that’s stock on the 2017 and power limited on the 2018 that should be fine.

The tool I use does not have the ability to set a higher power limit from OSX. The actual ‘peak performance’ setting might be higher than 45W.

 

[joptimus]

Thanks for the review! Do you also have the 13“ model or only the one with the i9?

 

[ezerez]

Thx for the review!

I have i7-8850H and i also get alot better results in cinebench when i put 45W in Volta. cinebench gets around 950 and my cpu is stable at around 3ghz and around 85c. So there even seems to be some headroom! Would be great if i could put the limit higher then 45W to test. Maybe with another tool or if Volta gets an update?

greets

ezerez

 

[dylin]

Great information. Thanks for taking the time to run these tests

 

[SoyCapitanSoyCapitan]

A desktop 8700K overclocked to 5GHZ gets 1650 Cinebench points for context. So laptops are not ideal for final CG rendering. Only pre-viz.

 

[shavou]

TLDR: Power limit to 45W and you'll see a 15% performance jump.

I ran ten Cinebench CPU Benchmarks in a row, giving the machine adequate time to cool down to idle between each run. No software changes were made and the CPU was idle otherwise (only running the software below & some small menu bar apps)

Software Used:
* Cinebench https://www.maxon.net/en/products/cinebench/
* Macs Fan Control https://www.crystalidea.com/macs-fan-control
* Volta http://volta.garymathews.com/

Great test and amazing results, thank you very much!! I'm waiting for my i9 to be delivered next week and this results are very promising...I was 100% sure I wanted to return it for a i7 2.6 GHz but now after seeing your test I changed my mind.
My 90% use would be photo editing, extensive batch processing in Lightroom and Photoshop with very large files (around 80MB each). Do you think I should always use Volta or just when performing the most demanding tasks?

Also, have you seen this other software VoltageShift ? Probably with it we can limit the power above the 45w volta threshold...

https://sitechprog.blogspot.com/2017/06/voltageshift.html?m=1

 

[The Mercurian]

@leman @maflynn @CodeJoy @Queen6 check out above results!

@Aea - thanks for sharing these results!

 

[Schranke]

This is great @Aea thanks for doing this.
Now the question becomes if apple will do a firmware update to change the values and set some limits...

 

[ofarlig]

Great testing, looks good that much of it can be solved with software.

 

[The Mercurian]

This is great @Aea thanks for doing this.
Now the question becomes if apple will do a firmware update to change the values and set some limits...

One would assume they should - but.....I feel like they would not like to admit the users managed thermal management better than they did.....

 

[Schranke]

One would assume they should - but.....I feel like they would not like to admit the users managed thermal management better than they did.....

I do not think apple will see it as users managing thermals better then them, but rather as a sign that they didn't test the machines enough before selling them. Apple should have found this, and they should know that their customers would too and that they will not accept it. So limiting power at different levels to see where it would perform best should have been done on a range of machines before.

 

[leman]

Great stuff! (and thanks Mercurian for tagging me).

I'd like to try it out as well, but the fact that this app is closed source is a dealbreaker for me. Does anyone know how this was achieved, maybe with some open source code?

 

[The Mercurian]

Great stuff! (and thanks Mercurian for tagging me).

I'd like to try it out as well, but the fact that this app is closed source is a dealbreaker for me. Does anyone know how this was achieved, maybe with some open source code?

Opensource would also be nice because then presumably you wold not be stuck to 45W and could try 50W for example

 

[maflynn]

I have no problems paying for an application that a developer worked hard on and given that Volta does exactly what I need, I'm happy to pay the small amount and reward the developer for his hard work.

I'm not down on open source, but there's a time and place for everything and I think in this instance the closed source application is better then not having anything

 

[Schranke]

I have no problems paying for an application that a developer worked hard on and given that Volta does exactly what I need, I'm happy to pay the small amount and reward the developer for his hard work.

Ohh I always love when people feel like that, it seems that more and more expect free application, not only on iOS but also MacOS.

 

[leman]

I have no problems paying for an application that a developer worked hard on and given that Volta does exactly what I need, I'm happy to pay the small amount and reward the developer for his hard work.

Oh, don't misunderstand me, I will gladly pay for good quality apps, but I made a habit not to use system control apps like these if I can't review the code.

 

[CodeJoy]

@leman @maflynn @CodeJoy @Queen6 check out above results!

@Aea - thanks for sharing these results!

Looks interesting, very good data collection. Can Volta only set the limit to 45W? What's encouraging is that power limit is something Apple could fix via firmware. What's potentially concerning is that there are physical limits both to power and to thermals in the current design, and neither is probably very easy to fix without a more significant rethink of the system.