How does the Air manage throttling and performance loss in games?!

[Diversion]

How’s the fan noise on the mbp?

Depends on the type of load.. during WoW it's literally so quiet you can't hear the fan unless you put your ear up to the back of the machine.. for like x265 encoding with Handbrake, it's audible and you can hear it in a quiet room easily but at least it's mostly a pleasant woosh sound.

 

[pento]

You have to turn of the "Target DPS" setting, as this will seemlessly change the system settings, lowing them to maintain a FPS of 60.

You can have all the settings turned up high, but if Target DPS is turned on, then the game will automatically adapt the settings to attempt to get to the desired FPS.

See a copy of your photo:

Good point! I've played around with it off and it still seemed to get a pretty stable 60 FPS, but not for any long periods of time. I'll probably play for a few hours tonight, and I'll turn that off and see what the numbers look like an hour or so in.

 

[moep]

I received my own MBA today, the 8GB/512GB base model. It's very impressive to say the least.

Regarding the topic I've got a few minor observations myself so far:

coconutBattery will display the battery power draw in watts when the device is disconnected from the charger. Power draw should reflect throttling, so we might be able to use this until better monitoring tools are available.

In my case I found that under high load (moderate CPU and full GPU) this starts out at ~25W total draw with the display at 50% and goes down to 14W as the SoC throttles down. I don't think I've let it run long enough to reach equilibrium yet.
pmset -g sysload still reported "thermal level = Great" at that point.

Based on a little bit of testing the display itself only seems to consume around 2W. The whole machine is idling at 3.5W total power draw as I write this.

Putting a large fan (140mm Noctua at 12V) near the USB-C ports and letting it blow across the keyboard/topcase, display and bottom case completely eliminated all throttling at 22°C ambient temperature.
Discharge rate in coconutBattery went from ~15W back up to ~25W within a few minutes of having the fan blow across the device during combined CPU/GPU loads. The entire device felt cold to the touch. When I say cold I mean it, think of the way a Macbook feels before you turn it on.

I think this is a neat solution for summer when stationary, because the keyboard and your hands stay completely cool, unlike with my old 15" MBP which always got a bit warm despite the HSFs.

The same tool will also display battery temperature, which gets up to 46°C as the whole assembly heats up without a fan. It remains to be seen how these temperatures will affect battery life. I wonder if battery temperatures also affect throttling. The extra fan also drops battery temperatures back down to the mid to low 30°C under full load.

Activity Monitor has a window that shows GPU load. (Command-4) This can be useful to gauge how efficient or straining certain GPU loads are.

 

[wyatterp]

@wyatterp thanks for doing this test, it's really interesting.

But to be honest, I don't understand how this is possible.
1568/1157~= 1.35
When throttling, you observed the 8 core gpu to be about 35% faster.

The difference between the two setups is one gpu core.
So naively, one could expect a difference of about 14% (8/7~=1.14)
When thermal throttling, the clockspeeds are limited by tdp.
The loss of 14% should occur, when the 7 core and 8 core gpu run at the same clockspeed.
But this doesn't really make sense to me.
Shouldn't the 7 core use less power then?
So it should be able to clock a little bit higher?

What I would have expected is this: peak performance of the 8 core gpu is 14% higher.
When throttling, both setups become slower and the difference between them decreases a little bit.
The advantage of the 8 core gpu could be something like 10% then.

Sorry, but could you maybe rerun those benchmarks?
Benchmarking is notoriously brittle. Maybe there was some os task going on in the back.

I can't rerun them as I returned the base MBA - but I have continued on and off to test the 8-core MBA I have - and it just throttles less and performs significantly better in the Uningine Valley loop over time. I was showing first and then last benchmark after 90 minutes looping for both the base MBA and the 8-core MBA I have. I understand that you would think the 7-core should run cooler and throttle less, but that's not what I'm observing. I repeated the Uningine valley test earlier today (didn't write the numbers down) but after time, it still wasn't throttling that much from the first run.

 

[wyatterp]

Goes to show the MBA does throttle.. I notice the fan does turn on under GPU oriented tasks on my MBP m1 so it's probably safe to assume that if you do want to game a bit, you might consider the MBP for it's fan and ability to prevent throttling (as much) over time.

Finding the 8-core model doesn't throttle as hard in GPU limited tasks - but it does throttle CPU similarly I think, for example, the 7 and 8 core in Cities Skylines seems pretty close on my very unscientific observation, but I don't have the base model MBA anymore to do a string of tests.

 

[wyatterp]

Another graph; same test as before, but lower settings, and using a script to take a screenshot every minute. It's less of a real-world test as there's no movement in the game but that's the trade-off.

The result this time is a 15% drop in framerate over 50-ish minutes until it levels out.

View attachment 1677418

As you might predict from the graph, limiting framerate to 60 results in a constant 60 fps for an hour.

Interesting - my uningine valley loop test for 90 minutes on the 8-core was ~13% drop. 7-core was a lot more.

 

[Wolff Weber]

Interesting - my uningine valley loop test for 90 minutes on the 8-core was ~13% drop. 7-core was a lot more.

Maybe 8-core GPU M1 is better piece of silicon than 7-core (selected, from the central part of the wafer). This means lower voltage for stable run, thus better performance.

 

[pento]

I received my base model(8-core CPU, 7-core GPU, 8GB RAM, 256GB SSD), MacBook Air(M1) yesterday, and have been playing World of Warcraft quite a bit over the past few days. The game runs great. I've played several multi-hour long sessions and I am just blown away by this computer.

With the settings shown below, I get a very consistent 60FPS everywhere in the world that I've been(I haven't done any raids, mostly solo or small group play). The bottom of the case definitely does get hot, and the area above the function keys on the top of the case, but no hotter than any other MBP or MBA I've used or owned through the years. I wouldn't want to leave the bare metal on the skin of my lap for long, because it would become uncomfortable, but not in a burning hot sort of way. It also cools down very quickly. I'm sure there was some throttling that happened, but none that I noticed. The frame rates held steady.

For some additional perspective on my idea of performance, I use both a 6-core 2018 MBP and a 12-core 2019 Mac Pro on a daily basis for work and personal uses. This thing is fast. I had originally opted for the M1 MBP with 16GB of RAM but changed my mind and went with the MBA.

If anyone is curious, I've been playing WoW for about 50 minutes(just finished a 5 person dungeon), and I'm still getting a consistent 60FPS(+/- 1 or 2 frames). Snapped a couple more screenshots of my config without the target FPS setting enabled. I've also been running on battery the entire time, started out at 100%, and am now at 78%.

 

[Mechanism]

Interesting - my uningine valley loop test for 90 minutes on the 8-core was ~13% drop. 7-core was a lot more.

That sort of % performance drop does seem to be about right, when pushed.

A repeated multicore CPU task (video conversion in ffmpeg that took about 2 minutes per iteration) also saw a 13% drop over the first 30 minutes.

Although for anyone thinking about getting the MBA, it would be wrong to think it will throttle in all scenarios, even games. I think it's a great balance between performance (it's great), heat (also great; it gets warm but not really hot) and noise (there is none).

 

[nameste]

That sort of % performance drop does seem to be about right, when pushed.

A repeated multicore CPU task (video conversion in ffmpeg that took about 2 minutes per iteration) also saw a 13% drop over the first 30 minutes.

Although for anyone thinking about getting the MBA, it would be wrong to think it will throttle in all scenarios, even games. I think it's a great balance between performance (it's great), heat (also great; it gets warm but not really hot) and noise (there is none).

Thanks for the update I am really excited to use a laptop without any noise

 

[jescolano]

If anyone is curious, I've been playing WoW for about 50 minutes(just finished a 5 person dungeon), and I'm still getting a consistent 60FPS(+/- 1 or 2 frames). Snapped a couple more screenshots of my config without the target FPS setting enabled. I've also been running on battery the entire time, started out at 100%, and am now at 78%.

Hi

I was very interested in your comments about WoW on the MBA 7 core. Just wanted to know if you've kept playing WoW for an extra period of time with the latest setting you mentioned, and if so, could you please let me know if you still keep getting around 60 FPS? Was the computer hot or more than you expected/experienced before?

I'm about to buy the new MBA and I would appreciate your input
Thanks

 

[pento]

Hi

I was very interested in your comments about WoW on the MBA 7 core. Just wanted to know if you've kept playing WoW for an extra period of time with the latest setting you mentioned, and if so, could you please let me know if you still keep getting around 60 FPS? Was the computer hot or more than you expected/experienced before?

I'm about to buy the new MBA and I would appreciate your input
Thanks

I played another dungeon tonight, with the same settings I mentioned in my last post, and further into the dungeon the FPS did dip to the 40's at times during larger fights. I was playing on battery, FWIW. Felt very playable to me but YMMV!

 

[Serban55]

If anyone is curious, I've been playing WoW for about 50 minutes(just finished a 5 person dungeon), and I'm still getting a consistent 60FPS(+/- 1 or 2 frames). Snapped a couple more screenshots of my config without the target FPS setting enabled. I've also been running on battery the entire time, started out at 100%, and am now at 78%.

for 50 min of WoW from 100% to 78%....comparing the Intel one that for 1 hour from 100% you are down to 44-46%

 

[MacMike081]

Has anyone tried shadow gaming? I guess it should be fine.

 

[wyatterp]

Enough reviews coming out that I think largely everyone was wrong about the MBA - it throttles, but only to a minor degree. Somewhere around a 10-15% maximum drop in CPU throughput, but it hold steady once it hits a certain point and DOES not continue to ramp down over time. It's absolutely incredible. I've been playing plenty of games on mine, and it holds up just fine over time. Really incredible to see a fanless device maintain this kind of performance!