First M2 benchmarks

[JPack]

It looks like the fans did not kick in until the CPU core temps reached 100°C and with the fans at around half RPM, temps peaked at around 104°C.

My guess is that the Air would allow temps in the 100°C range before it started reducing wattage to keep it around that temp. And even at 100°C, it was still running over 3GHz so we will likely not see severe throttling (where we lose a GHz or so).

I would say it's unlikely because of the difference in VRM components between MBA and MBP. That is one of the reasons why one has a larger logic board.

Keeping temps at 100C at 3GHz requires more power than 80C at 3GHz. Basically, the characteristics we see on MBP can't be directly carried to MBA.

The MBA is gated by both power delivery and temperature factors.

 

[dgdosen]

Such speculation. We'll know the whole story in a few weeks... My gut says throttling, temps and battery life won't be an issue on the M2 Air - even when compared to the M1 Pro/Air differences.

I don't think cats are going to be sitting on them.

 

[jav6454]

Such speculation. We'll know the whole story in a few weeks... My gut says throttling, temps and battery life won't be an issue on the M2 Air - even when compared to the M1 Pro/Air differences.

I don't think cats are going to be sitting on them.

How is this speculation when there is video evidence of a M2 heating up? MBAs are passively cooled, likely they'll throttle hard when heat soaked.

 

[DoFoT9]

I mean, we're going to have throttling regardless of the enclosure the devices are in - that's how they are designed. Do we know yet how larger the M2 die is compared to the M1? Whist there are thermal efficiencies with M2, they can't be so good as to remove the main bottleneck (heat & power requirements). Or at least, that's what I would assume...

 

[jav6454]

I mean, we're going to have throttling regardless of the enclosure the devices are in - that's how they are designed. Do we know yet how larger the M2 die is compared to the M1? Whist there are thermal efficiencies with M2, they can't be so good as to remove the main bottleneck (heat & power requirements). Or at least, that's what I would assume...

In transistor count I know it's ~20% bigger. As per power wise, it has been confirmed it consumes more power when reaching peak performance vs M1.

 

[DoFoT9]

In transistor count I know it's ~20% bigger. As per power wise, it has been confirmed it consumes more power when reaching peak performance vs M1.

Ah so 85W on M1 to 100W on M2 (roughly random numbers picked out of my head under load)? Such a power hog

I am waiting for the longer term cost implications (i.e. electricity) and comparisons to come out. Whilst the M1 Ultra wasn't really close to say an i9+3090 RTX in most things apart from ProRes encoding, it uses significantly less power. What does that equal in the long term?

 

[jav6454]

Ah so 85W on M1 to 100W on M2 (roughly random numbers picked out of my head under load)? Such a power hog

I am waiting for the longer term cost implications (i.e. electricity) and comparisons to come out. Whilst the M1 Ultra wasn't really close to say an i9+3090 RTX in most things apart from ProRes encoding, it uses significantly less power. What does that equal in the long term?

Something like that. In the end the power gaps between both at peak are 0W when at peak CPU performance, but at the lower end there is a 1W gap with the M2 consuming more.

In the GPU front, there is a ~4W gap at peak with the M2 being the high consumer. At lower end there is a ~1.5W, again with M2 consuming more.

How does this translate? Well, given the recent video reviews, it seems those minute power draws are creating hot spots on the fan cooled MBP. Albeit that can be attributed to the fan curve being a disaster. But in a finless operation? Translates to quicker throttling.

As per the Ultra... Apple just exaggerated their claims and likely stuck to their ProRes or ProRaw performance.

 

[DoFoT9]

Something like that. In the end the power gaps between both at peak are 0W when at peak CPU performance, but at the lower end there is a 1W gap with the M2 consuming more.

In the GPU front, there is a ~4W gap at peak with the M2 being the high consumer. At lower end there is a ~1.5W, again with M2 consuming more.

How does this translate? Well, given the recent video reviews, it seems those minute power draws are creating hot spots on the fan cooled MBP. Albeit that can be attributed to the fan curve being a disaster. But in a finless operation? Translates to quicker throttling.

As per the Ultra... Apple just exaggerated their claims and likely stuck to their ProRes or ProRaw performance.

Awesome info - thanks!

I recall a recent video on the Ultura where the fans did not go above ~3,500 RPM under load, where they were able to do ~6,000 RPM. If this were a conversation regarding server-grade platforms, it would be in shambles. Given we are discussing consumer devices, that have batteries and demand performance vs convenience (battery life, heat of the chassis, etc), I'm fairly okay with it overall - especially those without them whatsoever. They are clearly aimed at a different market (including myself!).

I mean, remember the days of Intel chips where they would throttle after tens of seconds? Don't the M1's not do that until a few minutes, if not longer, now?

DoFo

 

[jav6454]

Awesome info - thanks!

I recall a recent video on the Ultura where the fans did not go above ~3,500 RPM under load, where they were able to do ~6,000 RPM. If this were a conversation regarding server-grade platforms, it would be in shambles. Given we are discussing consumer devices, that have batteries and demand performance vs convenience (battery life, heat of the chassis, etc), I'm fairly okay with it overall - especially those without them whatsoever. They are clearly aimed at a different market (including myself!).

I mean, remember the days of Intel chips where they would throttle after tens of seconds? Don't the M1's not do that until a few minutes, if not longer, now?

DoFo

Quick graphs:

M1 on 13" MBPs do not throttle at all under heavy loads; on the M1 MBA it did after 1-2 minutes.

 

[JimmyjamesEU]

Something like that. In the end the power gaps between both at peak are 0W when at peak CPU performance, but at the lower end there is a 1W gap with the M2 consuming more.

In the GPU front, there is a ~4W gap at peak with the M2 being the high consumer. At lower end there is a ~1.5W, again with M2 consuming more.

How does this translate? Well, given the recent video reviews, it seems those minute power draws are creating hot spots on the fan cooled MBP. Albeit that can be attributed to the fan curve being a disaster. But in a finless operation? Translates to quicker throttling.

As per the Ultra... Apple just exaggerated their claims and likely stuck to their ProRes or ProRaw performance.

You keep saying throttling. Where Is it occurring? Max tech’s video didn’t show that.

 

[jav6454]

You keep saying throttling. Where Is it occurring? Max tech’s video didn’t show that.

For MBAs. For MBPs it won't happen, but it did sorta become capped and was even acknowledged in the video.

 

[JimmyjamesEU]

For MBAs. For MBPs it won't happen, but it did sorta become capped and was even acknowledged in the video.

I didn’t see that. Performance stayed level regardless of fan speed.

 

[jav6454]

I didn’t see that. Performance stayed level regardless of fan speed.

There are two videos on the same subject, probably you watched pt 1 or 2.

 

[DoFoT9]

Quick graphs:

M1 on 13" MBPs do not throttle at all under heavy loads; on the M1 MBA it did after 1-2 minutes.

Fair point - so there is overhead (or right on the limit) for the M1 13" MBP in terms of TDP. We can assume that to the be the same for the M2 in the same chassis?

For reference, the M1 does throttle down at 4% battery (For obvious reasons)

 

[JimmyjamesEU]

There are two videos on the same subject, probably you watched pt 1 or 2.

By Max tech?

 

[jav6454]

By Max tech?

Yup, I'll link them:

Thermals Video

M1 v M2 Video

 

[jav6454]

Fair point - so there is overhead (or right on the limit) for the M1 13" MBP in terms of TDP. We can assume that to the be the same for the M2 in the same chassis?

For reference, the M1 does throttle down at 4% battery (For obvious reasons)

In theory, yes there is overhead, but testing as showed that for some reason there is a micro-throttle in the MBP. Unsure why.

Unsure about the second part.

 

[JimmyjamesEU]

Yup, I'll link them:

Thermals Video

M1 v M2 Video

I’ve watched both of those. I don’t recall any examples of throttling.

 

[DoFoT9]

In theory, yes there is overhead, but testing as showed that for some reason there is a micro-throttle in the MBP. Unsure why.

Unsure about the second part.

Depends how much you trust our fellow posters.

 

[jav6454]

I’ve watched both of those. I don’t recall any examples of throttling.

On the Thermal video @ minute 11:49

 

[jav6454]

Depends how much you trust our fellow posters.

I mean, it would make sense as MacOS would try to prevent the an unintended shutdown due to power loss. At such low capacity, the battery could likely not respond to the demands in power.

I haven't seen any articles confirming it, but it wouldn't be farfetched to think of it as true.

 

[JimmyjamesEU]

On the Thermal video @ minute 11:49

Uhh…that shows Max stating it throttles but the chart contradicts that statement. More a less a straight line with a little run to run variation. Can’t see that qualifies as throttling, rather Max‘s lack of understanding.

 

[jav6454]

Uhh…that shows Max stating it throttles but the chart contradicts that statement. More a less a straight line with a little run to run variation. Can’t see that qualifies as throttling, rather Max‘s lack of understanding.

Well, I'd suggest you write to him asking for clarification as from his viewpoint, the heat caused throttling. I can concur with him given the GHz limitations from practical/tested vs advertised.

 

[JimmyjamesEU]

Well, I'd suggest you write to him asking for clarification as from his viewpoint, the heat caused throttling. I can concur with him given the GHz limitations from practical/tested vs advertised.

I’m not sure what you think throttling is, but I understand it as a drop in performance due to running at a certain temperature for an extended period. Max’s video shows a rise in temperature but no drop in performance.

 

[jav6454]

I’m not sure what you think throttling is, but I understand it as a drop in performance due to running at a certain temperature for an extended period. Max’s video shows a rise in temperature but no drop in performance.

The M2 is rated for 3.5 (3.49) GHz. The tests showed that it ran only up to 3.3GHz consistently. For me the missing 200MHz is a throttle.

Due to what? I would say temperature due to seeing 90+ there.