The nominal figures for GPUs tend to be higher than what you see in practice. For example, the 6700 XT in my gaming PC is nominally a 230 W GPU, but I haven't seen it go above ~180 W in any game. If the measured peak for M1 Max GPU is ~45 W, the theoretical maximum is probably somewhere in the 50-60 W range.
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Anandtech: M1 Max multithread performance is like Ryzen 5950x
- Thread starter senttoschool
- Start date
- Sort by reaction score
?I just added the CPU to itNo harm in being generous. I rather overestimate the power consumption than underestimate it, or the bean counters will haunt me to the end of my days.
Thing is, in the inevitable comparisons in the upcoming months I can swear that we will, repeatedly, see the power draw of dGPUs be compared to whole system powerdraws of the Macbooks. I took the opportunity to put Ryans observation in these forums.
That's the package size. The die size itself is much more moderate 698mm2. I am sure that M1 Max package is much larger than 4000mm2, those RAM modules are ginormous.
That's (approx) what he said. 76.12 x 56.2 = 4,424 mm^2 . That's the area in mm^2wait...the Intel W 3175X is 4 metres ^2 ??
i think you meant
Dimension 76.16 mm × 56.6 mm
Not far off 4,294 mm^2.
It's mm^2 not m^2
At least someone understands arithmetic a 10-year-old should know.... unlike several people it appears
Yeah, but there you go - you claim that your 6700xt actually draws less while gaming and present that number (For what it’s worth, since I use a similar solution in my gaming PC, it may be that you are using the ”GPU power” number that only looks at the GPU chip in isolation, ignoring the power circuitry and memory contributions), while taking the peak number Anandtech saw across the entirety of their M1Max testing and slapping another 30+% on top of it based on…?
That’s simply biased.
There’s a scarcity of data at the moment, but a nice thing with Anandtech data is that they could do some subsystem power tracking, as opposed to wall power. That’s important, since efficiency is the true story here. (I hope both AMD and intel get a bit more ambitious with their APUs rather than protecting their CPU and GPU business. After all, since Nvidia is the big fish in the dGPU market, cannibalized sales would mostly be from them.)
That makes much more sense. 700-800mm^2 is about the limit of die reticles these days I understand.
Correct. The reticle limit with the steppers used for these processes is 26 mm by 33 mm or 858 mm2.
Apple could conceivably double up the Max on a monolithic part, but I find that proposition doubtful. Then again, I never thought they would go 512 bits LPDDR5, so what do I know! ?
Well of course, that's how quick charging works. This is a very silly way of measuring things, unless the battery is 100% charged and one can show that the computer is only drawing power from the power adapter. Macs come with an array of diagnostic tools that will accurately show package power consumption.
i always love to know the maximum,overall package battery included fast charge and everything...in lower power mode it has max of 120W, and it shows how efficient Apple is...
The 5k fully loaded imac is around 224W
I meant that the TDP figures for GPUs tend to be conservative, because GPUs are often constrained by the power they can draw from power connectors. Giving low figures that could be exceeded under some circumstances is not a good idea. My ~180 W number comes from a ~2.4 GHz clock rate, and I believe it's a halfway number between the power consumption of the chip and the entire card. Benchmarks using special hardware for measuring the power delivered to the card report 210-220 W with >2.5 GHz clock rates.
For an apples-to-apples comparison, the numbers for M1 Max should include memory controller, RAM, and Thunderbolt controller, or at least some of them. And the ~45 W figure Anandtech reports is not the peak number but the number from the only meaningful GPU workload they tested.
For a consumer CPU, not a workstation-type one (with 10+ cores). And definitely not for a desktop-level GPU, which starts at ~150W and can easily go up to 300W for the GPU alone.
FWIW, my i9-9940X, albeit a few years older, 14nm+, overclocked, can suck ~600W!!! It tripped the over-voltage protection on my 850W PSU!! I had to upgrade it to a 1200W to be able to even run the GPU.
Of course these are synthetic benchmarks, but <100W for a CPU+GPU at full tilt is hardly high end laptop territory, let alone "desktop territory"… ?
I understand most people don't know this, but TDP does NOT mean "max wattage at max performance" by any means! Again, the TDP for my 9940X is 165W. I can pass that EASILY, without (technically) overclocking it, just by setting my BIOS to an "XMP profile", just to run the memory at their own rated clocks (which, technically, is overclocking by Intel's standards and does void your warranty FWIW).
Those <100W numbers for the M1 Max at full tilt, CPU **and** GPU, are observed results, not spec sheet numbers.
And that, my friend, is impressive!
Last edited:
Congratulations: you just leveled up your "Pointless Sarcasm" stat to 69,420, a new personal record, and a coveted achievement among your 4chan peers. Enjoy your prize.
Our big boss is an Apple fan. He has asked for the maxed-out configuration. If you guys wonder what does he do?
He reads important e-mails, occasionally asks silly questions (sorry, there are no silly questions just silly people asking questions); however, loves technology. To be fair we like him.
He reads important e-mails, occasionally asks silly questions (sorry, there are no silly questions just silly people asking questions); however, loves technology. To be fair we like him.
I rather have that boss than someone who thinks a low end HP is sufficient for all work processes! Let him have his fun and then ask for something similar for workprocesses that really needs it.
In the early days of PCs, the cases were large so the big bosses could brag with a large PC computer. The larger the case, the more important was the person.
In our workplace that is involved in supply chain the managers get the $700 laptops and the rank & file gets the hand me downs that are less than 1 decade old.
For preventitive maintenance on the 5th year we increase memory to 8GB & replace the HDD wit ha 512GB SSD.
Once the PC is 10yo we liquidate.
