How am I supposed to interpret this? A MacBook Air surpasses my i7-8700k in single and (almost) multi core performance?
replies(1):
Intel is now #3
Check out for example the per core power charts that Anandtech does: https://www.anandtech.com/show/16214/amd-zen-3-ryzen-deep-di...
Compare for example the 1 core power numbers between the chips. The 5600X 1 core result is 11w @ 4.6ghz, whereas the other two chips boost higher and hit 4.8-4.9ghz 1 core turbos, but it costs 17-18w to do it. Huge increase in power for that last 1-2% performance. So you really can't or shouldn't compare more power-concious configurations with the top end desktop where power is infinite and well worth spending for even single digit percentage gains.
And then of course you should also note that the single-core power draw in all of those is vastly lower than their TDP numbers (65w for the 5600x, and 125w for the 5800x/5900x).
The reason that graph doesn't include the A14 Firestorm -> M1 jump was simply timing. We know the thermal envelopes of the M1 and the cooling designs. We now have clock info thanks to GB5. So yes, the data is pretty solid. No one's saying that the iPhone beats the Mac (or a PC) at performance when you consider the whole system. Just that the CPU architecture can and will deliver higher performance given the M1 clock, thermals and cooling. Remember that The A14/M1 CPUs are faster at lower clock speeds.
That's why A14 only runs at 1.8ghz base, 3ghz boost. That's how it has low power consumption. And similarly Intel pushing 5ghz is why it has high power consumption.
TSMC's 5nm will have a raw transistor performance/watt advantage, but it's not huge
it's unfortunately drowned out by the cpu throttling scandal on google, but, its well-known in ar dev (and if you get to talk to an apple engineer away from stage lights at wwdc) that you have to proactively choose to tune performance, or you'll get killed after a minute or two due to thermal throttling.
Yeah comparing TDP is meaningless even within the same processor. The 4 core workload in this table uses 94W and the 16 core workload uses 98W. There is also an anomaly at 5 cores where the CPU uses less power than if it only used 4 cores.
If you tried to derive conclusions about the power efficiency of the CPU you would end up making statements like "This CPU is 3-4 times more power efficient than itself"
Imagine going on a hike and climbing an exponential slope like 2^x. You go up to 2^4 and then go down again and repeat this three times so you have hiked 12km (43) in total. Then there is a athlete who is going up to 2^8. He says he has hiked 8km and you laugh at him because of how sweaty he is despite having walked a shorter distance than you. In reality 32^4 (48) is nowhere near 2^8 (256). The athlete put in a lot more effort than you.
