AMD Zen 3 outperforms on workloads that can take advantage of having more cores, but for single threaded workloads, it leads in some SPEC subtests and falls behind in others.

https://www.anandtech.com/show/16252/mac-mini-apple-m1-teste...

> AMD still generally outperforms Apple Silicon despite larger lithographies

The single-core benchmarks I saw put the M1 against CPUs with amd64 cores running SMT (which is fair - SMT extracts about 20% of extra parallelism of an amd64 workload, something the M1 does using a humongous reorder buffer on a single thread) with the AMD (and latest Intel) parts showing a slight advantage.

They do so, however, at a much higher TDP which, if my own computers (all but the server under the desk) are any indication, will cause the x86's to throttle down in a couple seconds while the M1 will continue at full steam.

They have a much higher TDP because you are comparing PC parts that are not optimized for power consumption. Their uncore is made using power-inefficient lithographies to save on cost and contribute a ton to power usage.

They also have a lot more cores.

Per high-performance core, M1 macs are tied or less efficient depending on the workload.

> Per high-performance core, M1 macs are tied or less efficient depending on the workload.

That's not what the benchmarks I see say. AMD's chips with 8 amd64 cores beat the M1, which has 4 big cores and 4 low-power ones. In some benchmarks, the x86 laptop CPUs record higher performance, but at a higher clock and with a higher TDP. I don't have the IPC numbers for either, but with a lot of execution ports, its huge reorder buffer and the less strict memory consistency model, I suspect the IPC to be higher on the M1.

The low performance cores of the M1 go anywhere from useless to actively harmful in critical multi thread workloads.

When you are doing serious work, the M1 is really a four core CPU and nothing else.

M1 cores have higher IPC, yes. They are also bigger and slower. Larger reorder buffers and more flexible memory models mean bigger cores which mean slower clock speeds at the same power consumption.

The M1 also doesn't have SMT, which is an easy +20% IPC.

If you measure the performance per core and the power consumption per core, and of course I'm talking only high performance cores, then last gen x86 laptop CPUs are right there despite higher performance.