Exclusive insight: Apple M1 Single "Core" comparisons miss the mark (with reference points)
Exclusive insight: Apple M1 Single "Core" comparisons miss the mark (with reference points)
I believe since Apple has proven it can do it, others will follow suit, making it simpler to develop a second wheel instead of the first. More investment and research will likely follow in the ARM area because Apple has demonstrated potential there. It feels like a win for everyone, regardless of personal preference.
I might be mistaken since Tiger Lake is a newer design, but I’m very unlikely it’s using all cores simultaneously. The i9-9880H would generate around 50 watts at 4.8GHz on one core, not full turbo across all cores. Under heavy load, all cores only reached about 3GHz, and performance per watt dropped sharply as you approached 5GHz. This matches my experience—single-core power stayed the same while multi-core speed declined significantly.
Identifying the relevant benchmarks is key. Also, focusing on four-core performance makes sense since the M1 supports eight cores with power efficiency considerations.
The efficiency cores underperform when it comes to speed. They play no significant role in boosting performance. The four tiny cores combined don’t match the output of one large core. You’ll see similar results whether you test just the big cores or all of them. Including the small ones might even hurt performance because they’ll vie for memory bandwidth with the bigger ones. In short, the M1 acts like a quad-core processor.
While using Rosetta 2 seems to reveal it as a quad-core system, this adds some logic. It might suggest that extra memory translation steps aren't available in the efficiency cores. In CB23, all eight cores are used natively, giving an MP core ratio just over five. During testing, I limited CB23 to four threads and observed performance and energy consumption on my M1 Air. The same was done with eight threads. CB23 4T Total Package Power: 14593 mW Score: 5169pts CB23 8T: Total Package Power: 15394 mW Score: 7120pts Additional details for those interested: E-Cluster Power: 1256 mW Active frequency: 2064 MHz Active usage: 100% at 600 MHz idle 972 MHz 0%, 1332 MHz 0%, 1704 MHz 0%, 2064 MHz 100%, 2064 MHz idle 0% CPU 0 Frequency: 2064 MHz Active; 1 Frequency: 2064 MHz Active; 2 Frequency: 2064 MHz Active; 3 Frequency: 2064 MHz Active; 4 Frequency: 3204 MHz Active; 5 Frequency: 3204 MHz Active; 6 Frequency: 3204 MHz Active; 7 Frequency: 3204 MHz Active cpu 0 idle 0% cpu 1 Active 100% (600 MHz: 0%, 972 MHz: 0%, 1332 MHz: 0%, 1704 MHz: 0%, 2064 MHz: 100%) cpu 1 idle 0% cpu 2 Active 100% (600 MHz: 0%, 972 MHz: 0%, 1332 MHz: 0%, 1704 MHz: 0%, 2064 MHz: 100%) cpu 2 idle 0% cpu 3 Active 100% (600 MHz: 0%, 972 MHz: 0%, 1332 MHz: 0%, 1704 MHz: 0%, 2064 MHz: 100%) cpu 3 idle 0% cpu 4 Active 100% (600 MHz: 0%, 828 MHz: 0%, 1056 MHz: 0%, 1284 MHz: 0%, 1500 MHz: 0%, 1728 MHz: 0%, 1956 MHz: 0%, 2184 MHz: 0%, 2388 MHz: 1.2%, 2592 MHz: 0%, 2772 MHz: 0%, 2988 MHz: 99%, 3096 MHz: 0%, 3144 MHz: 0%, 3204 MHz: 0%) P-Cluster Power: 14138 mW Active frequency: 2983 MHz Residency: 100% cpu 0 cpu 0 cpu 1 cpu 2 cpu 3 cpu 4 cpu 5 cpu 6 cpu 7 cpu 8 cpu 7 cpu 0 cpu 4 cpu 5 cpu 6 cpu 8 cpu 9 cpu 9 idle residency: 0.00% cpu 4 cpu 6 active residency: 100% cpu 5 active residency: 100% cpu 7 active residency: 100% cpu 3 active residency: 100% cpu 2 active residency: 100% cpu 1 active residency: 100% cpu 2 idle residency: 0.00% cpu 2 cpu 3 active residency: 100% cpu 3 active residency: 100% cpu 4 idle residency: 0.00% cpu 6 active residency: 100% cpu 7 active residency: 100% cpu 8 active residency: 100% cpu 9 idle residency: 0.00% cpu 5 idle residency: 0% cpu 8 idle residency: 0%
They likely made a mistake by introducing the M1 chip, which could have caused issues for the Mac Mini line previously.
Intel's tenth generation includes VNNI, essentially identical to previous versions. The key distinction lies in Apple's ability to mandate app developers to utilize their NPU, whereas Intel does not have this restriction.