4 ms·

Toe to toe? It starches every Ryzen chip in single core performance, and every mobile Ryzen under 45 watts TDP in multicore despite only having 4 performance co

Toe to toe? It starches every Ryzen chip in single core performance, and every mobile Ryzen under 45 watts TDP in multicore despite only having 4 performance cores and 15 Watts TDP.

adrian_b 6y ago
You are correct only about Ryzens based on the old Zen 2. Any new Ryzen 5xxx based on Zen 3 is faster in single-thread than the best Apple M1. In general-purpose benchmarks, like Geekbench 5, the maximum Apple M1 score is 1752, while Ryzen 5xxx score between 1800 and 1876. In computational benchmarks, where the IPC is limited by some execution resources reaching 100% utilization, so the higher average IPC of Apple M1 does matter much less, Apple M1 is even slower compared to Ryzens. For example Ryzen 7 5800X is 22% faster than Apple M1 in gmpbench (the benchmark was run for both CPUs after tuning specific to Apple M1 and tuning specific to Zen 3 were added to libgmp, so it is a comparison of apples with apples).
- valuearb 6y ago
  https://browser.geekbench.com/processors/amd-ryzen-7-5800x https://browser.geekbench.com/processors/amd-ryzen-7-5800x The 5800x is a 105 watt part(!!!), still gets starched by the M1 in single core, and is only about 40% faster in multicore despite using all that power to drive double the performance core count. But by all means, go off to cherry pick more niche benchmarks...
  - adrian_b 6y ago
    I did not cherry pick any benchmarks, I have picked the only 2 benchmarks that have been published for both Apple M1 and for Zen 3, and that were also freely available, so I could also run them on my own computer, to verify that the published benchmark results are correct. They are not niche benchmarks. Geekbench 5 covers many common application domains and its only flaw is that the GB5 tests are very short so they finish before the CPU overheats and slows down, so this benchmark is much more favorable for Apple M1 than for a desktop CPU. Even if GB5 is advantageous for M1, it still shows that Zen 3 is faster. gmpbench tests a fundamental feature of any CPU, its ability to compute operations with large numbers. That may not matter for you but it is definitely one of the important criteria for measuring the single-thread speed of a processor, so when Apple M1 scores 6422 and a Ryzen 7 5800X (not the fastest Ryzen) scores 7816, there is no doubt about which starches which. Moreover, the multi-thread results that I have seen for Geekbench 5, which very strongly favors Apple M1 (because in a real application M1 would overheat soon and it would slow down) show a 60% higher speed for 5800X, not of 40%, e.g. 12339 for 5800X vs. 7716 for Apple M1. If you want to live with the illusion that Apple M1 is faster in single-thread execution than Zen 3, even if all the real numbers from benchmark results contradict you, that is your privilege.
  - xirbeosbwo1234 6y ago
    https://browser.geekbench.com/v5/cpu/5806255 https://browser.geekbench.com/v5/cpu/5806255 The 5950X has double the core count of the 5800X but only manages a 1.6x improvement in multithreaded performance. https://browser.geekbench.com/processors/amd-ryzen-threadripper-3990x https://browser.geekbench.com/processors/amd-ryzen-threadrip... The 3990X is the most powerful desktop part on the market. It gets merely a 1.4x speedup over the 5950X despite having four times the core count. (Granted, the 5950X core is somewhat better). Geekbench isn't terribly useful at the best of times. It's particularly bad when comparing across architectures and operating systems. And, as should be clear, its multithreaded test does not scale linearly with core count. You should never take the ratio between two scores to mean the ratio between performance, especially when comparing multithreaded scores with different core counts.
mhh__ 6y ago
What does "starches" mean?
xirbeosbwo1234 6y ago
>Toe to toe? It starches every Ryzen chip in single core performance It most certainly does not. It trades blows, winning on some benchmarks and losing others. It seems a bit ahead on floating-point workloads and a bit behind on integer workloads. https://www.anandtech.com/show/16252/mac-mini-apple-m1-tested/4 https://www.anandtech.com/show/16252/mac-mini-apple-m1-teste... It does simply outperform every Intel design. That's great, but not enough to make it the fastest core in the world. It, to no one's surprise, falls behind the parts with higher core counts. Caveats: 1. These benchmarks are preliminary. We are, by necessity, comparing across operating systems and architectures. 2. Comparing single-threaded performance between a SMT system and one without SMT is a little misleading, since a non-SMT core is worse than an SMT core even if the single-threaded performance is similar. 3. I have no clue how to handle the big.LITTLE architecture. I've never used a machine with heterogeneous cores and have never written any software for it. It is certainly a benefit to efficiency. It probably improves performance on highly-threaded workloads, but I am not sure. I find myself annoyed by these discussions. I love new hardware! I find this chip very impressive! It is the best chip in the world in performance-per-watt! It is clearly the best mobile chip on the market! It has a lot of interesting accelerators! Please stop forcing me to be the naysayer.
- valuearb 6y ago
  M1 is either first or very close to first in all the Anandtech benchmarks against CPUs using up to seven times more power.
  - adrian_b 6y ago
    Apple M1 is definitely faster than any other CPU whose power consumption is reduced at the same value. At identical power consumptions, an Apple M1 core is faster by about 30% to 40% than an Intel Tiger Lake or an AMD Zen 3. All older Intel/AMD CPUs are even much slower. Nevertheless, the power consumptions of Intel and AMD CPUs can be raised to much higher levels and above 20 W per core, which is probably a little more than 3 times what an Apple M1 core consumes, they match the Apple M1 performance and at a higher power (i.e. in the current desktop Zen 3 and in the soon to be introduced Intel Rocket Lake and Intel Tiger Lake H) they surpass the speed of Apple M1. So in single-thread execution, AMD Zen 3 is already faster and the Intel CPUs that are expected for end of March will also be faster, but only at much higher power consumptions (3 to 4 times more per core in single-thread). Nonetheless, in desktops and large laptops those power consumptions do not matter.

Toe to toe? It starches every Ryzen chip in single core performance, and every mobile Ryzen under 45 watts TDP in multicore despite only having 4 performance cores and 15 Watts TDP.

adrian_b 6y ago

You are correct only about Ryzens based on the old Zen 2. Any new Ryzen 5xxx based on Zen 3 is faster in single-thread than the best Apple M1. In general-purpose benchmarks, like Geekbench 5, the maximum Apple M1 score is 1752, while Ryzen 5xxx score between 1800 and 1876. In computational benchmarks, where the IPC is limited by some execution resources reaching 100% utilization, so the higher average IPC of Apple M1 does matter much less, Apple M1 is even slower compared to Ryzens. For example Ryzen 7 5800X is 22% faster than Apple M1 in gmpbench (the benchmark was run for both CPUs after tuning specific to Apple M1 and tuning specific to Zen 3 were added to libgmp, so it is a comparison of apples with apples).

valuearb 6y ago
https://browser.geekbench.com/processors/amd-ryzen-7-5800x https://browser.geekbench.com/processors/amd-ryzen-7-5800x The 5800x is a 105 watt part(!!!), still gets starched by the M1 in single core, and is only about 40% faster in multicore despite using all that power to drive double the performance core count. But by all means, go off to cherry pick more niche benchmarks...
- adrian_b 6y ago
  I did not cherry pick any benchmarks, I have picked the only 2 benchmarks that have been published for both Apple M1 and for Zen 3, and that were also freely available, so I could also run them on my own computer, to verify that the published benchmark results are correct. They are not niche benchmarks. Geekbench 5 covers many common application domains and its only flaw is that the GB5 tests are very short so they finish before the CPU overheats and slows down, so this benchmark is much more favorable for Apple M1 than for a desktop CPU. Even if GB5 is advantageous for M1, it still shows that Zen 3 is faster. gmpbench tests a fundamental feature of any CPU, its ability to compute operations with large numbers. That may not matter for you but it is definitely one of the important criteria for measuring the single-thread speed of a processor, so when Apple M1 scores 6422 and a Ryzen 7 5800X (not the fastest Ryzen) scores 7816, there is no doubt about which starches which. Moreover, the multi-thread results that I have seen for Geekbench 5, which very strongly favors Apple M1 (because in a real application M1 would overheat soon and it would slow down) show a 60% higher speed for 5800X, not of 40%, e.g. 12339 for 5800X vs. 7716 for Apple M1. If you want to live with the illusion that Apple M1 is faster in single-thread execution than Zen 3, even if all the real numbers from benchmark results contradict you, that is your privilege.
- xirbeosbwo1234 6y ago
  https://browser.geekbench.com/v5/cpu/5806255 https://browser.geekbench.com/v5/cpu/5806255 The 5950X has double the core count of the 5800X but only manages a 1.6x improvement in multithreaded performance. https://browser.geekbench.com/processors/amd-ryzen-threadripper-3990x https://browser.geekbench.com/processors/amd-ryzen-threadrip... The 3990X is the most powerful desktop part on the market. It gets merely a 1.4x speedup over the 5950X despite having four times the core count. (Granted, the 5950X core is somewhat better). Geekbench isn't terribly useful at the best of times. It's particularly bad when comparing across architectures and operating systems. And, as should be clear, its multithreaded test does not scale linearly with core count. You should never take the ratio between two scores to mean the ratio between performance, especially when comparing multithreaded scores with different core counts.

mhh__ 6y ago

What does "starches" mean?

xirbeosbwo1234 6y ago

>Toe to toe? It starches every Ryzen chip in single core performance It most certainly does not. It trades blows, winning on some benchmarks and losing others. It seems a bit ahead on floating-point workloads and a bit behind on integer workloads. https://www.anandtech.com/show/16252/mac-mini-apple-m1-tested/4 https://www.anandtech.com/show/16252/mac-mini-apple-m1-teste... It does simply outperform every Intel design. That's great, but not enough to make it the fastest core in the world. It, to no one's surprise, falls behind the parts with higher core counts. Caveats: 1. These benchmarks are preliminary. We are, by necessity, comparing across operating systems and architectures. 2. Comparing single-threaded performance between a SMT system and one without SMT is a little misleading, since a non-SMT core is worse than an SMT core even if the single-threaded performance is similar. 3. I have no clue how to handle the big.LITTLE architecture. I've never used a machine with heterogeneous cores and have never written any software for it. It is certainly a benefit to efficiency. It probably improves performance on highly-threaded workloads, but I am not sure. I find myself annoyed by these discussions. I love new hardware! I find this chip very impressive! It is the best chip in the world in performance-per-watt! It is clearly the best mobile chip on the market! It has a lot of interesting accelerators! Please stop forcing me to be the naysayer.

valuearb 6y ago
M1 is either first or very close to first in all the Anandtech benchmarks against CPUs using up to seven times more power.
- adrian_b 6y ago
  Apple M1 is definitely faster than any other CPU whose power consumption is reduced at the same value. At identical power consumptions, an Apple M1 core is faster by about 30% to 40% than an Intel Tiger Lake or an AMD Zen 3. All older Intel/AMD CPUs are even much slower. Nevertheless, the power consumptions of Intel and AMD CPUs can be raised to much higher levels and above 20 W per core, which is probably a little more than 3 times what an Apple M1 core consumes, they match the Apple M1 performance and at a higher power (i.e. in the current desktop Zen 3 and in the soon to be introduced Intel Rocket Lake and Intel Tiger Lake H) they surpass the speed of Apple M1. So in single-thread execution, AMD Zen 3 is already faster and the Intel CPUs that are expected for end of March will also be faster, but only at much higher power consumptions (3 to 4 times more per core in single-thread). Nonetheless, in desktops and large laptops those power consumptions do not matter.