3 ms·

Surely there's got to be something like that available already. I'm mostly dealing with low powered M0/M4 boards, but the selection in this range is massive.

jandrese 8y ago
There's an absolute ocean of old obsolete ARM cores clocked slowly. The difference in performance is so vast it would be hard to compare with an A11 chip, especially once you include graphics acceleration in the mix. Apple has been doubling down every year on the "we can make it go faster", and the rest of the ARM chip makers have responded with a "meh". It's so bad that people get really excited for a Samsung chip that is no slower than a two generation old A series chip.
- svantana 8y ago
  Performance is highly dependent on task and graphics is extra tricky to compare, but looking at geekbench, it seems the Samsung S9 is about halfway between iPhone 7 and 8/X. So half a generation behind is more accurate.
deaddodo 8y ago
Well, the price is wrong, but nvidia's Jetson platform is very powerful compared to the *board SBCs: https://www.phoronix.com/scan.php?page=article&item=march-2017-arm&num=2 https://www.phoronix.com/scan.php?page=article&item=march-20...
- pizza234 8y ago
  Jetson CPU performance is on par with chips like the RK3399 or the Exynos5422, which are 4/5+ times as fast as the RPi 3b (I owned RPi 2/3b series, and the Exynos). Note that the first benchmark in the mentioned article is an outlier.
  - deaddodo 8y ago
    No, it's not. Having worked with the Jetson platform extensively, the TK1 was equivalent to a 5422 (because they're both built on the vanilla A15 architecture). However, the TX1 and TX2 ("Denver" cores) are closer to (though, still slighty behind) the current SD 835: https://www.youtube.com/watch?v=l6y4atVl-mc https://www.youtube.com/watch?v=l6y4atVl-mc And when it comes to GPU compute, it's much more powerful than all the alternatives.
    - pizza234 8y ago
      I'll correct to "comparable", as it's more appropriate. However, it's not a "clearcut no" or "very powerful in comparison", due to the number of cores (no doubt about the GPU, I had made this explicit in the parent post). I've run the Phoronix test suite (which uses a set of real-world applications)[⁰], and, excluding the outliars for both parts (Redis/OpenSSL), we're talking about 10 to 40% advantage for the TX1, and little more for the TX2; XU4 has even the edge in one case. They're definitely faster for the majority of use cases, however, I wouldn't classify them as "very powerful in comparison", at least, when considering the performance of an RPi 3B. [⁰] https://openbenchmarking.org/result/1805276-FO-1703199RI52 https://openbenchmarking.org/result/1805276-FO-1703199RI52
      - deaddodo 8y ago
        The TX2 is a dual-core chip (ignoring the 4 "littles" since they're unused in benchmarks). The 5422 is a quad core. And even then, it's outperforming the 5422 by a decent margin in most tests. Even c-ray, an intrinsically multithreaded application is close. So you're correct, they're "comparable" in that if you take twice as many 40%-as-powerful cores, you might shrink the gap. But for any fundamentally single threaded tasks, a massive gap exists. And if/when you leverage the GPU (which can't be ignored, despite your handwaving), it becomes a canyon.

jandrese 8y ago

There's an absolute ocean of old obsolete ARM cores clocked slowly. The difference in performance is so vast it would be hard to compare with an A11 chip, especially once you include graphics acceleration in the mix. Apple has been doubling down every year on the "we can make it go faster", and the rest of the ARM chip makers have responded with a "meh". It's so bad that people get really excited for a Samsung chip that is no slower than a two generation old A series chip.

svantana 8y ago
Performance is highly dependent on task and graphics is extra tricky to compare, but looking at geekbench, it seems the Samsung S9 is about halfway between iPhone 7 and 8/X. So half a generation behind is more accurate.

deaddodo 8y ago

Well, the price is wrong, but nvidia's Jetson platform is very powerful compared to the *board SBCs: https://www.phoronix.com/scan.php?page=article&item=march-2017-arm&num=2 https://www.phoronix.com/scan.php?page=article&item=march-20...

pizza234 8y ago
Jetson CPU performance is on par with chips like the RK3399 or the Exynos5422, which are 4/5+ times as fast as the RPi 3b (I owned RPi 2/3b series, and the Exynos). Note that the first benchmark in the mentioned article is an outlier.
- deaddodo 8y ago
  No, it's not. Having worked with the Jetson platform extensively, the TK1 was equivalent to a 5422 (because they're both built on the vanilla A15 architecture). However, the TX1 and TX2 ("Denver" cores) are closer to (though, still slighty behind) the current SD 835: https://www.youtube.com/watch?v=l6y4atVl-mc https://www.youtube.com/watch?v=l6y4atVl-mc And when it comes to GPU compute, it's much more powerful than all the alternatives.
  - pizza234 8y ago
    I'll correct to "comparable", as it's more appropriate. However, it's not a "clearcut no" or "very powerful in comparison", due to the number of cores (no doubt about the GPU, I had made this explicit in the parent post). I've run the Phoronix test suite (which uses a set of real-world applications)[⁰], and, excluding the outliars for both parts (Redis/OpenSSL), we're talking about 10 to 40% advantage for the TX1, and little more for the TX2; XU4 has even the edge in one case. They're definitely faster for the majority of use cases, however, I wouldn't classify them as "very powerful in comparison", at least, when considering the performance of an RPi 3B. [⁰] https://openbenchmarking.org/result/1805276-FO-1703199RI52 https://openbenchmarking.org/result/1805276-FO-1703199RI52
    - deaddodo 8y ago
      The TX2 is a dual-core chip (ignoring the 4 "littles" since they're unused in benchmarks). The 5422 is a quad core. And even then, it's outperforming the 5422 by a decent margin in most tests. Even c-ray, an intrinsically multithreaded application is close. So you're correct, they're "comparable" in that if you take twice as many 40%-as-powerful cores, you might shrink the gap. But for any fundamentally single threaded tasks, a massive gap exists. And if/when you leverage the GPU (which can't be ignored, despite your handwaving), it becomes a canyon.