4 ms·

Thanks for linking the A1100 board. I was searching for one of those but couldn't find one. It's always a bit disappointing when these small companies release c

by mastax 11y ago

Thanks for linking the A1100 board. I was searching for one of those but couldn't find one. It's always a bit disappointing when these small companies release complicated boards like this, though, the price is never competitive with more mass market stuff. I could get an Intel desktop CPU and motherboard for $150, though I guess that's missing the point.

rwmj 11y ago
There are a lot of differences between Odroid/Dragonboard and the AMD A1100-based Cello. The former boards are essentially Android phone/tablet chips with the minimal amount of support circuitry to put them on a development board, and the result is of course very cheap, owing to the sheer volume of Android chips being sold. However boards like the AMD Cello, the Gigabyte MP30-AR0, the Huskyboard, have server chips in them, external DIMMs (the Gigabyte takes up to 128GB of RAM), real SATA, PCIe slots, multiple ethernet (Gigabyte has 5 x ethernet, 2 running at 10Gbps), IPMI and so on. They also conform to the SBSA/SBBR standard so you can boot RHEL and (in the near future) Windows on them. The cost of this is significant - a fully equipped AMD Cello (with case, RAM, SDD, etc) is going to be US$600. A fully equipped Gigabyte w/32 GB of RAM and an SDD will set you back US$1000+, but you're getting a real server for that money. More about the Gigabyte board on my blog: https://rwmj.wordpress.com/?s=mp30 https://rwmj.wordpress.com/?s=mp30
- hyc_symas 11y ago
  That looks pretty good too, was wondering how X-Gene fares. In the meantime, there's still quite a gap between cheap phone SOCs and server boards. Nothing really in the middle, desktop/laptop space - 2 DIMM slots and a modest set of SATA/PCIe ports.
  - rwmj 11y ago
    The X-gene 1 is a 3 year old design in 40nm [1]. It's not blazingly fast, but it's a reliable workhorse that boots the upstream kernel out of the box. The AMD A1100 is much newer, with the excellent superscalar Cortex-A57 core (but only 4 of them). I believe that's the space that the Huskyboard and Cello are meant to occupy, although you may well think the price -- $300 just for the board -- is not exactly in that gap, and I'd agree. I have the Cello on order, and will blog about it once I receive it, expected in a month or two. [1] http://www.hotchips.org/wp-content/uploads/hc_archives/hc26/HC26-11-day1-epub/HC26.11-4-ARM-Servers-epub/HC26.11.430-X-Gene-Singh-AppMicro-HotChips-2014-v5.pdf http://www.hotchips.org/wp-content/uploads/hc_archives/hc26/...
    - hyc_symas 11y ago
      Was reading the dmesg listings on your blog. Seems that X-gene 1 doesn't have AES instructions, is that correct? I'm in the midst of optimizing the CryptoNight hash algorithm for ARMv8. Fortunately the Cortex-A53 in the RK3368 has AES instructions otherwise it would be pretty dismal.
      - rwmj 11y ago
        How to find out? /proc/cpuinfo says: Features : fp asimd evtstrm The X-gene 1 has a custom core developed by APM (called Potenza). Edit: If you mail me (_source!_) code to rjones redhat com I will run it.
        hyc_symas 11y ago
        Willdo, when I have something ready to test. Thanks! Here's the rk3368: geekbox@geekbox:/home/software/bitmonero/build$ cat /proc/cpuinfo processor : 0 BogoMIPS : 48.00 Features : fp asimd evtstrm aes pmull sha1 sha2 crc32 CPU implementer : 0x41 CPU architecture: AArch64 CPU variant : 0x0 CPU part : 0xd03 CPU revision : 3 (and so on for the other 7 cores)

by mastax 11y ago

rwmj 11y ago

There are a lot of differences between Odroid/Dragonboard and the AMD A1100-based Cello. The former boards are essentially Android phone/tablet chips with the minimal amount of support circuitry to put them on a development board, and the result is of course very cheap, owing to the sheer volume of Android chips being sold. However boards like the AMD Cello, the Gigabyte MP30-AR0, the Huskyboard, have server chips in them, external DIMMs (the Gigabyte takes up to 128GB of RAM), real SATA, PCIe slots, multiple ethernet (Gigabyte has 5 x ethernet, 2 running at 10Gbps), IPMI and so on. They also conform to the SBSA/SBBR standard so you can boot RHEL and (in the near future) Windows on them. The cost of this is significant - a fully equipped AMD Cello (with case, RAM, SDD, etc) is going to be US$600. A fully equipped Gigabyte w/32 GB of RAM and an SDD will set you back US$1000+, but you're getting a real server for that money. More about the Gigabyte board on my blog: https://rwmj.wordpress.com/?s=mp30 https://rwmj.wordpress.com/?s=mp30

hyc_symas 11y ago
That looks pretty good too, was wondering how X-Gene fares. In the meantime, there's still quite a gap between cheap phone SOCs and server boards. Nothing really in the middle, desktop/laptop space - 2 DIMM slots and a modest set of SATA/PCIe ports.
- rwmj 11y ago
  The X-gene 1 is a 3 year old design in 40nm [1]. It's not blazingly fast, but it's a reliable workhorse that boots the upstream kernel out of the box. The AMD A1100 is much newer, with the excellent superscalar Cortex-A57 core (but only 4 of them). I believe that's the space that the Huskyboard and Cello are meant to occupy, although you may well think the price -- $300 just for the board -- is not exactly in that gap, and I'd agree. I have the Cello on order, and will blog about it once I receive it, expected in a month or two. [1] http://www.hotchips.org/wp-content/uploads/hc_archives/hc26/HC26-11-day1-epub/HC26.11-4-ARM-Servers-epub/HC26.11.430-X-Gene-Singh-AppMicro-HotChips-2014-v5.pdf http://www.hotchips.org/wp-content/uploads/hc_archives/hc26/...
  - hyc_symas 11y ago
    Was reading the dmesg listings on your blog. Seems that X-gene 1 doesn't have AES instructions, is that correct? I'm in the midst of optimizing the CryptoNight hash algorithm for ARMv8. Fortunately the Cortex-A53 in the RK3368 has AES instructions otherwise it would be pretty dismal.
    - rwmj 11y ago
      How to find out? /proc/cpuinfo says: Features : fp asimd evtstrm The X-gene 1 has a custom core developed by APM (called Potenza). Edit: If you mail me (_source!_) code to rjones redhat com I will run it.
      - hyc_symas 11y ago
        Willdo, when I have something ready to test. Thanks! Here's the rk3368: geekbox@geekbox:/home/software/bitmonero/build$ cat /proc/cpuinfo processor : 0 BogoMIPS : 48.00 Features : fp asimd evtstrm aes pmull sha1 sha2 crc32 CPU implementer : 0x41 CPU architecture: AArch64 CPU variant : 0x0 CPU part : 0xd03 CPU revision : 3 (and so on for the other 7 cores)