Howdy, Stranger!

It looks like you're new here. If you want to get involved, click one of these buttons!


Hetzner new cloud instances (ARM64) - Page 2
New on LowEndTalk? Please Register and read our Community Rules.

All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.

Hetzner new cloud instances (ARM64)

2

Comments

  • rm_rm_ IPv6 Advocate, Veteran
    edited April 2023

    @amarc said: Did you select country with VAT ?

    Interestingly I selected Russia and it still shows the 20% VAT to be charged. Did Hetzner find a way to avoid sanctions and wire-transfer it somehow to Putin? Or they just pocket the extra money charged from customers as "VAT"? @Hetzner_OL what is your take?

    Thanked by 1SashkaPro
  • @rm_ said:

    @amarc said: Did you select country with VAT ?

    Interestingly I selected Russia and it still shows the 20% VAT to be charged. Did Hetzner find a way to avoid sanctions and wire-transfer it somehow to Putin? Or they just pocket the extra money charged from customers as "VAT"? @Hetzner_OL what is your take?

    i'm really interested in knowing that. i mean i'm not russian but there's no VAT for my country. for example when i order at ovh they put VAT for my country automatically, but then i can contact support and they will remove the VAT from invoice but its weird how do they calculate VAT by default. just taking some extra money i guess. and 20% is not that little in big invoices.

  • ErisaErisa Member

    @darkimmortal said:
    Any nested virt on these?

    The cpu theyre using (Ampere Altra) doesn't support nested virtualisation, so not possible for them to offer it at the moment.

    That said, Hetzner Cloud doesnt support nested virt on their x86 machines either even though the cpus are capable, so I wouldn't get your hopes up.

    Thanked by 2darkimmortal nick_
  • cpsdcpsd Member

    @Erisa said:

    @darkimmortal said:
    Any nested virt on these?

    The cpu theyre using (Ampere Altra) doesn't support nested virtualisation, so not possible for them to offer it at the moment.

    That said, Hetzner Cloud doesnt support nested virt on their x86 machines either even though the cpus are capable, so I wouldn't get your hopes up.

    I was using in their x86 architecture both virt-manager and phpvirtualbox (old), but with poor performance.

  • tjntjn Member

    @emgh said:
    1. You sure RunCloud even runs on ARM?

    Didn't think it would, but just in case I had a go and the installer failed with:
    This installer only support x86_64 architecture

    Didn't try harder than that tbh.

    Thanked by 1emgh
  • eriseris Member

    Hestia supports ARM

  • I just installed Cloudpanel on cax11 and it works fine. Cloudpanel supports arm64

  • @thuanbui said: I just installed Cloudpanel on cax11 and it works fine. Cloudpanel supports arm64

    Try WordOps please.

  • HostWildHostWild Member, Host Rep

    For more about ARM Servers:

    Intel and AMD processors are generally X86 architecture where as ARM processors are RISC processors based on ARM architecture. ARM processors offer lower performance than AMD and Intel processors but, they also consume lower power. If you need a processor for a desktop, you have to go with INTEL or AMD. ARM processors are generally used in mobile phones, smart watches and other lpw power devices.

  • m4num4nu Member, Patron Provider

    @HostWild said:
    For more about ARM Servers:...

    Thanks, GPT4

    Thanked by 1bjo
  • emghemgh Member

    @HostWild said: ARM processors offer lower performance than AMD and Intel processors

    Thanked by 1kheng86
  • ARM processors offer lower performance than AMD and Intel processors but, they also consume lower power.

    Apple silicon: Interesting...

  • Any other hosts selling ARM Ampere servers?

  • KebabKebab Member

    @NewbieInTraining said:
    Any other hosts selling ARM Ampere servers?

    Oracle and scaleway has some kind of beta

  • davidedavide Member
    edited April 2023

    I moved a 15GB mariadb database from x86 to arm, which is the instance provided by oracle's free tier, by copying the binary DB files over to the new server as-is, without passing thru mysqldump and import. Everything worked smooth, even the replication resumed without flaws.

    I recompiled my c++ programs from x86 to this arm instance, and all worked fine, both cpu types use IEEE 754 for floating point representation, both have 32 and 64 bits integer and float types. From my c++ perspective, the only difference between the two cpus is the cacheline size, 64B on x86 and 128B on arm. No need to say, all nodejs, bash and other scripts work without modifications. The transition is completely smooth. For years I had been afraid of moving to arm for risking any kind of incompatibility in the binary representation. Turns out that fear was unmotivated

  • AWS also offer ARM but use their own Graviton CPUs.

    Though I'm sure they're probably just modified existing ARM CPUs rather than ground-up Amazon designed

    Thanked by 1Erisa
  • @davide said:
    I moved a 15GB mariadb database from x86 to arm, which is the instance provided by oracle's free tier, by copying the binary DB files over to the new server as-is, without passing thru mysqldump and import. Everything worked smooth, even the replication resumed without flaws.

    I recompiled my c++ programs from x86 to this arm instance, and all worked fine, both cpu types use IEEE 754 for floating point representation, both have 32 and 64 bits integer and float types. From my c++ perspective, the only difference between the two cpus is the cacheline size, 64B on x86 and 128B on arm. No need to say, all nodejs, bash and other scripts work without modifications. The transition is completely smooth. For years I had been afraid of moving to arm for risking any kind of incompatibility in the binary representation. Turns out that fear was unmotivated

    Did you write any SIMD intrinsics manually and converted them to Neon?

  • davidedavide Member
    edited April 2023

    @noisycode said:
    Did you write any SIMD intrinsics manually and converted them to Neon?

    No, the most I do is #pragma unroll, prefetch, and data structures optimization with callgrind and linux-perf to flatten bottlenecks. I don't want to descend into that darkness. It certainly helps with portability.

    Perhaps your SIMD requirements can be fulfilled by some portable BLAS library.

  • @davide said:

    @noisycode said:
    Did you write any SIMD intrinsics manually and converted them to Neon?

    No, the most I do is #pragma unroll, prefetch, and data structures optimization with callgrind and linux-perf to flatten bottlenecks. I don't want to descend into that darkness. It certainly helps with portability.

    Perhaps your SIMD requirements can be fulfilled by some portable BLAS library.

    Understood. For me, BLAS lib, especially MKL, works well with most of the parts, but when it comes to CPU cache reuse under specific situation, I have to make my hands dirty, 'cause BLAS libs are usually for general purpose, and do not offer the FASTEST solution.

  • davidedavide Member
    edited April 2023

    I can totally see your perf stat reports 16 instructions per cycle and the cpus run at 95°C. I'm defeated at 2 ins/cycle at 85°C :(

  • BasToTheMaxBasToTheMax Member, Host Rep

    @davide said:
    I moved a 15GB mariadb database from x86 to arm, which is the instance provided by oracle's free tier, by copying the binary DB files over to the new server as-is, without passing thru mysqldump and import. Everything worked smooth, even the replication resumed without flaws.

    I recompiled my c++ programs from x86 to this arm instance, and all worked fine, both cpu types use IEEE 754 for floating point representation, both have 32 and 64 bits integer and float types. From my c++ perspective, the only difference between the two cpus is the cacheline size, 64B on x86 and 128B on arm. No need to say, all nodejs, bash and other scripts work without modifications. The transition is completely smooth. For years I had been afraid of moving to arm for risking any kind of incompatibility in the binary representation. Turns out that fear was unmotivated

    Did performance improve? (like faster queries ect)

    And how much resources do you have (ram / cpu)?

    Thanked by 1pbx
  • Hetzner CAX11 ARM Server

    https://pastebin.com/pMxSy4E7

  • alt_alt_ Member

    Tested via bench.sh

    Thanked by 1maverick
  • febryanvaldofebryanvaldo Member
    edited April 2023

    Actually it's pretty good on some benchs.

    Benchmark by KangServer.id
    Performed: April 18th 2023

    ----------------------------------------------------------------------------
    ***** Passmark's PerformanceTest *****
    ----------------------------------------------------------------------------
                    Hetzner CAX31 - 11.99 EURO      | Hetzner CPX31 - 13.1 EURO                          
                         8 vCPU 16GB RAM            |      4 vCPU 8GB RAM   
    ----------------------------------------------------------------------------                
    Single Threaded:        1320                    |          2045
    CPU Mark       :        6224                    |          8232
    ----------------------------------------------------------------------------
    
    ----------------------------------------------------------------------------
    ***** Geekbench 5 *****
    ----------------------------------------------------------------------------
                    Hetzner CAX31 - 11.99 EURO      | Hetzner CPX31 - 13.1 EURO                          
                         8 vCPU 16GB RAM            |      4 vCPU 8GB RAM   
    ----------------------------------------------------------------------------
    Single Core:               803                  |           974
    Multi Core :              5909                  |          3686
    https://browser.geekbench.com/v5/cpu/21076774   | https://browser.geekbench.com/v5/cpu/21076766
    ----------------------------------------------------------------------------
    
    ----------------------------------------------------------------------------
    ***** Geekbench 6 *****
    ----------------------------------------------------------------------------
                    Hetzner CAX31 - 11.99 EURO      | Hetzner CPX31 - 13.1 EURO                          
                         8 vCPU 16GB RAM            |      4 vCPU 8GB RAM   
    ----------------------------------------------------------------------------                
    Single Core:              1034                  |          1242
    Multi Core :              5649                  |          4006
    https://browser.geekbench.com/v6/cpu/960880     | https://browser.geekbench.com/v6/cpu/960862
    ----------------------------------------------------------------------------
    
    ----------------------------------------------------------------------------
    ***** UnixBench *****
    ----------------------------------------------------------------------------
                    Hetzner CAX31 - 11.99 EURO      | Hetzner CPX31 - 13.1 EURO                          
                         8 vCPU 16GB RAM            |      4 vCPU 8GB RAM   
    ----------------------------------------------------------------------------
    Single SBIS:             1288.1                 |         1850.2
    Multi SBIS :             8948.2                 |         6825.2
    ----------------------------------------------------------------------------
    
    ----------------------------------------------------------------------------
    ***** PTS PostgreSQL (pgbench) *****
    Scaling Factor: 100 | Clients: 100 | Mode: Read-Only
    ----------------------------------------------------------------------------
                    Hetzner CAX31 - 11.99 EURO      | Hetzner CPX31 - 13.1 EURO                          
                         8 vCPU 16GB RAM            |      4 vCPU 8GB RAM   
    ----------------------------------------------------------------------------                
    Score      :           133666 TPS               |       138406 TPS  
    Avg Latency:            0.748 ms                |        0.723 ms
    ----------------------------------------------------------------------------
    
    ----------------------------------------------------------------------------
    ***** PTS Nginx Benchmark *****
    Connections: 1000
    ----------------------------------------------------------------------------
                    Hetzner CAX31 - 11.99 EURO      | Hetzner CPX31 - 13.1 EURO                          
                         8 vCPU 16GB RAM            |      4 vCPU 8GB RAM  
    ----------------------------------------------------------------------------                
    Score                 21617.43 RPS              |      19791.11 TPS
    ----------------------------------------------------------------------------
    
    ----------------------------------------------------------------------------
    ***** PTS Apache HTTP Server *****
    Connections: 1000
    ----------------------------------------------------------------------------
                    Hetzner CAX31 - 11.99 EURO      | Hetzner CPX31 - 13.1 EURO                          
                         8 vCPU 16GB RAM            |      4 vCPU 8GB RAM  
    ----------------------------------------------------------------------------                
    Score:                23250.16 RPS              |      25327.43 TPS
    ----------------------------------------------------------------------------
    
    ----------------------------------------------------------------------------
    ***** PTS PHPBench *****
    ----------------------------------------------------------------------------
                    Hetzner CAX31 - 11.99 EURO      | Hetzner CPX31 - 13.1 EURO                          
                         8 vCPU 16GB RAM            |      4 vCPU 8GB RAM   
    ----------------------------------------------------------------------------                
    Score:                   530440                 |         650093
    ----------------------------------------------------------------------------
    
    ----------------------------------------------------------------------------
    ***** PTS Redis *****
    Test options: SET & GET | Parallel Connections: 50
    ----------------------------------------------------------------------------
                    Hetzner CAX31 - 11.99 EURO      | Hetzner CPX31 - 13.1 EURO                          
                         8 vCPU 16GB RAM            |      4 vCPU 8GB RAM         
    ----------------------------------------------------------------------------                
    GET:                 1575098.00 RPS             |        2149429.92 RPS
    SET:                 1156443.41 RPS             |        1600233.71 RPS
    ----------------------------------------------------------------------------
    
    ----------------------------------------------------------------------------
    ***** PTS MariaDB (mysqlslap) *****
    Clients: 1024
    ----------------------------------------------------------------------------
                    Hetzner CAX31 - 11.99 EURO      | Hetzner CPX31 - 13.1 EURO                          
                         8 vCPU 16GB RAM            |      4 vCPU 8GB RAM        
    ----------------------------------------------------------------------------                
    Score:                  188 QPS                 |          223 QPS
    ----------------------------------------------------------------------------
    
    ----------------------------------------------------------------------------
    *** PTS 7-Zip Compression ***
    ----------------------------------------------------------------------------
                    Hetzner CAX31 - 11.99 EURO      | Hetzner CPX31 - 13.1 EURO                          
                         8 vCPU 16GB RAM            |      4 vCPU 8GB RAM   
    ----------------------------------------------------------------------------
    Compression  :         35697 MIPS               |       25581 MIPS
    Decompression:         32493 MIPS               |       16605 MIPS
    ----------------------------------------------------------------------------
    
    ----------------------------------------------------------------------------
    Hetzner CAX31 - 11.99 EURO
          8 vCPU 16GB RAM
          
    fio Disk Speed Tests (Mixed R/W 50/50):
    ---------------------------------
    Block Size | 4k            (IOPS) | 64k           (IOPS)
      ------   | ---            ----  | ----           ----
    Read       | 115.29 MB/s  (28.8k) | 1.11 GB/s    (17.4k)
    Write      | 115.21 MB/s  (28.8k) | 1.15 GB/s    (17.9k)
    Total      | 230.50 MB/s  (57.6k) | 2.26 GB/s    (35.4k)
               |                      |
    Block Size | 512k          (IOPS) | 1m            (IOPS)
      ------   | ---            ----  | ----           ----
    Read       | 1.88 GB/s     (3.6k) | 2.29 GB/s     (2.2k)
    Write      | 2.04 GB/s     (3.9k) | 2.56 GB/s     (2.5k)
    Total      | 3.92 GB/s     (7.6k) | 4.85 GB/s     (4.7k)
    
    ----------------------------------------------------------------------
     I/O Speed(1st run) : 1.0 GB/s
     I/O Speed(2nd run) : 979 MB/s
     I/O Speed(3rd run) : 1.1 GB/s
     I/O Speed(average) : 1043.1 MB/s
    ----------------------------------------------------------------------
    
    ----------------------------------------------------------------------------
    Hetzner CPX31 - 13.1 EURO
          4 vCPU 8GB RAM
          
    fio Disk Speed Tests (Mixed R/W 50/50):
    ---------------------------------
    Block Size | 4k            (IOPS) | 64k           (IOPS)
      ------   | ---            ----  | ----           ----
    Read       | 132.48 MB/s  (33.1k) | 1.66 GB/s    (25.9k)
    Write      | 132.83 MB/s  (33.2k) | 1.67 GB/s    (26.1k)
    Total      | 265.32 MB/s  (66.3k) | 3.33 GB/s    (52.0k)
               |                      |
    Block Size | 512k          (IOPS) | 1m            (IOPS)
      ------   | ---            ----  | ----           ----
    Read       | 2.39 GB/s     (4.6k) | 2.36 GB/s     (2.3k)
    Write      | 2.52 GB/s     (4.9k) | 2.52 GB/s     (2.4k)
    Total      | 4.92 GB/s     (9.6k) | 4.88 GB/s     (4.7k)
    
    ----------------------------------------------------------------------
     I/O Speed(1st run) : 1.4 GB/s
     I/O Speed(2nd run) : 1.5 GB/s
     I/O Speed(3rd run) : 1.5 GB/s
     I/O Speed(average) : 1501.9 MB/s
    ----------------------------------------------------------------------------
    
    Thanked by 2sebkehl atomi
  • CAX31 is not 4 vCPU 8GB RAM, but 8 core 16 GB RAM
    CPX31 is not 8 vCPU 16GB RAM, but 4 core 8 GB RAM

    At this point it's hard to believe any of the results above, unless you go through all of it once again and make neccessary corrections. But, thanks for the effort.

  • Also if I may suggest comparing similar models? Say, both AMD & ARM models with 4 cores (or both with 8 cores)? So you're not comparing oranges and apples. :wink:

  • febryanvaldofebryanvaldo Member
    edited April 2023

    @maverick said: CAX31 is not 4 vCPU 8GB RAM, but 8 core 16 GB RAM
    CPX31 is not 8 vCPU 16GB RAM, but 4 core 8 GB RAM

    Yes, already edited.

  • @maverick said: Also if I may suggest comparing similar models? Say, both AMD & ARM models with 4 cores (or both with 8 cores)? So you're not comparing oranges and apples.

    I'm comparing the pricing. The closest one.

  • mrcznmrczn Member

    Hi. Hope it's not necroing, this topic seemed the most appropriate
    I've spun up a CAX11 of my own and gb5 results are very different from what is usually reported, even in this thread

    For example here's vpsbenchmarks run: https://browser.geekbench.com/v5/cpu/21146539
    805 Single-Core Score, 1522 Multi-Core Score

    and here's mine: https://browser.geekbench.com/v5/cpu/21514819
    288 Single-Core Score, 542 Multi-Core Score

    what could be the reason for that? It's basically a fresh server, no CPU utilization with an expected spike to ~200% during the rest run, maybe ~150MB RAM taken
    I expected that a shared server would sometimes have to compete for resources, but these results are 3x times worse than previous benchmarks

    Thanks

Sign In or Register to comment.