Hetzner Cloud - CPU optimized servers with AMD EPYC

iandkiandk Hosting ProviderOG

Just noticed they added some new Cloud Servers to their lineup.

Did anyone test it yet?

https://www.hetzner.com/cloud

AMD EPYC / NVMe / 10GBPs KVM in Frankfurt - https://v6node.com
Looking for an unbeatable AMD EPYC Baremetal Server in Frankfurt? Drop me a PM

Thanked by (1)hey

Comments

  • Hetzner_OLHetzner_OL Hosting ProviderOG

    Hi, thanks for posting this news! You beat me to it! :D --Katie

    Thanked by (1)iandk

    We're Katie and Lea and we'll do our best to answer questions you have about Hetzner Online. We and not our employer are responsible for any horrible puns and dated cultural references.

  • heyhey OG
    edited April 2020

    just spin 1 server

    processor   : 0
    vendor_id   : AuthenticAMD
    cpu family  : 23
    model       : 1
    model name  : AMD EPYC Processor (with IBPB)
    stepping    : 2
    microcode   : 0x1000065
    cpu MHz     : 2495.312
    cache size  : 512 KB
    physical id : 0
    siblings    : 2
    core id     : 0
    cpu cores   : 2
    apicid      : 0
    initial apicid  : 0
    fpu     : yes
    fpu_exception   : yes
    cpuid level : 13
    wp      : yes
    flags       : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx mmxext fxsr_opt pdpe1gb rdtscp lm rep_good nopl cpuid extd_apicid tsc_known_freq pni pclmulqdq ssse3 fma cx16 sse4_1 sse4_2 x2apic movbe popcnt aes xsave avx f16c rdrand hypervisor lahf_lm cmp_legacy cr8_legacy abm sse4a misalignsse 3dnowprefetch osvw ssbd ibpb vmmcall fsgsbase bmi1 avx2 smep bmi2 rdseed adx smap clflushopt sha_ni xsaveopt xsavec xgetbv1 arat
    bugs        : fxsave_leak sysret_ss_attrs spectre_v1 spectre_v2 spec_store_bypass
    bogomips    : 4990.62
    TLB size    : 1024 4K pages
    clflush size    : 64
    cache_alignment : 64
    address sizes   : 40 bits physical, 48 bits virtual
    power management:
    
    processor   : 1
    vendor_id   : AuthenticAMD
    cpu family  : 23
    model       : 1
    model name  : AMD EPYC Processor (with IBPB)
    stepping    : 2
    microcode   : 0x1000065
    cpu MHz     : 2495.312
    cache size  : 512 KB
    physical id : 0
    siblings    : 2
    core id     : 1
    cpu cores   : 2
    apicid      : 1
    initial apicid  : 1
    fpu     : yes
    fpu_exception   : yes
    cpuid level : 13
    wp      : yes
    flags       : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx mmxext fxsr_opt pdpe1gb rdtscp lm rep_good nopl cpuid extd_apicid tsc_known_freq pni pclmulqdq ssse3 fma cx16 sse4_1 sse4_2 x2apic movbe popcnt aes xsave avx f16c rdrand hypervisor lahf_lm cmp_legacy cr8_legacy abm sse4a misalignsse 3dnowprefetch osvw ssbd ibpb vmmcall fsgsbase bmi1 avx2 smep bmi2 rdseed adx smap clflushopt sha_ni xsaveopt xsavec xgetbv1 arat
    bugs        : fxsave_leak sysret_ss_attrs spectre_v1 spectre_v2 spec_store_bypass
    bogomips    : 4990.62
    TLB size    : 1024 4K pages
    clflush size    : 64
    cache_alignment : 64
    address sizes   : 40 bits physical, 48 bits virtual
    power management:
    
    

    YABS

    # ## ## ## ## ## ## ## ## ## ## ## ## ## ## ## ## ## #
    #              Yet-Another-Bench-Script              #
    #                     v2020-02-10                    #
    # https://github.com/masonr/yet-another-bench-script #
    # ## ## ## ## ## ## ## ## ## ## ## ## ## ## ## ## ## #
    
    Wed 15 Apr 2020 04:40:57 PM CEST
    
    Basic System Information:
    ---------------------------------
    Processor  : AMD EPYC Processor (with IBPB)
    CPU cores  : 2 @ 2495.312 MHz
    AES-NI     : ✔ Enabled
    VM-x/AMD-V : ❌ Disabled
    RAM        : 1.9Gi
    Swap       : 0B
    Disk       : 38G
    
    fio Disk Speed Tests (Mixed R/W 50/50):
    ---------------------------------
    Block Size | 4kb           (IOPS) | 64kb          (IOPS)
      ------   | ---            ----  | ----           ---- 
    Read       | 130.53 MB/s  (32.6k) | 1.84 GB/s    (28.7k)
    Write      | 130.87 MB/s  (32.7k) | 1.85 GB/s    (28.9k)
    Total      | 261.40 MB/s  (65.3k) | 3.69 GB/s    (57.7k)
               |                      |                     
    Block Size | 512kb         (IOPS) | 1mb           (IOPS)
      ------   | -----          ----  | ---            ---- 
    Read       | 1.98 GB/s     (3.8k) | 2.08 GB/s     (2.0k)
    Write      | 2.08 GB/s     (4.0k) | 2.22 GB/s     (2.1k)
    Total      | 4.06 GB/s     (7.9k) | 4.30 GB/s     (4.2k)
    
    iperf3 Network Speed Tests (IPv4):
    ---------------------------------
    Provider                  | Location (Link)           | Send Speed      | Recv Speed     
                              |                           |                 |                
    Bouygues Telecom          | Paris, FR (10G)           | 7.93 Gbits/sec  | 2.92 Gbits/sec 
    Online.net                | Paris, FR (10G)           | 8.66 Gbits/sec  | 2.91 Gbits/sec 
    WorldStream               | The Netherlands (10G)     | 10.6 Gbits/sec  | 3.18 Gbits/sec 
    wilhelm.tel               | Hamburg, DE (10G)         | 9.29 Gbits/sec  | 3.23 Gbits/sec 
    Biznet                    | Bogor, Indonesia (1G)     | 781 Mbits/sec   | 75.1 Mbits/sec 
    Hostkey                   | Moscow, RU (1G)           | 926 Mbits/sec   | 697 Mbits/sec  
    Velocity Online           | Tallahassee, FL, US (10G) | 1.86 Gbits/sec  | 1.72 Gbits/sec 
    Airstream Communications  | Eau Claire, WI, US (10G)  | 1.34 Gbits/sec  | 494 Mbits/sec  
    Hurricane Electric        | Fremont, CA, US (10G)     | 653 Mbits/sec   | 1.17 Gbits/sec 
    
    iperf3 Network Speed Tests (IPv6):
    ---------------------------------
    Provider                  | Location (Link)           | Send Speed      | Recv Speed     
                              |                           |                 |                
    Bouygues Telecom          | Paris, FR (10G)           | 5.13 Gbits/sec  | 2.77 Gbits/sec 
    Online.net                | Paris, FR (10G)           | 8.58 Gbits/sec  | 2.63 Gbits/sec 
    WorldStream               | The Netherlands (10G)     | 11.4 Gbits/sec  | 3.30 Gbits/sec 
    wilhelm.tel               | Hamburg, DE (10G)         | 8.92 Gbits/sec  | 3.46 Gbits/sec 
    Airstream Communications  | Eau Claire, WI, US (10G)  | 1.34 Gbits/sec  | 67.6 Mbits/sec 
    Hurricane Electric        | Fremont, CA, US (10G)     | 1.32 Gbits/sec  | 1.08 Gbits/sec 
    
    Geekbench 5 Benchmark Test:
    ---------------------------------
    Test            | Value                         
                    |                               
    Single Core     | 630                           
    Multi Core      | 1185                          
    Full Test       | https://browser.geekbench.com/v5/cpu/1803078
    
    
  • @hey said:
    just spin 1 server

    Coud you please run Geekbench 4 please?

    Thanked by (2)vimalware vish
  • @sonic said: Coud you please run Geekbench 4 please?

    sure https://browser.geekbench.com/v4/cpu/15402202

    Thanked by (3)vimalware sonic benz
  • Rescaled one of my servers to use the EPYC CPU:s:

    Awesome @Hetzner_OL :smile:

  • This is great! I might celebrate by cancelling a couple of Scaleway dedis that I have idling ;).

    Thanked by (1)Hetzner_OL
  • @hey said:

    @sonic said: Coud you please run Geekbench 4 please?

    sure https://browser.geekbench.com/v4/cpu/15402202

    Loos promise!

  • What's the best single thread Geekbench4 score so far on these?

  • Geekbench score is a good 25-50% lower than I’d expect from what’s probably an EPYC 7502(P). But this may be because @Hetzner_OL doesn't pass through the full cpu?

  • Cpu contention already? I guess I'm not too surprised. But that geekbench score looked pretty good to me, given the relatively low ghz.

  • debaserdebaser OG
    edited April 2020

    @willie said:
    Cpu contention already? I guess I'm not too surprised. But that geekbench score looked pretty good to me, given the relatively low ghz.

    I’m definitely not saying it’s bad, especially for this kind of money.

    But on a second gen EPYC you’d expect a single core score of around 4000-4700 and what I’m seeing in the Geekbench browser are scores between 2700-3200. I don’t know what causes this. Is it overbooking, noisy neighbors or just because the cpu core isn’t passed through?

  • Each "core" is actually a hardware thread (vcore), ~ 50% of a physical core. So if the box is fully utilized then even without noisy neighbors you're doing good with 60% of what you'd see on an empty machine. I tested a CX51 (8 Intel vcores, 32gb ram) a while back, doing a sustained calculation on all 8 vcores for about 3 hours, and was happy to find that I really seemed to get close to the performance of 8 physical cores. I wonder how the CPX51 (16 Epyc vcores, 32gb ram) will compare. I guess I shouldn't expect too much, since all competing products cost quite a lot more.

    Thanked by (1)Hetzner_OL
  • @sonic said:

    @hey said:
    just spin 1 server

    Coud you please run Geekbench 4 please?

    yah, I got all these gb4 numbers in my head

  • Hetzner_OLHetzner_OL Hosting ProviderOG

    @debaser We also have Hetzner Cloud options with dedicated virtual CPUs. These plans (CCX line) are based on Intel Xeon Skylake processors.
    A few other customers have asked about dedicated plans for EPYC based Cloud Servers (CPX line). If you would be interested in this, I can add a +1 to the customer wishlist for you. Helena pass these onto the dev team so they can consider what they can work on for the future. --Katie

    We're Katie and Lea and we'll do our best to answer questions you have about Hetzner Online. We and not our employer are responsible for any horrible puns and dated cultural references.

  • I know about CCX but they are way more expensive than CX/CPX. I like the CPX idea. The main interest of dedicated cpu dedicated Epyc instances (CCPX?) would be very large ones, like with 1TB ram. I couldn't afford to use those myself, but having them available would extend Hetzner's range to be more competitive with the big providers like AWS, which do have them. As for CPX, I'll wait to see how the benchmarks play out and will run some of my own tests when I can, but having more vcores at increased contention levels (so without more real speed) doesn't seem like that great a proposition.

    Meanwhile, there are some new Epyc CPU's (7F series) that run at higher ghz than the standard ones, at not that large a price premium (article. This is pretty cool. I'm sure your hardware team knows about them so I hope you'll use them at some point. I've foound Vultr's high frequency instances (similar idea with Intel cpus) to be quite impressive.

    Anyway, congrats on the new product! As new variants get introduced I'm sure some will have the right balance even if the current ones turn out not to.

    Thanked by (1)Hetzner_OL
  • cybertechcybertech OGBenchmark King

    My YABS bench thread has one on falkenstein. Not as well performing as the one in Helsinki as shown above.

    Overall decent, good iops with adequate storage.

    Thanked by (1)Hetzner_OL

    I bench YABS 24/7/365 unless it's a leap year.

Sign In or Register to comment.