Howdy, Stranger!

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


Hetzner launch dedi-vCPU cloud instances- starting €19.9/ 2vCPUs
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 launch dedi-vCPU cloud instances- starting €19.9/ 2vCPUs

vimalwarevimalware Member
edited August 2018 in Providers

Just noticed this when i logged in to my console.

Edit: all prices excluding VAT.

Here's the official release:

https://www.hetzner.com/news/08-18-dedicated-cpu-hc/

Hetzner Cloud has just launched 5 Dedicated vCPU Server models to give you better performance and flexibility.

The servers house between 2 and 32 dedicated vCPUs and therefore provide you with a maximum amount of CPU performance as well as generous I/O and network speeds.

These models are ideal for systems with large production loads and CPU intensive applications.

The new Dedicated vCPU Servers start at €19.90 a month or € 0.03 per hour.

You can find an overview of all of our cloud servers [here](https://www.hetzner.com/cloud).
Thanked by 2beagle Hetzner_OL
«1

Comments

  • @hetzner_OL are these dedicated cores, or dedicated threads sold as cores?

  • YuraYura Member

    At these prices what are advantages over Heinz's very own dedis? I guess, built in hypervisor is one...

    Thanked by 1pike
  • kbapkbap Member
    edited August 2018

    teamacc said:
    @hetzner_OL are these dedicated cores, or dedicated threads sold as cores?

    It's a dedicated thread.

    What are the dedicated vCPU server plans?
    Every dedicated vCPU instance has its own dedicated CPU resources (1 vCPU = 1 hyper-thread)

    https://wiki.hetzner.de/index.php/CloudServer/en#What_are_the_dedicated_vCPU_server_plans.3F

  • NeoonNeoon Community Contributor, Veteran
    edited August 2018

    So, you pay for 2 Dedicated cores, 19EUR instead of paying 21.85EUR for a whole dedi with 32gig, I do not see the point.

    Must be the "cloud".

  • @Neoon said:
    So, you pay for 2 Dedicated cores, 19EUR instead of paying 21.85EUR for a whole dedi with 32gig, I do not see the point.

    Must be the "cloud".

    something something nvme

    Thanked by 1netomx
  • NeoonNeoon Community Contributor, Veteran

    @teamacc said:

    @Neoon said:
    So, you pay for 2 Dedicated cores, 19EUR instead of paying 21.85EUR for a whole dedi with 32gig, I do not see the point.

    Must be the "cloud".

    something something nvme

    "speedy NVMe SSDs "

    I did benchmarks, I never went higher then 400MB/sec, so does not look like NVMe to me.

  • Neoon said: 19EUR instead of paying 21.85EUR for a whole dedi with 32gig

    you can pay hourly on this dedicated vcore cloud

    Thanked by 3pike Hetzner_OL Wolveix
  • @Neoon said:
    so does not look like NVMe to me.

    According to [citation needed] the main benefit of nvme is lower latency. Did you test io latency?

  • vfusevfuse Member, Host Rep
    edited August 2018
    8vcpu 32gb ram dedi: https://browser.geekbench.com/v4/cpu/9321798
    8vcpu 32gb ram shared: https://browser.geekbench.com/v4/cpu/9321826
    

    Great performance on the dedicated vcpu one, but price/performance wise the shared one is better choice.

    Thanked by 2Falzo Aluminat
  • NeoonNeoon Community Contributor, Veteran

    @teamacc said:

    @Neoon said:
    so does not look like NVMe to me.

    According to [citation needed] the main benefit of nvme is lower latency. Did you test io latency?

    Well, SSD has lower latency then HDD, but the main difference form NVMe to SSD is, SSD is connected and limited to SATA, NVMe uses PCIe so can reach higher speeds.

    Could be that NVMe is a bit lower then SSD in latency, but I would not say that much.

    Well, I did checked it with ioping but not compared NVMe to SSD.

  • kbapkbap Member

    Were you on CEPH or local storage?

  • NeoonNeoon Community Contributor, Veteran

    @kbap said:
    Were you on CEPH or local storage?

    Local.

    Thanked by 1Hetzner_OL
  • Just for everyone i got some benchmarks for you :)

    CCX11

    Processor:    Intel Xeon Processor (Skylake, IBRS)
    CPU cores:    2
    Frequency:    2100.000 MHz
    RAM:          7.6G
    Swap:         135M
    Kernel:       Linux 4.15.0-29-generic x86_64
    
    Disks:
    sda   76.3G  HDD
    
    CPU: SHA256-hashing 500 MB
        2.442 seconds
    CPU: bzip2-compressing 500 MB
        4.070 seconds
    CPU: AES-encrypting 500 MB
        0.832 seconds
    
    ioping: seek rate
        min/avg/max/mdev = 31.3 us / 55.3 us / 2.78 ms / 21.4 us
    ioping: sequential read speed
        generated 31.2 k requests in 5.00 s, 7.62 GiB, 6.24 k iops, 1.52 GiB/s
    
    dd: sequential write speed
        1st run:    383.38 MiB/s
        2nd run:    302.31 MiB/s
        3rd run:    404.36 MiB/s
        average:    363.35 MiB/s
    
    IPv4 speedtests
        your IPv4:    195.201.37.xxxx
    
        Cachefly CDN:         317.46 MiB/s
        Leaseweb (NL):        114.08 MiB/s
        Softlayer DAL (US):   1.08 MiB/s
        Online.net (FR):      55.64 MiB/s
        OVH BHS (CA):         4.53 MiB/s
    
    IPv6 speedtests
        your IPv6:    2a01:4f8:1c1c:xxxx
    
        Leaseweb (NL):        68.31 MiB/s
        Softlayer DAL (US):   2.00 MiB/s
        Online.net (FR):      66.83 MiB/s
        OVH BHS (CA):         10.24 MiB/s
    

    Let me know if you want to see any of the other ones benchmarked or any other tests :)

    Thanked by 1Waldo19
  • I feel this should have been deployed to E3 cpus.

    Maybe an E3 variant is in the planning pipeline.

  • FHRFHR Member, Host Rep
    edited August 2018

    Neoon said:

    @Neoon said: so does not look like NVMe to me.

    @teamacc said:
    According to [citation needed] the main benefit of nvme is lower latency. Did you test io latency?

    Well, SSD has lower latency then HDD, but the main difference form NVMe to SSD is, SSD is connected and limited to SATA, NVMe uses PCIe so can reach higher speeds.
    Could be that NVMe is a bit lower then SSD in latency, but I would not say that much.
    Well, I did checked it with ioping but not compared NVMe to SSD.

    The main difference between NVMe and SATA SSDs is actually the protocol used, not the interface. SATA SSDs use AHCI, while NVMe SSDs use, as one could guess, the NVMe protocol.

    NVMe promises a much lower latency compared to AHCI, much better efficiency, ability to do more things at once (parallelism) and much larger queues.

    This means that NVMe should perform much better especially in heavy workloads compared to their SATA counterparts.

    Back to the topic, I would call 400MB/s on a shared server which probably houses tens of other VMs a pretty good number.

    Thanked by 1levnode
  • williewillie Member
    edited August 2018

    I ran some tasks on an 8-core 32gb instance a few months back and got close to 100% cpu utilization, i.e. equivalent to 8 dedicated cores (not threads) for around 3 hours. I guess it was too good to last and the miners (or whatever) moved in. The current pricing is still great vs. comparable hourly products (OVH public cloud B2 series I guess) but not shocking like before. At the monthly level Netcup rootservers look good, as do actual dedis.

    Also gotta test the core vs vcore situation. The 3 hour task I ran is a monthly data cleanup thing that I normally do on my auction dedi so I guess I'll try it on one of the new instances. I think OVH public cloud may literally allocate dedicated cores, i.e. 8 vcpu = 4 physical cores and you get all the threads, but that's not so great compared to 8 shared threads on an underutilized box. Typical server workloads aren't that cpu intensive most of the time, so running a short lived cpu-heavy application on a shared box can work great (other people get the ram, you get the cpu). Long lived applications of that type should run on dedis of course.

    It's been possible for a while to upgrade/downgrade existing instance sizes. I wonder if moving between dedicated-core and non-dedicated works.

    It's nice that bigger instances (64 and 128gb) are available since that's needed sometimes. The hourly rates are high enough that I start wishing for minute by minute billing as some of the other "hourly" services now implement. Still waiting for GPU instances. And I'm sure the dev team is working on more "cloud" features (load balancing, VLAN, object store, etc), so just saying, some of us are watching for that too.

  • FHR said:

    Back to the topic, I would call 400MB/s on a shared server which probably houses tens of other VMs a pretty good number.

    If you're feeling extravagant, I wonder if you could try that test on a 128GB instance, so there wouldn't be tens of other VMs. Parallel cpu tests would also be great.

  • FHRFHR Member, Host Rep

    willie said: If you're feeling extravagant, I wonder if you could try that test on a 128GB instance, so there wouldn't be tens of other VMs. Parallel cpu tests would also be great.

    Can't do 32 cores, as accounts are limit do 8 by default. Anyway, did the 8 core one.

    Can't say I like these dedicated packages at all - they use exactly the same CPU as on their normal "cloud".

    Hetzner CX11 NUREMBERG
    ----------------------------------------------------------------------
    CPU model            : Intel Xeon Processor (Skylake, IBRS)
    Number of cores      : 1
    CPU frequency        : 2099.998 MHz
    Total size of Disk   : 19.0 GB (1.2 GB Used)
    Total amount of Mem  : 1790 MB (53 MB Used)
    Total amount of Swap : 0 MB (0 MB Used)
    System uptime        : 0 days, 0 hour 1 min
    Load average         : 0.60, 0.27, 0.10
    OS                   : CentOS 7.5.1804
    Arch                 : x86_64 (64 Bit)
    Kernel               : 3.10.0-862.6.3.el7.x86_64
    ----------------------------------------------------------------------
    I/O speed(1st run)   : 56.3 MB/s
    I/O speed(2nd run)   : 224 MB/s
    I/O speed(3rd run)   : 290 MB/s
    Average I/O speed    : 190.1 MB/s
    ----------------------------------------------------------------------
    Node Name                       IPv4 address            Download Speed
    CacheFly                        205.234.175.175         154MB/s       
    Linode, Tokyo, JP               106.187.96.148          5.35MB/s      
    Linode, Singapore, SG           139.162.23.4            5.16MB/s      
    Linode, London, UK              176.58.107.39           80.5MB/s      
    Linode, Frankfurt, DE           139.162.130.8           141MB/s       
    Linode, Fremont, CA             50.116.14.9             9.28MB/s      
    Softlayer, Dallas, TX           173.192.68.18           8.66MB/s      
    Softlayer, Seattle, WA          67.228.112.250          7.10MB/s      
    Softlayer, Frankfurt, DE        159.122.69.4            53.3MB/s      
    Softlayer, Singapore, SG        119.81.28.170           5.72MB/s      
    Softlayer, HongKong, CN         119.81.130.170          4.97MB/s      
    ----------------------------------------------------------------------
    Node Name                       IPv6 address            Download Speed
    Linode, Atlanta, GA             2600:3c02::4b           12.3MB/s      
    Linode, Dallas, TX              2600:3c00::4b           10.7MB/s      
    Linode, Newark, NJ              2600:3c03::4b           7.93MB/s      
    Linode, Singapore, SG           2400:8901::4b           8.48MB/s      
    Linode, Tokyo, JP               2400:8900::4b           5.20MB/s      
    Softlayer, San Jose, CA         2607:f0d0:2601:2a::4    5.70MB/s      
    Softlayer, Washington, WA       2607:f0d0:3001:78::2    4.58MB/s      
    Softlayer, Paris, FR            2a03:8180:1301:8::4     49.6MB/s      
    Softlayer, Singapore, SG        2401:c900:1101:8::2     6.13MB/s      
    Softlayer, Tokyo, JP            2401:c900:1001:16::4    3.64MB/s      
    ----------------------------------------------------------------------
    
    
    
    HETZNER CCX31 NUREMBERG
    ----------------------------------------------------------------------
    CPU model            : Intel Xeon Processor (Skylake, IBRS)
    Number of cores      : 8
    CPU frequency        : 2099.998 MHz
    Total size of Disk   : 226.0 GB (1.2 GB Used)
    Total amount of Mem  : 31256 MB (229 MB Used)
    Total amount of Swap : 0 MB (0 MB Used)
    System uptime        : 0 days, 0 hour 0 min
    Load average         : 0.30, 0.08, 0.03
    OS                   : CentOS 7.5.1804
    Arch                 : x86_64 (64 Bit)
    Kernel               : 3.10.0-862.9.1.el7.x86_64
    ----------------------------------------------------------------------
    I/O speed(1st run)   : 349 MB/s
    I/O speed(2nd run)   : 366 MB/s
    I/O speed(3rd run)   : 351 MB/s
    Average I/O speed    : 355.3 MB/s
    ----------------------------------------------------------------------
    Node Name                       IPv4 address            Download Speed
    CacheFly                        205.234.175.175         123MB/s       
    Linode, Tokyo, JP               106.187.96.148          5.35MB/s      
    Linode, Singapore, SG           139.162.23.4            7.63MB/s      
    Linode, London, UK              176.58.107.39           76.1MB/s      
    Linode, Frankfurt, DE           139.162.130.8           133MB/s       
    Linode, Fremont, CA             50.116.14.9             9.70MB/s      
    Softlayer, Dallas, TX           173.192.68.18           8.02MB/s      
    Softlayer, Seattle, WA          67.228.112.250          6.48MB/s      
    Softlayer, Frankfurt, DE        159.122.69.4            44.4MB/s      
    Softlayer, Singapore, SG        119.81.28.170           5.24MB/s      
    Softlayer, HongKong, CN         119.81.130.170          4.42MB/s      
    ----------------------------------------------------------------------
    Node Name                       IPv6 address            Download Speed
    Linode, Atlanta, GA             2600:3c02::4b           12.8MB/s      
    Linode, Dallas, TX              2600:3c00::4b           10.4MB/s      
    Linode, Newark, NJ              2600:3c03::4b           14.2MB/s      
    Linode, Singapore, SG           2400:8901::4b           6.87MB/s      
    Linode, Tokyo, JP               2400:8900::4b           5.16MB/s      
    Softlayer, San Jose, CA         2607:f0d0:2601:2a::4    4.45MB/s      
    Softlayer, Washington, WA       2607:f0d0:3001:78::2    12.0MB/s      
    Softlayer, Paris, FR            2a03:8180:1301:8::4     48.1MB/s      
    Softlayer, Singapore, SG        2401:c900:1101:8::2     5.42MB/s      
    Softlayer, Tokyo, JP            2401:c900:1001:16::4    3.48MB/s      
    ----------------------------------------------------------------------
    
  • FHR said: Can't say I like these dedicated packages at all - they use exactly the same CPU as on their normal "cloud".

    Yeah the presumed difference is the available cpu share. The scary idea is that the available "burst" cpu might be -lower- with the dedicated core plans than with the shared ones. I put "burst" in quotes because again my previous test on a 32gb kept the 8 cores pegged for 3 hours, which is above what I'd think of as a burst. I don't know if that could have been sustained for say 1 week though.

    I agree these are less impressive than the earlier versions and with the dedis. I'd be surprised if the 128gb/32 vcpu instance can outperform an AX160 which is considerably cheaper, has more SSD, can be upgraded to 1TB(?) of ram and tens of TB of SSD or HDD,, and currently has no setup fee. Similarly my i7-3770 auction dedi is around midway between the 4 core and 8 core cloud instances in cpu performance, at less than half the cost of the 4 core dedi-vcpu instance.

  • @FHR said:

    Can't do 32 cores, as accounts are limit do 8 by default. Anyway, did the 8 core one.

    Can't say I like these dedicated packages at all - they use exactly the same CPU as on their normal "cloud".

    For the 8 vCPU.. maybe - but for 32 vCPU, they can't use the same CPU - Skylake (E3v5 CPUs) ain't available in a 16C/32T config - so they'd either have to do E5 or xeon scalable for the bigger instances.

  • FHRFHR Member, Host Rep
    edited August 2018

    @Zerpy said:

    @FHR said:

    Can't do 32 cores, as accounts are limit do 8 by default. Anyway, did the 8 core one.

    Can't say I like these dedicated packages at all - they use exactly the same CPU as on their normal "cloud".

    For the 8 vCPU.. maybe - but for 32 vCPU, they can't use the same CPU - Skylake (E3v5 CPUs) ain't available in a 16C/32T config - so they'd either have to do E5 or xeon scalable for the bigger instances.

    Those are not E3 though. Judging by the frequency of 2100MHz those are E5s

    They could be using Xeon Silver 4110 chips, or something similar

  • neikneik Member
    edited August 2018

    8vcpu 32gb ram shared: https://browser.geekbench.com/v4/cpu/9321826
    Great performance on the dedicated vcpu one, but price/performance wise the shared one is better choice.

    Sorry guys for this dumb question but how would I run Geekbench? Do I need to install it or is there some sort of script available like bench.sh?

    Would like to run Geekbench on my Netcup Rootserver but can't figure out how to do it.

  • Hetzner_OLHetzner_OL Member, Top Host

    Neoon said: Local.

    The Hetzner Cloud Dedicated vCPU Cervers, "are only available with local (NVMe SSD) storage." https://wiki.hetzner.de/index.php/CloudServer/en#What_are_the_dedicated_vCPU_server_plans.3F --Katie, Marketing

  • HaxHax Member

    @neik said:
    Sorry guys for this dumb question but how would I run Geekbench? Do I need to install it or is there some sort of script available like bench.sh?

    Would like to run Geekbench on my Netcup Rootserver but can't figure out how to do it.

    wget http://cdn.geekbench.com/Geekbench-4.2.3-Linux.tar.gz
    tar -zxvf Geekbench-4.2.3-Linux.tar.gz
    cd Geekbench-4.2.3-Linux/
    ./geekbench4
    
  • Hetzner_OLHetzner_OL Member, Top Host

    willie said: It's been possible for a while to upgrade/downgrade existing instance sizes. I wonder if moving between dedicated-core and non-dedicated works.

    Yes, this is possible. You should already be able to do this using the rescaling feature on Cloud Console. --Katie, Marketing

  • neikneik Member

    @Sixell, thanks mate!

  • vfusevfuse Member, Host Rep

    Another geekbench from the 32 core plan: https://browser.geekbench.com/v4/cpu/9344676

    32 core: 47821
    8 core: 24352
    8 core shared: 12648

    ----------------------------------------------------------------------
    CPU model            : Intel Xeon Processor (Skylake, IBRS)
    Number of cores      : 32
    CPU frequency        : 2099.998 MHz
    Total size of Disk   : 508.0 GB (1.9 GB Used)
    Total amount of Mem  : 125907 MB (285 MB Used)
    Total amount of Swap : 135 MB (0 MB Used)
    System uptime        : 0 days, 0 hour 9 min
    Load average         : 0.13, 1.96, 1.32
    OS                   : Ubuntu 18.04.1 LTS
    Arch                 : x86_64 (64 Bit)
    Kernel               : 4.15.0-29-generic
    ----------------------------------------------------------------------
    I/O speed(1st run)   : 344 MB/s
    I/O speed(2nd run)   : 366 MB/s
    I/O speed(3rd run)   : 384 MB/s
    Average I/O speed    : 364.7 MB/s
    ----------------------------------------------------------------------
    Node Name                       IPv4 address            Download Speed
    CacheFly                        205.234.175.175         154MB/s
    Linode, Tokyo, JP               106.187.96.148          7.79MB/s
    Linode, Singapore, SG           139.162.23.4            8.33MB/s
    Linode, London, UK              176.58.107.39           81.9MB/s
    Linode, Frankfurt, DE           139.162.130.8           156MB/s
    Linode, Fremont, CA             50.116.14.9             14.7MB/s
    Softlayer, Dallas, TX           173.192.68.18           13.2MB/s
    Softlayer, Seattle, WA          67.228.112.250          9.08MB/s
    Softlayer, Frankfurt, DE        159.122.69.4            59.1MB/s
    Softlayer, Singapore, SG        119.81.28.170           9.36MB/s
    Softlayer, HongKong, CN         119.81.130.170          6.23MB/s
    ----------------------------------------------------------------------
    Node Name                       IPv6 address            Download Speed
    Linode, Atlanta, GA             2600:3c02::4b           12.5MB/s
    Linode, Dallas, TX              2600:3c00::4b           10.9MB/s
    Linode, Newark, NJ              2600:3c03::4b           13.9MB/s
    Linode, Singapore, SG           2400:8901::4b           5.36MB/s
    Linode, Tokyo, JP               2400:8900::4b           8.36MB/s
    Softlayer, San Jose, CA         2607:f0d0:2601:2a::4    4.21MB/s
    Softlayer, Washington, WA       2607:f0d0:3001:78::2    17.3MB/s
    Softlayer, Paris, FR            2a03:8180:1301:8::4     58.9MB/s
    Softlayer, Singapore, SG        2401:c900:1101:8::2     7.92MB/s
    Softlayer, Tokyo, JP            2401:c900:1001:16::4    3.15MB/s
    ----------------------------------------------------------------------
    
  • neikneik Member

    @pike, your results for the two dedi. cores look pretty good compared to my Netcup RS 1000 G8 which also has two dedi. cores.

    Single-Core: 4500 vs 3150 for Hetzner
    Multi-Core: 8300 vs 5500 for Hetzner

    Although the Hetzner CPU's clock 200mhz lower they perform significantly better than those on Netcup.

    Does anyone know which CPUs are used at Hetztner? Netcup uses Xeon Gold 6140.

    Thanked by 1vimalware
  • Netcup G8 root servers used to perform much better when first announced but not any more.
    I wonder how Hetzner dedi vCPU will perform when all cores of a node will be in (heavy) use...

    Thanked by 2vimalware willie
Sign In or Register to comment.