Howdy, Stranger!

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


My first post + benchmark of Hetzner EX62-NVMe (Falkenstein)
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.

My first post + benchmark of Hetzner EX62-NVMe (Falkenstein)

angeliusangelius Member
edited May 2019 in General

Hey Guys, I'm reading LET (without an account) since 1 year. Today I decided it was time for me to join and start taking part to the discussions ... and this is my first post :)

Few words about me, I'm a full-stack developer and also an entrepreneur.

I just took a few days ago a new dedicated server from Hetzner "EX62-NVMe"
I have deployed there LXD VM on it (IPV6 only), usage will be REDIS + DB hosting over wireguard for customers

I wanted to share with you a few bench results about this server, I was looking for it and was not able to find anything.

Hardware: Intel® Core™ i9-9900K Octa-Core (16 cores with HT)

RAM: 64 GB DDR4
Hard drive: 2 x 1 TB NVMe SSD (software RAID 1)
Connection: 1 GBit/s port - Guaranteed bandwidth: 1 GBit/s
Traffic: Unlimited

REDIS: redis-benchmark -t set -n 1000000 -r 1000000;

====== SET ======
1000000 requests completed in 4.34 seconds
50 parallel clients
3 bytes payload
keep alive: 1

99.96% <= 1 milliseconds
99.97% <= 2 milliseconds
100.00% <= 2 milliseconds
230520.98 requests per second

NGINX: Nearly 300K Requests per second!

wrk -t32 -c1024 -d5s http://localhost/
Running 5s test @ http://localhost/
32 threads and 1024 connections
Thread Stats Avg Stdev Max +/- Stdev
Latency 3.74ms 3.57ms 208.12ms 88.31%
Req/Sec 9.38k 3.77k 42.43k 77.63%
1523314 requests in 5.10s, 1.22GB read
Socket errors: connect 35, read 0, write 0, timeout 0
Requests/sec: 298829.76
Transfer/sec: 244.80MB

Node.js v12 on single core (44K RPS)

wrk -t1 -c16 -d5s http://localhost:8080/
Running 5s test @ http://localhost:8080/
1 threads and 16 connections
Thread Stats Avg Stdev Max +/- Stdev
Latency 363.23us 126.07us 10.01ms 99.26%
Req/Sec 44.54k 842.83 45.30k 98.00%
221615 requests in 5.00s, 32.34MB read
Requests/sec: 44306.35
Transfer/sec: 6.46MB

Node.js v12 + Fastify + PM2 cluster mode with 16 cores (352K RPS!)

wrk -t16 -c512 -d5s http://localhost:8080/
Running 5s test @ http://localhost:8080/
16 threads and 512 connections
Thread Stats Avg Stdev Max +/- Stdev
Latency 1.83ms 1.92ms 26.84ms 86.44%
Req/Sec 21.87k 11.30k 87.75k 63.75%
1795193 requests in 5.10s, 257.91MB read
Requests/sec: 352571.18
Transfer/sec: 51.44MB

Geekbench4: single core: 6629 / multi-core: 36920

https://browser.geekbench.com/v4/cpu/13096097

bench.sh:

CPU model : Intel(R) Core(TM) i9-9900K CPU @ 3.60GHz
Number of cores : 16
CPU frequency : 4368.783 MHz
Total size of Disk : 922.7 GB (8.0 GB Used)
Total amount of Mem : 64331 MB (3663 MB Used)
Total amount of Swap : 16367 MB (0 MB Used)
System uptime : 0 days, 21 hour 47 min
Load average : 0.00, 0.13, 0.24
OS : Ubuntu 18.04.2 LTS
Arch : x86_64 (64 Bit)
Kernel : 4.15.0-48-generic

I/O speed(1st run) : 1.3 GB/s
I/O speed(2nd run) : 1.4 GB/s
I/O speed(3rd run) : 1.4 GB/s
Average I/O speed : 1399.5 MB/s

Node Name IPv4 address Download Speed
CacheFly 205.234.175.175 108MB/s
Linode, Tokyo, JP 106.187.96.148 8.72MB/s
Linode, Singapore, SG 139.162.23.4 9.43MB/s
Linode, London, UK 176.58.107.39 73.2MB/s
Linode, Frankfurt, DE 139.162.130.8 108MB/s
Linode, Fremont, CA 50.116.14.9 13.6MB/s
Softlayer, Dallas, TX 173.192.68.18 10.7MB/s
Softlayer, Seattle, WA 67.228.112.250 12.2MB/s
Softlayer, Frankfurt, DE 159.122.69.4 95.1MB/s
Softlayer, Singapore, SG 119.81.28.170 10.4MB/s
Softlayer, HongKong, CN 119.81.130.170 8.66MB/s

Node Name IPv6 address Download Speed
Linode, Atlanta, GA 2600:3c02::4b 16.5MB/s
Linode, Dallas, TX 2600:3c00::4b 17.0MB/s
Linode, Newark, NJ 2600:3c03::4b 28.1MB/s
Linode, Singapore, SG 2400:8901::4b 13.3MB/s
Linode, Tokyo, JP 2400:8900::4b 8.57MB/s
Softlayer, San Jose, CA 2607:f0d0:2601:2a::4 9.55MB/s

Right now it's idling really hard ... htop in FullHD ^^

htop in FullHD

What would you do with that beast? :)

Comments

  • YmpkerYmpker Member
    edited May 2019

    Idling is the only correct answer.

  • williewillie Member

    The geekbench is impressive, don't know about the redis result, but I thought redis was single threaded.

  • JackHJackH Member
    edited May 2019

    Welcome to LET! Nice to see a decent quality first post that isn't shilling. Hope we do our best to drive you away make you feel welcome! :-)

    Thanked by 1angelius
  • @willie said:
    The geekbench is impressive, don't know about the redis result, but I thought redis was single threaded.

    Indeed, this result is using only 1 vcore, when using the 16 vcore (with 16 redis instances) the result is ~ 2.6M RPS (WOW!!)

  • nice pic of yah..

    Thanked by 1angelius
  • angeliusangelius Member
    edited May 2019

    @JackH said:
    Welcome to LET! Nice to see a decent quality first post that isn't shilling. Hope we do our best to drive you away make you feel welcome! :-)

    Thanks for this great community and cold warm welcome :)

  • eva2000eva2000 Veteran

    angelius said: What would you do with that beast?

    Very nice. I'd make it my GCC/Clang compiler & Centmin Mod test server. I'm actually deciding if I want to give up my existing OVH Core i7 4790K 4C/8T test server for this Hetzner Core i9 9900K due to having double number of cpu core/threads which beneficial for source compilation and my multi-threaded testing for backups/compression algorithms.

    Thanked by 1angelius
  • MikeAMikeA Member, Patron Provider

    @eva2000 said:

    angelius said: What would you do with that beast?

    Very nice. I'd make it my GCC/Clang compiler & Centmin Mod test server. I'm actually deciding if I want to give up my existing OVH Core i7 4790K 4C/8T test server for this Hetzner Core i9 9900K due to having double number of cpu core/threads which beneficial for source compilation and my multi-threaded testing for backups/compression algorithms.

    i9-9900K is great. The Hetzner one is so close to OVH's i7's so may as well switch to it. The i9's at OVH won't be similar priced to Hetzner either.

    Thanked by 2webcraft eva2000
  • williewillie Member
    edited May 2019

    This cpu seems hard to beat at the price. I'd be interested in comparison with Ryzen 2700X but I think the EX62 would win. Even the ridiculous 18 core i9 isn't much faster, because of its lower clock. And the hdd EX52 and EX62 versions have 2x8TB! The auction servers with 2x3TB don't look as good as they used to, especially if you also want cpu.

    Thanked by 2HostDoc eva2000
  • poissonpoisson Member

    @angelius said:
    Right now it's idling really hard ... htop in FullHD ^^

    Excellent. First post and you are already a certified LET hard idler.

    Thanked by 1angelius
  • alvinalvin Member

    great test... it seems i9-9900K better than 2700x

    Thanked by 1angelius
  • mrclownmrclown Member

    Good benchmark.

    Thanked by 1angelius
  • eva2000eva2000 Veteran

    MikeA said: i9-9900K is great. The Hetzner one is so close to OVH's i7's so may as well switch to it. The i9's at OVH won't be similar priced to Hetzner either.

    Indeed, surprised by Hetzner i9 9900K price as it's only ~$40 more than the price I am paying for me OVH i7 4790K i got on black friday for US$47/month way back :)

  • williewillie Member

    Yeah, and the EX52 is 10 euro less than the EX62 when the only difference is the 6 core cpu instead of 8 core. The EX52 is looking good. The whole EX series has 64GB of ram now so it would be great to see versions with less ram but more disk. I have a 32GB server now and really never do anything with it that makes good use of so much memory. 16GB would be fine for me, 64GB even worse overkill than 32GB.

  • jsgjsg Member, Resident Benchmarker

    @angelius

    Welcome - and: well done!

    Thanked by 1angelius
  • vfusevfuse Member, Host Rep

    We have a couple of them running the price/performance is as good as it gets. Very fast single thread performance as well.

    Thanked by 1vimalware
  • edited May 2019

    Hi, how do you setup your node test setups? Could you share using github?

  • If only it could run Debian 9 out of the box...the Kernel doesn't support it.
    Anyone managed to do so ?

  • Jona4sJona4s Member
    edited May 2019

    Hmm, well you did the wrk bench on the same machine. Thats cheating : P

    Doing it locally with a 16 core / 32HT I get 6 million req/sec, using compiled Golang.

    Running 1m test @ http://localhost:8080/
    20 threads and 256 connections
    Latency 6.36ms 4.34ms 56.30ms 66.95%
    Req/Sec 305.51k 35.46k 508.70k 74.70%
    96901968 requests in 15.99s, 8.39GB read
    Requests/sec: 6059823.31
    Transfer/sec: 537.46MB

    If done across networks, I get around ~1M

  • joepie91joepie91 Member, Patron Provider

    angelius said: Node.js v12 on single core (44K RPS)

    What benchmark is that?

  • waphirewaphire Member
    edited May 2019

    @mehargags said:
    If only it could run Debian 9 out of the box...the Kernel doesn't support it.
    Anyone managed to do so ?

    Debian 9 does run OOTB but the included kernel doesn't support the i9's iGPU. The iGPU is supported 4.20 onwards.

    Thanked by 1vimalware
  • FalzoFalzo Member

    @waphire said:

    @mehargags said:
    If only it could run Debian 9 out of the box...the Kernel doesn't support it.
    Anyone managed to do so ?

    Debian 9 does run OOTB but the included kernel doesn't support the i9's iGPU. The iGPU is supported 4.20 onwards.

    it's been reported to miss the neccessary network drivers/firmware as well for the boards hetzner uses (unless you use backports)... maybe meanwhile that has changed (on one end or the other) ;-)

    Thanked by 1waphire
  • Ref: https://helgeklein.com/blog/2019/04/windows-server-2019-hetzner-ex62-ax100-dedicated-servers/

    EX62: Motherboard is probably Gigabyte B360HD3PLM-CF (U3E1) which is essentially intel B360 Chipset

    Ref: https://forums.linuxmint.com/viewtopic.php?t=271239
    8th gen needs at least 4.13 kernel

    Thats the problem with Desktop grade hardware... Intel just didn't feel like providing Linux drivers for it at all.

    I wouldn't feel safe to put this server in production use if something so important like "Network" is "hacked" to work.

    I'd rather wait and have good solid testimonials that Linux kernel and drivers work for this chipset for a few months.

  • @greattomeetyou said:
    Hi, how do you setup your node test setups? Could you share using github?

    Here you go, I've just published the source (very basic) here:
    https://github.com/jbenguira/node-perf-test

    or you can run it directly on ubuntu/debian with this:
    wget -O - https://raw.githubusercontent.com/jbenguira/node-perf-test/master/start-test.sh | bash

  • @joepie91 said:

    angelius said: Node.js v12 on single core (44K RPS)

    What benchmark is that?

    source published here:
    https://github.com/jbenguira/node-perf-test

    Thanked by 1joepie91
  • @Jona4s said:
    Hmm, well you did the wrk bench on the same machine. Thats cheating : P

    Doing it locally with a 16 core / 32HT I get 6 million req/sec, using compiled Golang.

    Running 1m test @ http://localhost:8080/
    20 threads and 256 connections
    Latency 6.36ms 4.34ms 56.30ms 66.95%
    Req/Sec 305.51k 35.46k 508.70k 74.70%
    96901968 requests in 15.99s, 8.39GB read
    Requests/sec: 6059823.31
    Transfer/sec: 537.46MB

    If done across networks, I get around ~1M

    Not really cheating, it's just not measuring the same thing ;)
    About your go bench above, Wow! Would you mind to share on github your source?

Sign In or Register to comment.