Howdy, Stranger!

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


VPS Benchmark Compare Upcloud vs DigitalOcean vs Linode vs Vultr vs Hetzner
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.

VPS Benchmark Compare Upcloud vs DigitalOcean vs Linode vs Vultr vs Hetzner

Thought I'd share my latest 13-way KVM VPS provider benchmark comparison tests for Upcloud vs DigitalOcean vs Linode vs Vultr vs Hetzner https://community.centminmod.com/threads/17742/. Only criteria I used are

  • needs to be web host I use regularly
  • needs to have hourly billing

enjoy :)

Comments

  • donlidonli Member

    Conclusion: Get the Hetzner 4.9 Euro/month one.

  • Nice comparison. Thanks!

  • williewillie Member
    edited June 2019

    Interesting I guess. Remarks:

    • I find the graphical tables hard to read and it would be nice to have plain text, with everything in one table.

    • Geekbench seems to be a useless cpu benchmark and I don't understand its popularity. Results are all over the place on cpus whose real world application performance is similar. Passmark seems much more predictive of real world.

    • The key question for non-dedicated cpu vps is what happens if you burn cpu nonstop for days. I think it is not very friendly to benchmark that though. Some of my compute workload does look like that, so I use dedis for it.

    • I'd have liked to see some dedis in the comparison. I can give you access to some of mine if you want to try cpu tests on them, but the are all HDD based rather than SSD, so likely to do poorly on OLTP tests.

    • Actually, OLTP tests using ramdisk might have been interesting.

    • Maybe it's just me but I've always found concern about VPS performance on LET to be misplaced, given how much resource overallocation there is on the affordable ones. VPS as I think of them are for low-intensity services, disk storage, and a little bit of computation now and then. For anything more intensive I always prefer dedis.

    Thanked by 1uptime
  • eva2000eva2000 Veteran
    edited June 2019

    donli said: Conclusion: Get the Hetzner 4.9 Euro/month one.

    definitely can make that argument if performance to price ratio is the criteria used and you're on a budget and geographical location is suited

    MGarbis said: Nice comparison. Thanks!

    thanks

    willie said: I find the graphical tables hard to read and it would be nice to have plain text, with everything in one table.

    yeah hard to do for forums and amount of VPS servers being tabulated

    willie said: Geekbench seems to be a useless cpu benchmark and I don't understand its popularity. Results are all over the place on cpus whose real world application performance is similar. Passmark seems much more predictive of real world.

    Yeah i don't place too much stock in Geekbench or passmark, just threw it in for folks that seem to have interest heh

    willie said: The key question for non-dedicated cpu vps is what happens if you burn cpu nonstop for days. I think it is not very friendly to benchmark that though. Some of my compute workload does look like that, so I use dedis for it.

    true long term persistent load capabilities is one thing I'd like to look at if I had the $$$ budget for such heh. I think that's where in theory DigitalOcean cpu compute & general purpose, Linode dedicated cpu and Hetzner dedicated cpu named VPS plans should shine more.

    willie said: I'd have liked to see some dedis in the comparison. I can give you access to some of mine if you want to try cpu tests on them, but the are all HDD based rather than SSD, so likely to do poorly on OLTP tests.

    my benchmarks in this test can be replicated by anyone using CentOS 7 virgin server and my installnbench2.sh script as outlined at https://community.centminmod.com/threads/13-way-vps-server-benchmark-comparison-tests-discussion-thread.17744/#post-75077 so you can run on bare metal dedicated or vps server of your choice to see how your own servers stack up too :)

    though since the benchmark tests were done there's been a new CentOS linux kernel update for TCP SACK vulnerability so it wouldn't be 100% same OS wise but close I suspect.

    willie said: Actually, OLTP tests using ramdisk might have been interesting.

    sysbench tests can also be replicated using my sysbench.sh script https://github.com/centminmod/centminmod-sysbench - should work on CentOS and Ubuntu tested up to 16.x though the MySQL version you use and how it's configured setting wise would also factor into your own results

    willie said: Maybe it's just me but I've always found concern about VPS performance on LET to be misplaced, given how much resource overallocation there is on the affordable ones. VPS as I think of them are for low-intensity services, disk storage, and a little bit of computation now and then. For anything more intensive I always prefer dedis.

    Indeed it's fine line to balance usage requirements and price/affordability. But that's somewhat where benchmark testing is required to sort out which is best for your own performance and budget requirements i.e. if you can pay the same price for a different VPS plan/host but get 2-3x times the performance or pay 1/2 the price and get the same performance. Wouldn't you want to know about that ? :)

    Thanked by 2vimalware uptime
  • vimalwarevimalware Member
    edited June 2019

    Thank you @eva2000 .
    I tried to look for a way for my tightwad ass to thank you.
    In the past, I've found your compression benchmark nerdbait, very helpful.

    I'll use your Upcloud affiliate link when I give them a try.

    LET cheapos: Look at the right sidebar labelled 'Cheap VPS Hosting' . (edit: on the linked thread obviously)
    Support this prolific contributor to human knowledge , when you try a host there (for the first time obviously)

    Thanked by 1eva2000
  • eva2000eva2000 Veteran

    vimalware said: I tried to look for a way for my tightwad ass to thank you.

    In the past, I've found your compression benchmark nerdbait, very helpful.

    I'll use your Upcloud affiliate link when I give them a try.

    cheers much appreciated :)

  • uptimeuptime Member

    @eva2000 said:
    in theory DigitalOcean cpu compute & general purpose, Linode dedicated cpu and Hetzner dedicated cpu named VPS plans should shine more.

    also might consider the "compute intensive" option from Lunanode

  • williewillie Member

    eva2000 said: if you can pay the same price for a different VPS plan/host but get 2-3x times the performance or pay 1/2 the price and get the same performance. Wouldn't you want to know about that ?

    Thinking about it, truthfully I haven't cared. Dedicated-core VPS are so expensive compared to dedis that I haven't paid much attention to them. Shared-core VPS's can only get good burst CPU if overall utilization is low, which means they can't let CPU hogs run for too long, though having big multicore CPUs helps with that. And most of the stuff I run isn't that iops intensive (couldn't imagine running a serious database on a low end VPS).

    So the main VPS spec (other than location) that I look for if I compare them at all is $/TB for file storage, or SSD for non-storage. Otherwise it's more a matter of comfort level with the provider and basic reliability rather than performance. Things might be different if I didn't have a couple of long term dedis to throw compute tasks at, but I do like that setup.

  • So which is good?

  • @cybertech said:
    So which is good?

    New Vultr high frequency compute smashed em all.

    Thanked by 1cybertech
  • cybertech said: So which is good?

    Upcloud and vultr are the best, but hetzner provides the best value (but only in EU).

  • eva2000eva2000 Veteran

    smallbibi said: Upcloud and vultr are the best, but hetzner provides the best value (but only in EU).

    Pretty much based on their respective 2 cpu core plans.

  • SplitIceSplitIce Member, Host Rep
    edited June 2019

    Quite interesting indeed. Digitaloceans performance was kind of disapointing.

    @eva2000 I understand this is alot of work to put togeather but I would love to see this done at the low end (on the $5 instances).

  • eva2000eva2000 Veteran

    SplitIce said: I understand this is alot of work to put togeather but I would love to see this done at the low end (on the $5 instances).

    while I may not have free time, folks can replicate my tests on their own servers or VPSes too, quoting myself :)

    eva2000 said: my benchmarks in this test can be replicated by anyone using CentOS 7 virgin server and my installnbench2.sh script as outlined at https://community.centminmod.com/threads/13-way-vps-server-benchmark-comparison-tests-discussion-thread.17744/#post-75077 so you can run on bare metal dedicated or vps server of your choice to see how your own servers stack up too

    so it's quite easy to do own $5 instance tests too or run on full blown bare metal dedicated servers too :)

  • SplitIceSplitIce Member, Host Rep

    @eva2000 if I had the time I'd totally ansible it to automate the entire framework :)

    But yeah well aware that time is short. So very short of late.

    Thanked by 1eva2000
  • eva2000eva2000 Veteran

    SplitIce said: But yeah well aware that time is short. So very short of late.

    Yeah being time poor ^_^

  • @Hetzner_OL never fails to surprise me when it comes to bang for the buck!

  • desperanddesperand Member
    edited June 2019

    Graphomania. Very hard to read.
    Too much water. Really too much water.

    Sorry mate, I respect what you do, and reading your forum, and GitHub, and sources, and really enjoy it. But try to shrink duplicate things or hide some details, they are not representative, because very hard to compare or not even needed at all to write about them. Or maybe need to split to smaller things which easy to understand. Anyway, thank you for the hard work that you did within the table.

  • SplitIce said: Quite interesting indeed. Digitaloceans performance was kind of disapointing.

    They have always been like that. Always, since opening.

  • hjlowhjlow Member

    does anyone know the eta for vultr high compute instances in eu? so far only in US I think

  • ITLabsITLabs Member

    @hjlow said:
    does anyone know the eta for vultr high compute instances in eu? so far only in US I think

    "High Frequency compute is initially available in New Jersey and will be available worldwide later this summer!" - https://www.vultr.com/news/Now-Available---High-Frequency-Compute-and-a-Brand-New-Look/

    Thanked by 1hjlow
  • eva2000eva2000 Veteran

    @desperand said:
    Graphomania. Very hard to read.
    Too much water. Really too much water.

    Sorry mate, I respect what you do, and reading your forum, and GitHub, and sources, and really enjoy it. But try to shrink duplicate things or hide some details, they are not representative, because very hard to compare or not even needed at all to write about them. Or maybe need to split to smaller things which easy to understand. Anyway, thank you for the hard work that you did within the table.

    It's just that 13x VPS servers benchmarked will result in alot of details and data/results to digest. Guess it isn't for everyone - I personally like as much detail as I can consume when I read benchmarks/reviews so I can make up my own mind.

    hjlow said: does anyone know the eta for vultr high compute instances in eu? so far only in US I think

    no ETA, but Vultr on twitter said they're just waiting on server shipments to other datacenters to rack them up and deploy - so sounds like it's in progress.

    as to cpus used on Vultr high frequency line, in my above benchmark I made the suggestion they might be using Intel Xeon E-21xx Coffee Lake series

    Vultr High Frequency VPS servers seem to be clocked at ~3.80Ghz based and Centmin Mod's Nginx GCC 8.2.1 compiler is reporting cpu as Skylake march target and consulting Intel Ark database, there doesn't seem to be any Xeon Scalable Skylake cpus with 3.80Ghz base frequency. So it's likely it's the newer Intel Xeon workstation based Xeon E-2100 series Intel Coffee Lake cpus or the new 2nd generation Intel Xeon Scalable Cascade Lake cpus which do have 3.80Ghz base cpu clock frequency and since GCC 8.2 is older compiler it may not see Coffee Lake or Cascade Lake cpu target - I will confirm when I test GCC 9 compiler for Centmin Mod Nginx builds later on. Possible cpu models meeting this criteria would be Intel Cascade Lake based Xeon E-2244G and Xeon E-2276G, Xeon Gold 5222, Xeon Platinum 8256 and Intel Coffee Lake based, Intel Xeon E-2174G, Xeon E-2186G

    and it could be, one of my Centmin Mod users posted install times on Hostdoc NVMe VPS with 3x cpu E-2136 Coffee Lake and the times were very fast ~667 seconds https://community.centminmod.com/threads/post-your-centmin-mod-123-09beta01-install-time-stats.8866/page-16#post-75430 so could what Vultr is using ?

    Thanked by 1hjlow
  • williewillie Member
    edited June 2019

    The Vultr high frequency compute measurements are quite impressive for cheap vps. I wonder what overutilization and throttling will look like, since the product description is sure to attract cpu hogs (I'm one of course). Hey @Hetzner_OL you need a product like that ;-).

    Thanked by 1eva2000
  • eva2000eva2000 Veteran

    willie said: The Vultr high frequency compute measurements are quite impressive for cheap vps.

    indeed will be interesting to see long term stability for vultr high frequency compute :)

    The Intel Xeon E-2xxx Coffee Lake seems to be a beast especially for Nginx HTTP/2 HTTPS performance. One of my Centmin Mod users posted dedicated OVH Advance Xeon E-2136 Nginx HTTP/2 HTTPS benchmark numbers at https://community.centminmod.com/threads/post-share-your-centmin-mod-nginx-http-2-https-benchmarks.14832/#post-73159 and they're at least 2-2.5x times faster than any dedicated server I have benchmarked or seen benchmarked so far !

    Looking forward to when Ryzen 39xx Zen2 cpus arrive to see how they fair and also looking forward to seeing how VPS providers evolve over next 12-24 months cpu wise - so much power in VPS compared to even dedicated servers 8-15yrs ago !

    willie said: Hey @Hetzner_OL you need a product like that ;-)

    I'd benchmark/use that too :)

  • SplitIceSplitIce Member, Host Rep

    I've got a Vultr High Frequency running a software platform designed for a low powered ARM (for the purposes of fast testing) at the moment. Damn is it fast. 2x cores performing 10x faster than the quad core arm (of course everything is compiled for x64 vs armv7).

    Honestly though I think the plans are weighted too much towards storage and ram though. I want more of those fast cores, less of that ram and storage. I want to process!

    Thanked by 2eva2000 uptime
Sign In or Register to comment.