All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
Contabo V*D*S benchmark and review - special
Front up: this is a special in more than one regard. Look at the insanely awesome disk performance! That's special no. 1.
A while ago Contabo contacted me and offered me access to a VDS, the "small" one ("VDS S") with 6 vCores. Hint/warning: their web site mentions "3 physical cores" of a 16-core AMD Epyc but in the usual hosting parlance one would mention vCores, so don't be mistaken. Now the second special point: the list price is €36.99/mo plus a €19.99 one time setup fee which falls away with an annual contract. "Wut?" I hear some ask, and you are right, this is way beyond $10/mo - but so is what you get. Let me explain. Like most of us I'm focused on the best quality within a certain (rather low) price frame, but at the same time like probably more than just a few of us I also need a couple of really reliable iron for business, projects, etc, and of course the first place I look at is LET.
Second, compare the VDS to a dedi and you'll see why (real) VDS can be quite attractive although they seem to be expensive.
For example I have a Xeon (older model) based dedi with 8 GB memory and an SSD that costs about the same as the Contabo VDS which has more memory, has a significantly higher performance, both single core and in total, and an insanely fast NVMe. Now, granted I'm a dedi freak but looking at what I get for my money with a dedi or a good quality Zen based VDS makes me think (and no, as long as I get physical cores Spectre & friends are no dedi advantage over the VDS).
For those of you who are still here, let's go at it:
6 dedicated Epyc vCores, 24(!) GB memory, and a 180 GB (insanely fast) NVMe based disk.Review based on over 500 benchmark runs.
Machine: amd64, Arch.: amd64, Model: AMD EPYC 7282 16-Core Processor
OS, version: FreeBSD 12.2, Mem.: 23.982 GB
CPU - Cores: 6, Family/Model/Stepping: 23/49/0
Cache: 64K/64K L1d/L1i, 512K L2, 16M L3
Std. Flags: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat
pse36 cflsh mmx fxsr sse sse2 htt sse3 pclmulqdq ssse3 fma cx16
sse4_1 sse4_2 popcnt aes xsave osxsave avx f16c rdrnd hypervisor
Ext. Flags: syscall nx mmxext fxsr_opt pdpe1gb rdtscp lm lahf_lm cmp_legacy svm
cr8_legacy lzcnt sse4a misalignsse 3dnowprefetch osvw topoext
perfctr_core
----- Processor and Memory -----
ProcMem SC [MB/s]: avg 294.8 - min 291.7 (99.0 %), max 295.4 (100.2 %)
ProcMem MC [MB/s]: avg 939.4 - min 844.8 (89.9 %), max 1050.0 (111.8 %)
AES, hypervisor, etc. flags are available and the performance is, well, what one expects from an AMD Epyc, duh. And an amount of memory that frankly in my books belongs in dedi territory, not in virtual server territory. But it seems Contabo is dead serious about the 'D' in VDS ...
Also note the low spread. Less than 1% spread in single core confirms that this is a VDS (don't worry about the higher but still rather low spread in multi core; you'll see that even with a dedi)
Now on to the disk ... but be sure to fasten your seat belts!
--- Disk - Buffered ---
Write seq. [MB/s]: avg 3760.94 - min 2911.87 (77.4%), max 3884.18 (103.3%)
Write rnd. [MB/s]: avg 6196.52 - min 5009.81 (80.8%), max 6912.47 (111.6%)
Read seq. [MB/s]: avg 4672.86 - min 3771.61 (80.7%), max 5067.30 (108.4%)
Read rnd. [MB/s]: avg 8723.46 - min 7932.94 (90.9%), max 9775.28 (112.1%)
--- Disk - Sync/Direct ---
Write seq. [MB/s]: avg 1138.40 - min 742.56 (65.2%), max 1270.34 (111.6%)
Write rnd. [MB/s]: avg 1310.31 - min 655.94 (50.1%), max 1394.64 (106.4%)
Read seq. [MB/s]: avg 4508.40 - min 3019.81 (67.0%), max 4678.94 (103.8%)
Read rnd. [MB/s]: avg 9493.79 - min 7534.01 (79.4%), max 9813.65 (103.4%)
That's just insane! When I did what I always do after some first couple of runs that is, when I looked at the results I was sure that something was wrong with my software and quickly did a test run on my own rig with a Cardea PCIe 4 NVMe (the fastest thing I had in my hands so far). Nope. my software worked fine.
FYI: The fastest disk (NVMe) results I had seen so far with any kind of virtual server was in the 150 - 200 MB/s range with unbuffered sync sequential writing. This VDS's NVMe crosses over into 4 digits(!) territory! And I really rode it hard that disk and played almost dirtily with the parameters of my benchmark.
If you have a DB heavy project go for a Contabo VDS!. Bloody incredible that speed.
Let's look at the connectivity before I get horny again ...
--- Network ---
US LAX lax.download.datapacket.com [F: 0]
DL [Mb/s]: avg 42.1 - min 36.1 (85.7%), max 225.7 (535.7%)
Ping [ms]: avg 154.2 - min 12.6 (8.2%), max 159.6 (103.5%)
Web ping [ms]: avg 155.2 - min 12.7 (8.2%), max 160.7 (103.5%)
NO OSL speedtest.osl01.softlayer.com [F: 0]
DL [Mb/s]: avg 132.2 - min 78.7 (59.5%), max 157.3 (119.0%)
Ping [ms]: avg 39.2 - min 34.6 (88.2%), max 39.4 (100.5%)
Web ping [ms]: avg 39.3 - min 34.6 (88.1%), max 39.5 (100.6%)
US SJC speedtest.sjc01.softlayer.com [F: 0]
DL [Mb/s]: avg 32.5 - min 0.0 (0.0%), max 41.0 (126.3%)
Ping [ms]: avg 160.3 - min 0.0 (0.0%), max 164.4 (102.6%)
Web ping [ms]: avg 162.4 - min 0.0 (0.0%), max 170.8 (105.2%)
AU MEL speedtest.c1.mel1.dediserve.com [F: 0]
DL [Mb/s]: avg 24.8 - min 22.5 (90.7%), max 27.6 (111.1%)
Ping [ms]: avg 261.4 - min 0.0 (0.0%), max 264.3 (101.1%)
Web ping [ms]: avg 261.5 - min 0.0 (0.0%), max 266.8 (102.0%)
JP TOK speedtest.tokyo2.linode.com [F: 0]
DL [Mb/s]: avg 22.3 - min 0.0 (0.0%), max 27.6 (123.5%)
Ping [ms]: avg 267.1 - min 0.0 (0.0%), max 272.5 (102.0%)
Web ping [ms]: avg 269.4 - min 239.9 (89.1%), max 276.3 (102.6%)
IT MIL speedtest.mil01.softlayer.com [F: 0]
DL [Mb/s]: avg 210.7 - min 174.6 (82.8%), max 224.3 (106.4%)
Ping [ms]: avg 16.0 - min 0.0 (0.0%), max 17.5 (109.6%)
Web ping [ms]: avg 16.1 - min 15.2 (94.3%), max 18.0 (111.6%)
FR PAR speedtest.par01.softlayer.com [F: 0]
DL [Mb/s]: avg 214.8 - min 203.3 (94.7%), max 224.4 (104.5%)
Ping [ms]: avg 13.8 - min 0.0 (0.0%), max 15.7 (113.5%)
Web ping [ms]: avg 15.1 - min 12.9 (85.2%), max 49.6 (327.7%)
SG SGP mirror.sg.leaseweb.net [F: 0]
DL [Mb/s]: avg 21.0 - min 0.0 (0.0%), max 26.6 (126.8%)
Ping [ms]: avg 278.0 - min 0.0 (0.0%), max 354.7 (127.6%)
Web ping [ms]: avg 288.9 - min 0.0 (0.0%), max 354.7 (122.8%)
BR SAO speedtest.sao01.softlayer.com [F: 0]
DL [Mb/s]: avg 29.3 - min 23.0 (78.6%), max 31.7 (108.1%)
Ping [ms]: avg 203.5 - min 201.0 (98.8%), max 203.6 (100.1%)
Web ping [ms]: avg 203.5 - min 201.0 (98.8%), max 203.9 (100.2%)
IN CHN speedtest.che01.softlayer.com [F: 0]
DL [Mb/s]: avg 36.1 - min 26.2 (72.5%), max 41.4 (114.8%)
Ping [ms]: avg 158.1 - min 150.7 (95.3%), max 160.9 (101.8%)
Web ping [ms]: avg 160.7 - min 154.4 (96.1%), max 166.3 (103.5%)
GR UNK speedtest.ftp.otenet.gr [F: 0]
DL [Mb/s]: avg 80.1 - min 0.0 (0.0%), max 135.4 (169.0%)
Ping [ms]: avg 33.2 - min 0.0 (0.0%), max 51.8 (156.1%)
Web ping [ms]: avg 33.2 - min 0.0 (0.0%), max 52.1 (156.9%)
US WDC mirror.wdc1.us.leaseweb.net [F: 0]
DL [Mb/s]: avg 61.5 - min 44.8 (72.9%), max 68.7 (111.8%)
Ping [ms]: avg 99.3 - min 90.7 (91.4%), max 99.4 (100.1%)
Web ping [ms]: avg 99.8 - min 91.3 (91.5%), max 100.5 (100.7%)
RU MOS speedtest.hostkey.ru [F: 0]
DL [Mb/s]: avg 121.3 - min 107.2 (88.4%), max 136.2 (112.3%)
Ping [ms]: avg 44.2 - min 43.7 (99.0%), max 48.0 (108.7%)
Web ping [ms]: avg 48.2 - min 43.8 (90.8%), max 53.5 (111.0%)
US DAL speedtest.dal05.softlayer.com [F: 0]
DL [Mb/s]: avg 48.6 - min 45.8 (94.3%), max 51.6 (106.1%)
Ping [ms]: avg 123.3 - min 121.1 (98.2%), max 123.5 (100.2%)
Web ping [ms]: avg 125.4 - min 121.4 (96.8%), max 128.7 (102.6%)
UK LON speedtest.lon02.softlayer.com [F: 0]
DL [Mb/s]: avg 186.8 - min 178.9 (95.7%), max 196.6 (105.2%)
Ping [ms]: avg 22.5 - min 0.0 (0.0%), max 22.9 (101.6%)
Web ping [ms]: avg 23.5 - min 22.4 (95.3%), max 26.1 (111.0%)
US NYC nyc.download.datapacket.com [F: 0]
DL [Mb/s]: avg 65.7 - min 55.5 (84.4%), max 72.0 (109.5%)
Ping [ms]: avg 95.4 - min 0.0 (0.0%), max 97.1 (101.8%)
Web ping [ms]: avg 97.2 - min 92.4 (95.1%), max 108.7 (111.8%)
RO BUC 185.183.99.8 [F: 0]
DL [Mb/s]: avg 139.6 - min 77.7 (55.6%), max 161.0 (115.3%)
Ping [ms]: avg 39.3 - min 34.4 (87.6%), max 39.5 (100.6%)
Web ping [ms]: avg 39.7 - min 34.6 (87.2%), max 55.1 (138.8%)
CN_HK mirror.hk.leaseweb.net [F: 0]
DL [Mb/s]: avg 21.8 - min 0.0 (0.0%), max 25.8 (118.5%)
Ping [ms]: avg 266.1 - min 0.0 (0.0%), max 323.1 (121.4%)
Web ping [ms]: avg 270.8 - min 0.0 (0.0%), max 324.7 (119.9%)
DE FRA fra.lg.core-backbone.com [F: 0]
DL [Mb/s]: avg 239.6 - min 238.0 (99.3%), max 241.2 (100.7%)
Ping [ms]: avg 3.9 - min 3.8 (96.2%), max 4.0 (101.3%)
Web ping [ms]: avg 4.0 - min 3.8 (95.1%), max 4.1 (102.7%)
Remember the (constructive) criticism I received recently about the failure quote? Well, note the '[F:x]' at the end of each target line! It tells you how often a target host failed (download not possible or extremely slow (< ~ 2 Mb/s)).
Or, to me more precise, how often a failure occured beyond 5% of all runs. Why? Because there was something I didn't like since quite a while that I fixed in the new version of my result compiler: The internet is a "living" thing; sometimes a connection (as well as a target server) is a bit faster, sometimes it's a bit slower, and sometimes it simply has PMS. I didn't like that an occasional and damn normal hiccup made the results of a provider look worse than they actually were. And then it dawned on me: How do carriers often calculate their price? They cut out the top 5%, the highest peaks. So what works fine for carriers should work fine for me too, right. From this benchmark on the network results cut out occasional failures - but not results that are sh_tty beyond occasionally (up to 5% of runs).
As for the results, not really much to say other than decent, really decent, even nice. Particularly positive: most results 'avg' value is relatively close to 'max', except for a few targets which I know (and sometimes even chose for that reason) to be capricious or difficult (very PMSy).
Oh and btw, I'm very pleased to have learned that Contabo closely looked at my (former) benchmarks and finetuned some screws, particularly wrt network routes. And it shows. Kudos to Contabo for really listening to us and valuing our LET community. And of course thank you Contabo for providing a machine to test to me! Well noted, I myself "chose" the node via the usual lottery. I got a normal customer account and ordered the test system like everyone else, the only difference being that afterwards they take out the invoice (and for those mistrusting: NO, I have no system with Contabo, nor - as usual - did I get anything for free, other than the test system).
TL;DR Very nice system. Price way outside LET range but an attractive dedi alternative with an insanely fast NVMe from a provider who actually listens.
Comments
Contabo has gotten some nice machines in the last couple years.
They've gotten lots of hate for no reason. They've brought new machines with extremely good hardware while keeping the prices almost the same. I've had a VPS from them for about 3 years, near never downtimes, and good and stable internet speeds.
Would love them to implement some Ryzen servers, bet they'll sell like hotcakes.
Thats pretty average. Even netcup rootservers from easter deals had better value and power. At this price point, I cant see why anyone want any of these instead of a Herzner AX41 NVMe, which costs like 34 euros and is fully dedicated.
That's BS.
For one you can't compare some promo with a standard (priced) product. Also I actually do have a netcup VDS ("root server") and am quite happy but netcup VDS and Contabo VDS (as tested) are different classes of product. Maybe formertimes Contabo was different and/or had different products and quality but the VDS I benchmarked for some weeks is a different class of product.
As for the AX41, does that give you a processor as fast as an Epyc with 24 GB decent RAM (not 2133 or the like) and an NVMe that is really as fast as the one benchmarked here? I doubt it.
What I'd agree with is that Contabo certainly isn't on the cheap side; maybe they should run some attractive promos from time to time. But then, on the other hand, they seem to do and sell well with their prices. There is life outside of LET, after all ...
I have an AX41NVMe, supplied with 2x512GB Samsung NVMes which I am using in raid0. The passmark of R5 3600 is under below:
https://www.cpubenchmark.net/cpu.php?cpu=AMD+Ryzen+5+3600&id=3481
While, the epyc one:
https://www.cpubenchmark.net/cpu.php?cpu=AMD+EPYC+7282&id=3625
The AX41NVMe server is truly a beast and unlimited without any throttling whatsoever. It's a no brainer at this price point.
Single thread 7782 passmark: 2006, total for 6 cores: ~12000.
Single thread R5 3600: 2500+, total: 17864.
If you are being objective and not biased, you will find hetzner to be a clear winner!
Edit: Just checked ram clock, it's 3200MHz. The SSDs are Samsung PM981.
And regarding netcup, the standard price of equivalent server to the one in your review is RS4000 G9 for 32 euros/month and 27euros/m if paid for multiple months. It has 6 core 7702 CPU, 32GB DDR4, 800GB SSD, 2.5Gbps network with 120TB bandwidth. It blows the specs of the contabo on paper!
And with frequent sales of netcup every now and then on rootservers, it offers much better value to the point I used it in place of hetzner dedi for sometime before I got tired of their QoS feature. It was wrong of me to use a shared space for dedicated purpose, but it almost worked!
Well, if that system is available at a similar price then it seems to be a very attractive alternative.
As I said myself
...
I don't know and can't test the SSD speed because my netcup VDS has spindles (whose speed btw is excellent, pretty much like SSDs).
Let me put it like this: There's a reason why I have a netcup VPS.
But still: I was offered the opportunity to do an extensive benchmark series of their VDS by Contabo and I think that benchmarks are always valuable for our community, no matter whether some product is expensive or not.
Contabo was very transparent and seriously interested in playing fair and open. For that I thank them.
My reviews can serve as a basis to get to know more about a product and to find particularly nice spots as well as weak ones. Whether one considers a benchmarked product for purchase or not is not my concern. My concern is to provide data points that can help you to make the best decision for yourself.
Also note that I never get any commission and have no interest in influencing your decision. What I do is done as a service to our community and I laud all providers supporting me by providing free test systems..
Does Contabo has a looking glass?
Been using Contabo for a few years now for some of my projects and they've been great. The only time I had issues was within the first couple of weeks I joined them, the network had issues for a day or two with some significant packets losses only in the afternoon but they fixed that quite fast. I had zero problems or downtime ever since and the performance is great for the price I am paying.
Worth mentioning that I also emailed them about SSD speed limits and they lifted the restrictions right away. This is something I did right when I joined them as I knew it's a common practice but they are mainly doing this to avoid abusive customers.
Thank you @jsg for the great review of our VDS, we really appreciate it! Hope some of you will give our VDS a try as well when in need of a really powerful machine and we are grateful for any suggestion you have about our services
@contabo_m can you share a public IP or looking glass?
Just choose one of the IPs from here: https://ipinfo.io/AS51167#blocks