Howdy, Stranger!

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

Sign In Register

Quick Links


Shells Virtual Desktop
BMail.ag - Secure Email Service
Server.net
CPLicense.net
VPS Server
Buy VPN
Vultr
VMs for AI
HostDare
ReliableSite White-Label Dedicated Hosting for Resellers
InterServer VPS
BMail.ag - Secure Email Service
Best VPN
High-Performance Bare Metal Server Solutions
Karvl.com
Server Mania Cloud Hosting
DataWagon Hosting
AlphaVPS Hosting
Evoxt.com
Clouvider
VPS Hosting with NVMe
Residential IPs in the US & 4G Mobile Proxies in EU & US with Unlimited Bandwidth
ReliableSite White-Label Dedicated Hosting for Resellers
Rabisu - Hosting Solutions
Shells Virtual Desktop

Categories

In this Discussion

Home › General
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.

Netcup Root Server G11 (Annual Cycle of Repeats)

riomartinriomartin Member
in General

https://lowendbox.com/blog/netcup-faces-user-riot-over-throttling-supposedly-dedicated-cores/

It happens again in August - September 2024 , the 2nd time.
Impact on Root Server G11 Nuremberg
not just the performance of Root Server, but impact on the network speed also.

YABS dropped again.
I got GB6scores SingleCore 1400 and MultiCore 4800 during PeakTime after LunchHour at Germany Time

Looks to me,
Profits from over subscribing makes more sense to me.

Thanked by 1mustafamw3
«1

Comments

  • mustafamw3mustafamw3 Member, Patron Provider, LIR
    edited September 2024

    Netcup root server cores are not dedicated cores, but they allow unrestricted usage. You can use 100% of the core, but it’s not dedicated. The provider relies on the fact that not all customers will use 100% of their core all the time.

    Thanked by 2TODO nick_
  • faleddofaleddo Member

    Mine in Vienna also dropped to 1600 in GB6. This impacts to Wordpress (and Woocommerce) performance. WP bench score dropped from 9+ to 7+.

  • emghemgh Member, Megathread Squad
    edited September 2024

    It’s crazy that this is still a discussion.

    Even if you don’t overprovision (not saying that they don’t), single core (thread, actually) speed IS affected by overall load due to multithreading BUT mostly due to thermal throttling.

    If you want to maintain zero impact by other guests, get a dedicated server.

    I’m so tired of people paying for a shared envirement and then expecting zero guests in that.

    The above goes NO MATTER if they lie about dedicated resources or not. That’s a seperate discussion. But a drop from 1600 to 1400 when the hypervisor isn’t idle is rutine. Get a dedicated server.

    Emgh out.

    Thanked by 22Silvest Arkas nick_ MaxTakeba khalequzzaman dev_vps Porlam maverick WhiteRoseG angstrom sh97 Xrmaddness AXYZE host_c Maounique BilohBucks Shade TODO xms WebProject khadhafi1083 fohadeel
  • SilvestSilvest Member

    @emgh said:
    It’s crazy that this is still a discussion.

    Even if you don’t overprovision (not saying that they don’t), single core (thread, actually) speed IS affected by overall load due to multithreading BUT mostly due to thermal throttling.

    If you want to maintain zero impact by other guests, get a dedicated server.

    I’m so tired of people paying for a shared envirement and then expecting zero guests in that.

    The above goes NO MATTER if they lie about dedicated resources or not. That’s a seperate discussion. But a drop from 1600 to 1400 when the hypervisor isn’t idle is rutine. Get a dedicated server.

    Emgh out.

    Couldn't have said it better myself.

    Thanked by 1emgh
  • WhiteRoseGWhiteRoseG Member
    edited September 2024

    Price performance wise it's currently the best server I have, for this little money this amount of compute power with 2,5gbit line. Sorry but better than this it's not gonna get. Like @emgh said get a dedi

    Thanked by 6emgh loay Silvest Maounique khalequzzaman TODO
  • riomartinriomartin Member

    10 percents drop is acceptable
    more than that, ROBBING is what the suitable WORDS .. :))

  • riomartinriomartin Member

    @WhiteRoseG said:
    Price performance wise it's currently the best server I have, for this little money this amount of compute power with 2,5gbit line. Sorry but better than this it's not gonna get. Like @emgh said get a dedi

    Little money?
    If customers alerted with drop performance of 50%
    who going to choose this VPS or RootServer if the drop happens almost 12-16 hours a day.

    Maybe customers are targeted to be fooled.

    Thanked by 1mustafamw3
  • SilvestSilvest Member
    edited September 2024

    @riomartin said:

    @WhiteRoseG said:
    Price performance wise it's currently the best server I have, for this little money this amount of compute power with 2,5gbit line. Sorry but better than this it's not gonna get. Like @emgh said get a dedi

    Little money?
    If customers alerted with drop performance of 50%
    who going to choose this VPS or RootServer if the drop happens almost 12-16 hours a day.

    Maybe customers are targeted to be fooled.

    Get a dedi. You'd get performance drops everywhere when nodes starts to fill. That is to be expected. Geekbench benchmarks isnt actually meant for hosting setting.

    Thanked by 1emgh
  • AXYZEAXYZE Member

    What exactly RS server you got? RS1000?

  • riomartinriomartin Member

    @AXYZE said:
    What exactly RS server you got? RS1000?

    RS2000

  • tra10000tra10000 Member

    RS 1000 G11 4c8g512gb (vienna)

    10 days ago

    Geekbench 6 Benchmark Test:
---------------------------------
Test            | Value
                |
Single Core     | 1852
Multi Core      | 6224

    Now

    Geekbench 6 Benchmark Test:
---------------------------------
Test            | Value
                |
Single Core     | 1831
Multi Core      | 5748
  • AXYZEAXYZE Member

    @riomartin said:

    @AXYZE said:
    What exactly RS server you got? RS1000?

    RS2000

    Singlecore is normal (you get 1800-2000 if you get turbo boost for your core), but 4800 is very bad result for 8cores. Contant support.

  • AdvinAdvin Member, Host Rep
    edited September 2024

    @emgh said:
    It’s crazy that this is still a discussion.

    Even if you don’t overprovision (not saying that they don’t), single core (thread, actually) speed IS affected by overall load due to multithreading BUT mostly due to thermal throttling.

    If you want to maintain zero impact by other guests, get a dedicated server.

    I’m so tired of people paying for a shared envirement and then expecting zero guests in that.

    The above goes NO MATTER if they lie about dedicated resources or not. That’s a seperate discussion. But a drop from 1600 to 1400 when the hypervisor isn’t idle is rutine. Get a dedicated server.

    Emgh out.

    I highly doubt it's due to thermal throttling, almost every respectable VPS provider would try their best to keep their CPUs under the thermal limits (or at the very least, have enough cooling to the point where the difference in performance is very negligible). There is almost no incentive to intentionally thermal throttle processors.

    AMD EPYC processors are “easy” to cool because of the large IHS, and even the top end SKU (9654) can just be air cooled in a 2U.

    From our experience, even if we keep nodes under 50-60% CPU usage, we still see significant drops in GB6 performance on AMD EPYC 9004 series of processors. Despite having almost 40-50% of CPU leftover and not even reaching maximum CPU load on these host nodes, drops in GB6 scores can happen. It's important to keep in mind that they are simply artificial benchmarks that are typically not at all representative of real-world performance.

    Thanked by 2maverick riomartin
  • loayloay Member

    @tra10000 said:
    RS 1000 G11 4c8g512gb (vienna)

    10 days ago

    Geekbench 6 Benchmark Test:
---------------------------------
Test            | Value
                |
Single Core     | 1852
Multi Core      | 6224

    Now

    Geekbench 6 Benchmark Test:
---------------------------------
Test            | Value
                |
Single Core     | 1831
Multi Core      | 5748

    RS 1000 G11 SE VIE BSS24

    27/8

    Geekbench 6 Benchmark Test:
---------------------------------
Test            | Value
                |
Single Core     | 1969
Multi Core      | 6326
Full Test       | https://browser.geekbench.com/v6/cpu/7515176

    19/9

    Geekbench 6 Benchmark Test:
---------------------------------
Test            | Value
                |
Single Core     | 1729
Multi Core      | 5226
Full Test       | https://browser.geekbench.com/v6/cpu/7860005
    Thanked by 1riomartin
  • AXYZEAXYZE Member
    edited September 2024

    @tra10000 said:
    RS 1000 G11 4c8g512gb (vienna)

    10 days ago

    Geekbench 6 Benchmark Test:
---------------------------------
Test            | Value
                |
Single Core     | 1852
Multi Core      | 6224

    Now

    Geekbench 6 Benchmark Test:
---------------------------------
Test            | Value
                |
Single Core     | 1831
Multi Core      | 5748

    Its still hella good for 96core Genoa. These chips are used so much by providers, because it is dense chip.
    Infinity fabric is not fast enough for that much cores, this is why performance degrades even without any overselling, because communication inside CPU doesnt scale to good enough level.

    I dont know if there is any point of going above 48/64 cores on Genoa, I saw so many benchmarks where performance just stops to scale even tho cores are seemingly utilized. Maybe some provider can test it if KVM also doesnt scale, so far @ProHosting24 found degredation, but I dont know after how many cores were used

    Thanked by 1loay
  • advinserversadvinservers Member, Patron Provider
    edited September 2024

    @AXYZE said:

    @tra10000 said:
    RS 1000 G11 4c8g512gb (vienna)

    Its still hella good for 96core Genoa. These chips are used so much by providers, because it is dense chip.
    Infinity fabric is not fast enough for that much cores, this is why performance degrades even without any overselling, because communication inside CPU doesnt scale to good enough level.

    I dont know if there is any point of going above 48/64 cores on Genoa, I saw so many benchmarks where performance just stops to scale even tho cores are seemingly utilized.

    We started to notice that clocks tend to boost a lot more aggressively when using the high performance preset and setting TDP to 400W on the 9654 SKUs. We started doing that on more recent nodes and the situation had improved significantly.

    Thanked by 1AXYZE
  • angstromangstrom Moderator

    @riomartin said: It happens again in August - September 2024 , the 2nd time.
    Impact on Root Server G11 Nuremberg
    not just the performance of Root Server, but impact on the network speed also.

    This is your third thread expressing dissatisfaction with netcup

    (See https://lowendtalk.com/discussion/196611/i-was-shocked-using-netcup-vps and https://lowendtalk.com/discussion/197083/netcup-cpu-steal-and-the-low-quality-of-support )

    Why not cancel the server? Or do you have a contract for a year?

    Thanked by 1khadhafi1083
  • WhiteRoseGWhiteRoseG Member

    Open a topic in there forum, support is pretty good from NetCup

    Thanked by 1khadhafi1083
  • tra10000tra10000 Member 
    fio Disk Speed Tests (Mixed R/W 50/50) (Partition /dev/vda3):
---------------------------------
Block Size | 4k            (IOPS) | 64k           (IOPS)
  ------   | ---            ----  | ----           ----
Read       | 97.30 MB/s   (24.3k) | 209.16 MB/s   (3.2k)
Write      | 97.56 MB/s   (24.3k) | 210.26 MB/s   (3.2k)
Total      | 194.86 MB/s  (48.7k) | 419.42 MB/s   (6.5k)
           |                      |
Block Size | 512k          (IOPS) | 1m            (IOPS)
  ------   | ---            ----  | ----           ----
Read       | 233.84 MB/s    (456) | 420.54 MB/s    (410)
Write      | 246.27 MB/s    (480) | 448.55 MB/s    (438)
Total      | 480.11 MB/s    (936) | 869.10 MB/s    (848)

    Actually I think the real problem is not the cpu but the disk speeds

    It is decreasing every passing day. It will be like Contabo soon.

  • riomartinriomartin Member

    @Silvest said:

    @riomartin said:

    @WhiteRoseG said:
    Price performance wise it's currently the best server I have, for this little money this amount of compute power with 2,5gbit line. Sorry but better than this it's not gonna get. Like @emgh said get a dedi

    Little money?
    If customers alerted with drop performance of 50%
    who going to choose this VPS or RootServer if the drop happens almost 12-16 hours a day.

    Maybe customers are targeted to be fooled.

    Get a dedi. You'd get performance drops everywhere when nodes starts to fill. That is to be expected. Geekbench benchmarks isnt actually meant for hosting setting.

    @loay said:

    @tra10000 said:
    RS 1000 G11 4c8g512gb (vienna)

    10 days ago

    Geekbench 6 Benchmark Test:
---------------------------------
Test            | Value
                |
Single Core     | 1852
Multi Core      | 6224

    Now

    Geekbench 6 Benchmark Test:
---------------------------------
Test            | Value
                |
Single Core     | 1831
Multi Core      | 5748

    RS 1000 G11 SE VIE BSS24

    27/8

    Geekbench 6 Benchmark Test:
---------------------------------
Test            | Value
                |
Single Core     | 1969
Multi Core      | 6326
Full Test       | https://browser.geekbench.com/v6/cpu/7515176

    19/9

    Geekbench 6 Benchmark Test:
---------------------------------
Test            | Value
                |
Single Core     | 1729
Multi Core      | 5226
Full Test       | https://browser.geekbench.com/v6/cpu/7860005

    I agree on this one.
    Same experience.
    Specially after Germany lunch hour until 5-6pm,

  • riomartinriomartin Member

    @tra10000 said:
    fio Disk Speed Tests (Mixed R/W 50/50) (Partition /dev/vda3):
    ---------------------------------
    Block Size | 4k (IOPS) | 64k (IOPS)
    ------ | --- ---- | ---- ----
    Read | 97.30 MB/s (24.3k) | 209.16 MB/s (3.2k)
    Write | 97.56 MB/s (24.3k) | 210.26 MB/s (3.2k)
    Total | 194.86 MB/s (48.7k) | 419.42 MB/s (6.5k)
    | |
    Block Size | 512k (IOPS) | 1m (IOPS)
    ------ | --- ---- | ---- ----
    Read | 233.84 MB/s (456) | 420.54 MB/s (410)
    Write | 246.27 MB/s (480) | 448.55 MB/s (438)
    Total | 480.11 MB/s (936) | 869.10 MB/s (848)

    Actually I think the real problem is not the cpu but the disk speeds

    It is decreasing every passing day. It will be like Contabo soon.

    Correct ..
    The worst experience is with one my ARM @ NUE
    Disk Speed only 75% of this YABS

    And I don't know if disk speeds play a big role also in performance

    Thanks

  • WhiteRoseGWhiteRoseG Member

    Why you don't send support a mail

  • AXYZEAXYZE Member
    edited September 2024

    @advinservers said:

    @AXYZE said:

    @tra10000 said:
    RS 1000 G11 4c8g512gb (vienna)

    Its still hella good for 96core Genoa. These chips are used so much by providers, because it is dense chip.
    Infinity fabric is not fast enough for that much cores, this is why performance degrades even without any overselling, because communication inside CPU doesnt scale to good enough level.

    I dont know if there is any point of going above 48/64 cores on Genoa, I saw so many benchmarks where performance just stops to scale even tho cores are seemingly utilized.

    We started to notice that clocks tend to boost a lot more aggressively when using the high performance preset and setting TDP to 400W on the 9654 SKUs. We started doing that on more recent nodes and the situation had improved significantly.

    Have you tried 9754? This CPU suffers the most from the scaling problem, I didnt found out anything where it would be better than 9654. Not just because of clocks. When I added more parallel encodes to fill all of cores then more CPU cores were utilized, but combined FPS of encodes was the same, because each encode slowed down. More power used, same overall performance, so there is some bottleneck in architecture in IO or frontend of cores and you could go for cheaper 96core in a first place and get same(or better cause higher boost clocks) performance.

    I think Infinity Fabric is the problem as overclocking it even on 16cores increases performance.

    "up to 32Gbps" on Genoa doesnt sound that great for 128 cores with context switching all the time (cause more hypervisors deployed per node). Maybe it is enough and problem lies somewhere else like DDR5 bandwidth or limited L3 caches, idk, but for now I think Infinity Fabric just doesnt scale good enough.

    Thanked by 2riomartin techdragon
  • tra10000tra10000 Member

    @WhiteRoseG said:
    Why you don't send support a mail

    It's sitting idle. If it was for production I would contact support.

  • WhiteRoseGWhiteRoseG Member

    @tra10000 said:

    @WhiteRoseG said:
    Why you don't send support a mail

    It's sitting idle. If it was for production I would contact support.

    Or go to Contabo, very good for production

  • tra10000tra10000 Member

    @WhiteRoseG said:

    @tra10000 said:

    @WhiteRoseG said:
    Why you don't send support a mail

    It's sitting idle. If it was for production I would contact support.

    Or go to Contabo, very good for production

    I'll ask you who I should buy from next time.

  • riomartinriomartin Member

    @AXYZE said:

    @advinservers said:

    @AXYZE said:

    @tra10000 said:
    RS 1000 G11 4c8g512gb (vienna)

    Its still hella good for 96core Genoa. These chips are used so much by providers, because it is dense chip.
    Infinity fabric is not fast enough for that much cores, this is why performance degrades even without any overselling, because communication inside CPU doesnt scale to good enough level.

    I dont know if there is any point of going above 48/64 cores on Genoa, I saw so many benchmarks where performance just stops to scale even tho cores are seemingly utilized.

    We started to notice that clocks tend to boost a lot more aggressively when using the high performance preset and setting TDP to 400W on the 9654 SKUs. We started doing that on more recent nodes and the situation had improved significantly.

    Have you tried 9754? This CPU suffers the most from the scaling problem, I didnt found out anything where it would be better than 9654. Not just because of clocks. When I added more parallel encodes to fill all of cores then more CPU cores were utilized, but combined FPS of encodes was the same, because each encode slowed down. More power used, same overall performance, so there is some bottleneck in architecture in IO or frontend of cores and you could go for cheaper 96core in a first place and get same(or better cause higher boost clocks) performance.

    I think Infinity Fabric is the problem as overclocking it even on 16cores increases performance.

    "up to 32Gbps" on Genoa doesnt sound that great for 128 cores with context switching all the time (cause more hypervisors deployed per node). Maybe it is enough and problem lies somewhere else like DDR5 bandwidth or limited L3 caches, idk, but for now I think Infinity Fabric just doesnt scale good enough.

    Thanks for the references.
    That make sense they had so many issues with their Nuremberg location :))
    Soon Vienna
    and Amsterdam is just counting the month ... soon in 1 or 2

    I am planning to test their Netcup Milan VPS 2000 or 1000 instead of RootServer
    While advise for Dedicated Server would be good option.

  • riomartinriomartin Member

    Based on searching this LET these late 2 days , it seems Netcup had repeatable issues and YABS from 2020 showing the performance still in par with Contabo,
    while Contabo is so reputable in this LET (for their down side of course :)))

    Guess no other option, Hetzner or Dedicated

  • amarcamarc Veteran

    @riomartin said: Based on searching this LET these late 2 days

    Seriously, cancel service.. and save your health and us from Netcup nagging

    Thanked by 1riomartin
  • remyremy Member

    @riomartin said:
    Based on searching this LET these late 2 days , it seems Netcup had repeatable issues and YABS from 2020 showing the performance still in par with Contabo,
    while Contabo is so reputable in this LET (for their down side of course :)))

    Guess no other option, Hetzner or Dedicated

    That's a bit of an exaggeration. Netcup isn't perfect, but comparing it to contabo isn't right.

    Otherwise, if you have no constraints on processor architecture.
    You can test their ARM range, I've had stable performance since day one.

Sign In or Register to comment.