Howdy, Stranger!

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


Hetzner going no setup fee for AX41 NVMe - Page 2
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 going no setup fee for AX41 NVMe

2

Comments

  • @kzed said:
    @CalmDown, it fast, yes, but the Samsung one only has 100 TBW, i thought this was a lot and enough for me, but as soon as i using it for production, it wearout very fast,,
    i only using it less than a year but the wearout level is already at 70%..
    i plan to order another server, just kinda curious what nvme drives they are using now..

    maybe stop plotting ;-)

    afaik the 981 model with 512GB is rated with 300TBW. that might still be low compared to others but if you burned through more 200TB in just one year Hetzner will for sure ask what you were doing if asking for a replacement. they nowadays have things in their ToS that might have you pay for abused hardware...

  • @Falzo wait, where did you get the 300TBW info from? based on S.M.A.R.T values, mine says 67% wear out with 68TB data already written, so i assume the TBW is 100TB,
    i dont know why it wear out fast, probably because it being use for busy database server.

  • @kzed said:
    anyone ordered Ax41 recently? what nvme drives they using now?
    are they still using Samsung MZVLB512HBJQ?

    Ordered one and it was delivered with Samsung drives.

    Thanked by 1kzed
  • @YorkiS, thanks!, did you get new drives?

  • @kzed said:
    @Falzo wait, where did you get the 300TBW info from? based on S.M.A.R.T values, mine says 67% wear out with 68TB data already written, so i assume the TBW is 100TB,
    i dont know why it wear out fast, probably because it being use for busy database server.

    PM981a 512GB is the same as 970 Evo 500GB (just that it has diff firmware so they reserve diff amount of nand) which has 300TBW

    Thanked by 2Falzo kzed
  • FalzoFalzo Member
    edited September 2021

    @kzed said:
    @Falzo wait, where did you get the 300TBW info from? based on S.M.A.R.T values, mine says 67% wear out with 68TB data already written, so i assume the TBW is 100TB,
    i dont know why it wear out fast, probably because it being use for busy database server.

    usually their warranty limitation hold some relevant data... f.i. https://www.samsung.com/semiconductor/global.semi.static/SAMSUNG_SSD_Limited_Warranty_German.pdf

    most of the 500GB models are 300TBW or more. I wouldn't trust that wearout thingy in s.m.a.r.t anyway. it will drop to zero at some point for sure, but that doesn't mean anything.

    if at all the vendors will put in high buffers/tolerances to their protection. so it will tell you very early 'I am dead' while it probably is totally fine for years to come.

    if you have ~70TBW written for now I wouldn't worry much about the next 2-3 years here at all.

    Thanked by 1kzed
  • @Falzo, how do we know if an nvme already near its eol if stats from s.m.a.r.t incorrectly report the status?

  • @kzed said:
    @Falzo, how do we know if an nvme already near its eol if stats from s.m.a.r.t incorrectly report the status?

    simple. you don't, at least not for sure. there is no fixed EOL anyway. I have seen ssds with far more TB written, than what would be guarantued, running just fine.

    However, I do watch out for the regular 'reallocated sectors' which some disks still show and even more importantly the 'spare threshold' or 'unused blocks reserve' or things like that.
    reason for that is, that the memory cells itself might have a slightly different lifespan and for sure the wearout itself won't be spread equally across them anyway. so all ssds come with some additional space from which dying cells can be replaced, a reserve.

    only if this starts getting used I'd also start to think about a replacement.

    Thanked by 1kzed
  • thank you for the explanation!...

    Thanked by 1Falzo
  • darkimmortaldarkimmortal Member
    edited September 2021

    @AXYZE said:

    @kzed said:
    @Falzo wait, where did you get the 300TBW info from? based on S.M.A.R.T values, mine says 67% wear out with 68TB data already written, so i assume the TBW is 100TB,
    i dont know why it wear out fast, probably because it being use for busy database server.

    PM981a 512GB is the same as 970 Evo 500GB (just that it has diff firmware so they reserve diff amount of nand) which has 300TBW

    Enterprise server SSDs do minimal-to-no background rewriting, so for the same NAND chips TBW and unpowered data retention will be worse than SSDs designed for desktops/laptops. This is because it can never move static data implicitly, resulting in uneven spread of writes if some data is never changed. In exchange you get more consistent performance under continuous load

  • @darkimmortal said:

    @AXYZE said:

    @kzed said:
    @Falzo wait, where did you get the 300TBW info from? based on S.M.A.R.T values, mine says 67% wear out with 68TB data already written, so i assume the TBW is 100TB,
    i dont know why it wear out fast, probably because it being use for busy database server.

    PM981a 512GB is the same as 970 Evo 500GB (just that it has diff firmware so they reserve diff amount of nand) which has 300TBW

    Enterprise server SSDs do minimal-to-no background rewriting, so for the same NAND chips TBW and unpowered data retention will be worse than SSDs designed for desktops/laptops. This is because it can never move static data implicitly, resulting in uneven spread of writes if some data is never changed. In exchange you get more consistent performance under continuous load

    Samsung SSD for enterprise are named PMxxxx, not PMxxx which are consumer drives that you find in laptops, desktops etc. They sell them at lower prices, so thats why Hetzner is using them in cheapest Ryzen machines.

  • @kzed said: based on S.M.A.R.T values, mine says 67% wear out with 68TB data already written, so i assume the TBW is 100TB

    btw: did you check if this number is increasing or decreasing? could be that it started at 100 and is now only decreased to 67... which would make the wearout only 33% instead and translating to a theoretical ~200TBW ;-)

    however, doesn't sound in any way worrying to me yet. ;-)

  • darkimmortaldarkimmortal Member
    edited September 2021

    @AXYZE said:

    @darkimmortal said:

    @AXYZE said:

    @kzed said:
    @Falzo wait, where did you get the 300TBW info from? based on S.M.A.R.T values, mine says 67% wear out with 68TB data already written, so i assume the TBW is 100TB,
    i dont know why it wear out fast, probably because it being use for busy database server.

    PM981a 512GB is the same as 970 Evo 500GB (just that it has diff firmware so they reserve diff amount of nand) which has 300TBW

    Enterprise server SSDs do minimal-to-no background rewriting, so for the same NAND chips TBW and unpowered data retention will be worse than SSDs designed for desktops/laptops. This is because it can never move static data implicitly, resulting in uneven spread of writes if some data is never changed. In exchange you get more consistent performance under continuous load

    Samsung SSD for enterprise are named PMxxxx, not PMxxx which are consumer drives that you find in laptops, desktops etc. They sell them at lower prices, so thats why Hetzner is using them in cheapest Ryzen machines.

    I have an older hetzner server with PM863a which is a true enterprise drive (also using cheap nand), assumed PM981a would be similar but seems it’s used in some laptops so the jury is out on that one whether it has enterprise-like firmware or not

  • @darkimmortal said:

    @AXYZE said:

    @darkimmortal said:

    @AXYZE said:

    @kzed said:
    @Falzo wait, where did you get the 300TBW info from? based on S.M.A.R.T values, mine says 67% wear out with 68TB data already written, so i assume the TBW is 100TB,
    i dont know why it wear out fast, probably because it being use for busy database server.

    PM981a 512GB is the same as 970 Evo 500GB (just that it has diff firmware so they reserve diff amount of nand) which has 300TBW

    Enterprise server SSDs do minimal-to-no background rewriting, so for the same NAND chips TBW and unpowered data retention will be worse than SSDs designed for desktops/laptops. This is because it can never move static data implicitly, resulting in uneven spread of writes if some data is never changed. In exchange you get more consistent performance under continuous load

    Samsung SSD for enterprise are named PMxxxx, not PMxxx which are consumer drives that you find in laptops, desktops etc. They sell them at lower prices, so thats why Hetzner is using them in cheapest Ryzen machines.

    I have an older hetzner server with PM863a which is a true enterprise drive (also using cheap nand), assumed PM981a would be similar but seems it’s used in some laptops so the jury is out on that one whether it has enterprise-like firmware or not

    I was referring to NVMe drives where this 3/4 letter rule applies. PM863a is ofc enterprise sata ssd.

    PM981a is regular consumer drive with TurboWrite etc. It just doesnt have that much reversed space as 970 Evo. Same controller, same nand, green pcb, no marketing and cheaper cost if you buy in bulk.

  • Hi all. I've been a happy Hetzner Cloud customer for a few years but I have been thinking of switching to dedicated, and this offer makes it even more interesting. I know that I can connect multiple dedicated servers to each other in a private lan with a vSwitch, but does anyone know what kind of network speed I can expect with this network connection?
    I need to set up Kubernetes and will use either Longhorn or Ceph/Rook for persistent volumes, so the network speed can make a difference with replicated storage. Thanks in advance!

  • extremecarverextremecarver Member
    edited September 2021

    Damn, just got two 41-NVME servers two weeks ago paying setup fee.

    But my drives are different - In Helsinki I got the much faster
    PM9A1 - MZVL2512HCJQ
    plus a much more modern mainboard (B565D4-V1L and 3600Mhz RAM. But yeah it's also only 300 TBW, and I racked up the first 30TB in 2 weeks. Dunno really how. I compile stuff on those servers and thought I would not do over 3-4TB per week.

    Ordered a 1TB NVME too for it and got a rather old KXG60ZNV1T02.
    Well I would have been fine with non NVME SSD but the price difference is little. Do not know the TBW of this one but I have more trust in Samsung... Usually Samsung you can write much more than the stated endurance... Or is this not true anymore for the MLC drives?

    Actually the Ryzen 3600 is using really high voltage, do you think If I order KVM I can unblock in bios the overclocking settings? Would love to undervolt this one a bit and guess then I could get 200-300Mhz more out of it on all core performance.

    While on the other server in Falkenstein I got unlucky and got 3200MHz RAM, PCIE gen 3, and the usual PM981a
    but yeah - this server I ordered with ECC RAM. So maybe that's why you get slow stuff.

    And yeah - it's a bit a shame they have no option for Samsung Datacenter SSD, the SM883 is not that expensive and has like 8 times the TBW values of the PM9a1 drives...

  • karanchookaranchoo Member
    edited September 2021

    I guess its all chia left overs. before hetzner banned Chia . Ordered an additional 960 GB nvme for my dell dx181
    and this is what I got. (Today)

  • Quick question,

    I am not 100% which server would be fine for me (ax51 or ax61), so I was thinking about ordering both, then test it for a few days and cancel one once I see which one would be fine for me.

    So I would like to know is it still okey to cancel server in first 14 days and they won't charge me anything for that server?

  • CabbageCabbage Member
    edited September 2021

    @luka289 said:
    Quick question,

    I am not 100% which server would be fine for me (ax51 or ax61), so I was thinking about ordering both, then test it for a few days and cancel one once I see which one would be fine for me.

    So I would like to know is it still okey to cancel server in first 14 days and they won't charge me anything for that server?

    It should be okay as long as you don't abuse it. I've done it myself at least once.

  • I have such an itch to get 5-10 of these AX41s but the new IPv4 prices are depressing. Depressing that I can't avail this offer, I wish it was here last year, would have saved a good grand or so on setup fees. Depressing that I have to keep going back to OVH, luckily I don't have to deal with their incompetent support directly or get tied up in their commitments since I'm able to get them from a great reseller. o:)

  • @sunnyg Can you please share the reseller?

  • @comXyz said:
    @sunnyg Can you please share the reseller?

    His name is Andy10gbit, you can check his offers on his discord server or send him a message on there. https://discord.gg/2SvfnDgKqN

    Thanked by 1comXyz
  • What's the uptime/reliability like on these?

  • Has anyone tried to use their server with mostly SE Asian traffic? How was it? They have competitive pricing but do not know if this is good for SE Asian traffic.

  • @imnotarobot said:
    Has anyone tried to use their server with mostly SE Asian traffic? How was it? They have competitive pricing but do not know if this is good for SE Asian traffic.

    It's slow most of the time, both German and Finland. You should use CDN service if possible

  • ArkasArkas Moderator

    I'm going to get 2 of these. If it wasn't for the IP prices, this would have been the ultimate deal!

  • stefemanstefeman Member
    edited September 2021

    Mine is almost dead. both around 20%

    Couldn't resist the offer. But it seems like it was used for CHIA before.

    Thanked by 1vimalware
  • They will probably introduce new ryzen lineup. Good way to rid off old stock.

  • @stefeman said:

    Couldn't resist the offer. But it seems like it was used for CHIA before.

    also you:

    @stefeman said: I don't know about you guys, but I'm moving to another provider next month, and when enough people like me do this, it is gonna sting them no matter how much they tell themselves about "Only business customers matter to us".

    While moving my infra out is annoying to say the least, it is still quite easy decision to do when pricing model changes like this.

    aha. spineless mollusc you are, right? 😂

  • yeah it doesn't really make sense to keep offering 3600. Ryzen 5600 should actually use less power and doesn't cost much more. I usually keep my servers for 3-4 years, so I was really thinking of waiting a bit. But on my old 4 year old server the fan broke, and Hetzner is really crap if your hardware breaks and them figuring out what is broken (if a simple walk by would be enough, but standard procedure is them running some test protocol which takes up to 24 hours while your server is down......).

    It's clear they are not keen on offering intel anymore - as the intel CPUs on the EX line for the same monthly price use much more electricity.

Sign In or Register to comment.