Howdy, Stranger!

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


New Hetzner RX170/RX220 Arm 80 core - Avoid
New on LowEndTalk? Please Register and read our Community Rules.

New Hetzner RX170/RX220 Arm 80 core - Avoid

luckypenguinluckypenguin Member
edited July 20 in General

Essentially the CPU is ok, but nothing is adapted to it yet, unless you build own
software from your own sources, which is not for reselling purposes. Will do a great
job if you have parallel nginx workers, or compiler jobs, but not for the LowEnd market,
so do take into consideration the IPv4 prices and it will make this offer far from ideal.
So - no, dedicated ARM cores per customer isn't going to justify anything, still.
Actually, taking 6x M1 Mac Minis for that price at Hetzner would give you much more potential.
Or just get 6x Ryzens for now for the same price.

Even for NAT VPS, if you found some killer business plan, baremetal hypervisor panels are
not ready for it yet, so all Solus, Proxmox and Virtualizor resellers should wait a couple of years
more, at least. But 1v1 Ryzen cores still outperform Ampere ones. And you can get 6 AX41s for the price of a single RX170, think about it again.

Thanked by 1jar
«1

Comments

  • ericlsericls Member, Patron Provider
    edited July 20

    At least I don't need to cross compile anymore lol.

    Yea at that price, the M1 would be suitable. Also single node is not suitable for any production environment (unless you treat each node as ephemeral, but the set-up fee makes it not suitable) so you'll need more than 2 nodes to do anything. Hope they have ARM based cloud solutions.

  • jarjar Member, Patron Provider

    I do think arm is going to play a continually increasing role but yeah, I feel like if you don’t know up front precisely why you want those boxes they’re not for you. They’re unlikely to be anyone’s platform for dipping their toes in the water.

  • @ericls said: At least I don't need to cross compile anymore lol.

    Cross compiling on M1 is faster than natively compiling a project on x86_64>x86_64 for me.

  • ericlsericls Member, Patron Provider

    @luckypenguin said:

    @ericls said: At least I don't need to cross compile anymore lol.

    Cross compiling on M1 is faster than natively compiling a project on x86_64>x86_64 for me.

    woo, that's good news!

  • yoursunnyyoursunny Member, IPv6 Advocate

    RX220 is €3399.83 per year (with setup fee).
    If selling NAT VPS with 1 dedicated core each, the VPS spec would be:

    • 1 dedicated core
    • 3 GiB RAM
    • 44 GiB NVMe RAID1
    • €42 per year

    Not a good deal.

  • DPDP Member, Moderator, The Domain Guy

    @yoursunny said:
    RX220 is €3399.83 per year (with setup fee).
    If selling NAT VPS with 1 dedicated core each, the VPS spec would be:

    • 1 dedicated core
    • 3 GiB RAM
    • 44 GiB NVMe RAID1
    • €42 per year

    Not a good deal.

    💯

  • ralfralf Member

    @luckypenguin said:
    Actually, taking 6x M1 Mac Minis for that price at Hetzner would give you much more potential.

    Except you could only get 3x M1 for the price of RX170 and 4x M1 for the price of RX220.

    An M1 has 4 performance cores and 4 efficiency cores. I don't know how either benches to the Altera 80-core, but 16+16 M1 cores probably has less CPU power than 80 Altera cores.

    But sure, probably not great for the reselling market, but probably good for custom applications needing many cores.

  • eriseris Member

    I don't think Hetzner cares about the LET market when the decided to provide a RX series but business who really benefit form the use of ARM based CPU...

  • Daniel15Daniel15 Member
    edited July 20

    Another thing worth noting is that the price of ARM servers probably won't go up as much as the price of x86-64 servers, as they use less power (which is one of the major contributors to the cost of running a server).

    But yes, you need to make sure all your software can run on ARM.

  • lanefulanefu Member

    If your stacks are based on packages from OS distribution and interpreted languages the experience is pretty turn key these days.

    Java of course is ready to go.

    There are arm containers for most major runtimes.
    You'll see multiarch containers are becoming more common. Docker official images have been multiarch for a few years now. naturally bespoke unofficial containers won't work.

    Golang apps are the easiest to cross compile for different architectures. Rust apps aren't bad.

    Also real cores and no hyperthreading BS.

    These things are more ready for prime time than one may expect. Kick the tires and run a big ass postgres box

    Thanked by 2ericls Daniel15
  • ericlsericls Member, Patron Provider

    @lanefu said:
    If your stacks are based on packages from OS distribution and interpreted languages the experience is pretty turn key these days.

    Java of course is ready to go.

    There are arm containers for most major runtimes.
    You'll see multiarch containers are becoming more common. Docker official images have been multiarch for a few years now. naturally bespoke unofficial containers won't work.

    Golang apps are the easiest to cross compile for different architectures. Rust apps aren't bad.

    Also real cores and no hyperthreading BS.

    These things are more ready for prime time than one may expect. Kick the tires and run a big ass postgres box

    I absolutely agree, the future will be ARM (Q: what stock should I buy now?), especially now a lot of developers use ARM to develop software anyways. I think the problem is that the advantage of ARM in performance, density, efficiency isn't reflected on the pricing of this particular product.

  • serv_eeserv_ee Member

    @lanefu said:
    If your stacks are based on packages from OS distribution and interpreted languages the experience is pretty turn key these days.

    Java of course is ready to go.

    There are arm containers for most major runtimes.
    You'll see multiarch containers are becoming more common. Docker official images have been multiarch for a few years now. naturally bespoke unofficial containers won't work.

    Golang apps are the easiest to cross compile for different architectures. Rust apps aren't bad.

    Also real cores and no hyperthreading BS.

    These things are more ready for prime time than one may expect. Kick the tires and run a big ass postgres box

    It's so primetime that AWS and Oracle have to give them away for free each month. Much prime such time.

    "ARM WILL TAKE OVER" blah blah blah. Yeah, haven't heard that for a decade now, nah not at all.

  • jsgjsg Member, Resident Benchmarker

    The point of Ampere and similar based machines mainly is to have lots of cores and computing power in a single system. If one wants to compare that to Ryzen or Xeon one needs to look at 4 - 6 socket systems; 4 socket systems actually exist but just look at the price tag and you'll see that Ampere based machines actually are attractive (for certain use cases). And don't forget to compare el. power cost too, or in other words, sometimes the game is about many cores and a good performance per Watt ratio.

    Hetzner will find customers for those systems and in fact almost certainly already has quite a few and more coming.

  • eriseris Member

    It is allready out of stock ...

  • sotssots Member

    Which OS on ARM platform has the most applications? Android.
    If one can set up 'virtual phones' on the ARM servers, I think that's great. If you have some applications that you don't want them to run on your own phone, then you can get a virtual phone at reasonable pricing.

  • LordSpockLordSpock Member, Host Rep
    edited July 23

    I don't think avoid is quite the right phrase. It was never going to be meant for someone who just wants to plop their x86 shit somewhere.

    I think these Ampere boxes are a nice step in the right direction to getting ARM adoption higher. I personally enjoy the architecture primarily for its power efficiency. I use M1 and I use ARM machines in production for a number of things.

    Lots of software is supported, some of it requires a few workarounds or having to compile stuff yourself, but I haven't really bumped in to any significant issues.

    As for no hypervisors being available, 1 - so what? all the ingredients are there. 2 - Proxmox should be relatively trivial to get working, there is a project that's been going for a little while getting it working on the Raspberry Pi.

  • ralfralf Member

    Wow, that does look impressive. TLDW: GB5 score of 44000.

    Shame he took the chip to bits and exposed the die, but didn't shove it under a microscope and share pictures of that.

  • @luckypenguin, no yabs?

    44,000 of gb5, almost twice as much as AMD Ryzen Threadripper 3990X with its ~25,300
    So sweet!

  • ralfralf Member
    edited August 5

    Interesting review on BANNED Huawei Arm server:
    https://www.servethehome.com/the-forbidden-arm-server-that-is-banned-in-the-us/
    (TLDR: don't buy!)

  • rm_rm_ Member
    edited August 5

    @Daniel15 said: Another thing worth noting is that the price of ARM servers probably won't go up as much as the price of x86-64 servers, as they use less power (which is one of the major contributors to the cost of running a server).

    Do you have any measurements that the Ampere Altra Q80-30 server chip with a TDP of 210W uses "less power"? Less than what?

    Actually even 250W for Hetzner's model...

  • Daniel15Daniel15 Member
    edited August 5

    @rm_ said:

    @Daniel15 said: Another thing worth noting is that the price of ARM servers probably won't go up as much as the price of x86-64 servers, as they use less power (which is one of the major contributors to the cost of running a server).

    Do you have any measurements that the Ampere Altra Q80-30 server chip with a TDP of 210W uses "less power"? Less than what?

    I'm talking about ARM in general, not this specific chip. RISC architecture is generally going to use less power than CISC. This is a very powerful chip, so it's not as noticeable (compared to, for example, comparing a Raspberry Pi to an older PC with the same power).

    Having said that, the benchmarks show that this chip is faster than a Threadripper 3990X, which has a 280W TDP and uses quite a bit more when boosting. A 25% drop in power consumption is significant.

  • Ubuntu and Debian 11 work great on ARM server
    Use a PI4 for your tests and you will see the power of this architecture

  • jarjar Member, Patron Provider
    edited August 9

    We ready for a benchmark?

    https://browser.geekbench.com/v5/cpu/16518669

    For comparison, I benchmarked a totally different server, not a Hetzner box at all:

    https://browser.geekbench.com/v5/cpu/16518692

    GeekBench versions were a little different, don't think there's an updated ARM version but it's not that old.

    But quite simply, it outperformed a Xeon E5-2650 v2 in single threading, and just demolished it in multithreading.

    Thanked by 1ralf
  • ErisaErisa Member

    @jar said:
    We ready for a benchmark?

    https://browser.geekbench.com/v5/cpu/16518669

    For comparison, I benchmarked a totally different server, not a Hetzner box at all:

    https://browser.geekbench.com/v5/cpu/16518692

    Seems quite low compared to the typical, I wonder why
    https://browser.geekbench.com/search?utf8=✓&q=MP32-AR1-SW-HZ-001

    Thanked by 2jar ralf
  • jarjar Member, Patron Provider

    @Erisa said:

    @jar said:
    We ready for a benchmark?

    https://browser.geekbench.com/v5/cpu/16518669

    For comparison, I benchmarked a totally different server, not a Hetzner box at all:

    https://browser.geekbench.com/v5/cpu/16518692

    Seems quite low compared to the typical, I wonder why
    https://browser.geekbench.com/search?utf8=✓&q=MP32-AR1-SW-HZ-001

    I wonder if it's the difference of GB version, the top one on there seemed to be 5.4.4 and I used 5.4.0.

    Thanked by 1Erisa
  • ErisaErisa Member
    edited August 9

    @jar said:

    @Erisa said:

    @jar said:
    We ready for a benchmark?

    https://browser.geekbench.com/v5/cpu/16518669

    For comparison, I benchmarked a totally different server, not a Hetzner box at all:

    https://browser.geekbench.com/v5/cpu/16518692

    Seems quite low compared to the typical, I wonder why
    https://browser.geekbench.com/search?utf8=✓&q=MP32-AR1-SW-HZ-001

    I wonder if it's the difference of GB version, the top one on there seemed to be 5.4.4 and I used 5.4.0.

    Maybe, yeah.

    The treatment for ARM64 downloads is strange, the only link that comes up is https://www.geekbench.com/blog/2021/03/geekbench-54/ which serves https://cdn.geekbench.com/Geekbench-5.4.0-LinuxARMPreview.tar.gz

    And yet other versions exist, all the way through to 5.4.4, yet no mention of them that I can find.
    e.g. https://cdn.geekbench.com/Geekbench-5.4.4-LinuxARMPreview.tar.gz

    Thanked by 1jar
  • jarjar Member, Patron Provider
    edited August 9

    @Erisa said: And yet other versions exist, all the way through to 5.4.4, yet no mention of them that I can find.
    e.g. https://cdn.geekbench.com/Geekbench-5.4.4-LinuxARMPreview.tar.gz

    Sweet. For good measure: https://browser.geekbench.com/v5/cpu/16518814

    About the same though.

  • ErisaErisa Member

    @jar said: About the same though.

    Yeah I honestly don't know why it would be so much lower than the others, I haven't messed with these machines myself so don't have any experience to draw from.

    Thanked by 1jar
  • ralfralf Member

    @Erisa said:

    @jar said:
    We ready for a benchmark?

    https://browser.geekbench.com/v5/cpu/16518669

    For comparison, I benchmarked a totally different server, not a Hetzner box at all:

    https://browser.geekbench.com/v5/cpu/16518692

    Seems quite low compared to the typical, I wonder why
    https://browser.geekbench.com/search?utf8=✓&q=MP32-AR1-SW-HZ-001

    These results all seem to be from end of May. Maybe that's when they were still trying out different settings to find out what was most stable?

Sign In or Register to comment.