Howdy, Stranger!

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


Hetzner tech says CPU temperature of 100°C is "okay" - 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 tech says CPU temperature of 100°C is "okay"

2»

Comments

  • @CConner said:

    @FHR said:
    I would agree with Hetzner - as long as you're getting at least base clock, you're getting what you're paying for. The temperature doesn't need to concern you, it's not your hardware.

    No where is stated that you're paying for just the base clock. The boost clock is a standard feature of the CPU and can be considered false advertisement if it is not able to boost all of the time whilst customers are not being made aware of that. They're also listing CPU benchmark scores with their machines that did not have their CPU's boost clock disabled.

    What? You fundamentally don't understand what boost means and there is zero expectation of "boost all the time". Maybe you meant "base", not boost.

    Thanked by 4FHR rcxb jsg that_guy
  • Who do you think you are to question hetzner technician?

  • @sureiam said: OMG they must be so stupid loud!!

    In my youth, I was foolish enough to purchase a used 1u rack server for my home use. Regretted it since day one. Fans were so loud.

    Sold it as soon as I could and never looked back.

  • CConnerCConner Member, Host Rep

    @TimboJones said:

    @CConner said:

    @FHR said:
    I would agree with Hetzner - as long as you're getting at least base clock, you're getting what you're paying for. The temperature doesn't need to concern you, it's not your hardware.

    No where is stated that you're paying for just the base clock. The boost clock is a standard feature of the CPU and can be considered false advertisement if it is not able to boost all of the time whilst customers are not being made aware of that. They're also listing CPU benchmark scores with their machines that did not have their CPU's boost clock disabled.

    What? You fundamentally don't understand what boost means and there is zero expectation of "boost all the time". Maybe you meant "base", not boost.

    I think it is reasonable to assume that a provider uses a cooling solution that is rated higher than the max amount of power that said CPU can consume. Thus if the cooling capacity is adequate the CPU will be able to boost all of the time.

  • @CConner said:

    @TimboJones said:

    @CConner said:

    @FHR said:
    I would agree with Hetzner - as long as you're getting at least base clock, you're getting what you're paying for. The temperature doesn't need to concern you, it's not your hardware.

    No where is stated that you're paying for just the base clock. The boost clock is a standard feature of the CPU and can be considered false advertisement if it is not able to boost all of the time whilst customers are not being made aware of that. They're also listing CPU benchmark scores with their machines that did not have their CPU's boost clock disabled.

    What? You fundamentally don't understand what boost means and there is zero expectation of "boost all the time". Maybe you meant "base", not boost.

    I think it is reasonable to assume that a provider uses a cooling solution that is rated higher than the max amount of power that said CPU can consume. Thus if the cooling capacity is adequate the CPU will be able to boost all of the time.

    Any specification for CPU cooling is for base at a less than 100% load for less than 24/7 and boost performance is just gravy. You think the $6 OEM CPU cooler is in the same league as a $200 cooler?

    If you mean "often" and not "all of the time", then sure. But not 24/7.

  • rcxbrcxb Member
    edited July 2020

    @CConner said:
    I think it is reasonable to assume that a provider uses a cooling solution that is rated higher than the max amount of power that said CPU can consume. Thus if the cooling capacity is adequate the CPU will be able to boost all of the time.

    One of the biggest problems in modern (large) CPUs is getting heat from inside the chip out to the heatsink in the first place. No matter how oversize your cooling system, there are internal limitations. It's been problematic since the Pentium4 days. Intel even hedges on it:

    Due to varying power characteristics, some parts with Intel® Turbo Boost Technology 2.0 may not achieve maximum turbo frequencies when running heavy workloads and using multiple cores concurrently.
    https://www.intel.com/content/www/us/en/architecture-and-technology/turbo-boost/turbo-boost-technology.html

    Thanked by 1TimboJones
  • @sdglhm said:

    @sureiam said: OMG they must be so stupid loud!!

    In my youth, I was foolish enough to purchase a used 1u rack server for my home use. Regretted it since day one. Fans were so loud.

    Sold it as soon as I could and never looked back.

    Lol same. If it makes you feel better they are much quieter these days when the system is idle. But when you get even a little hot they get LOUD

  • @sureiam said: But when you get even a little hot they get LOUD

    It didn't help since I'm living in a tropical country with 25℃+ temps regularly

  • JustVPSJustVPS Member, Host Rep

    My opinion is that the end user should not be concerned about the temperature of the provider's equipment. It is the provider's job to monitor its temperature and maintain it within the required range. If overheating is causing underfrequency and performance issues, please ask technical support to correct the situation.

  • ctoomctoom Member

    It is oK to run at 100°C for a server CPU. I'm not a hardware engineer but I have an old server in my offic to mine some cryptocoins, the temperature is always too high even it have 6 fans and I also added an exptra heattsink on the CPU... It will keep mining for weeks and months and everything is OK...!

  • CConnerCConner Member, Host Rep

    @ctoom said:
    It is oK to run at 100°C for a server CPU. I'm not a hardware engineer but I have an old server in my offic to mine some cryptocoins, the temperature is always too high even it have 6 fans and I also added an exptra heattsink on the CPU... It will keep mining for weeks and months and everything is OK...!

    No one is concerned about the lifespan of the hardware. The concern is in regards to performance.

  • @sureiam said:

    @sdglhm said:

    @sureiam said: OMG they must be so stupid loud!!

    In my youth, I was foolish enough to purchase a used 1u rack server for my home use. Regretted it since day one. Fans were so loud.

    Sold it as soon as I could and never looked back.

    Lol same. If it makes you feel better they are much quieter these days when the system is idle. But when you get even a little hot they get LOUD

    The sound level is no joke!
    Noise complaints from our neighbors have played a factor in rent negotiations! Our employees wear hearing protection when they are anywhere near the office. We had to buy a decibel meter to safely work with it, and I've seen areas as high as 110. It wouldn't surprise me if a significant amount of the electrical energy we take in is converted to acoustic instead of heat! I suppose as the walls deaden it, it makes its way into heat, but quite a bit of sound energy leaves our datacenter.

  • jsgjsg Member, Resident Benchmarker
    edited July 2020

    Some possibly interesting remarks

    • Turbo/boost is largely a marketing trick that strives to make processors look faster. The basic idea is to use the upper part of the thermal envelope to make one or a few cores run faster for short amounts of time with short meaning a small fraction of the heat-time product (meaning e.g. one longer burst (per second) of say 10 ms or 100 longer bursts of 90 us). Note that the available "window" is strongly depending on the total load that is, not insignificant boost capacity is available only when the load is low. So, if you have compute intense jobs running (like e.g. video transcoding) there is no or only very little boost available.
    • Heat is one of the major problems for many semiconductors, one of the reasons being that the heat density is very high because the area is very small.
    • Heat is closely related to frequency and unfortunately that relation is not linear, so running a chip say 20% faster is not increasing heat by 20% but much more.
    • What temperature are we talking about anyway? On-die, case-junction or ambient? The sensor is almost always on-die but that does not mean that the temperature you are told by the system is the on-die temperature. 100°C on-die is not a problem, at the junction it may be, and ambient it definitely is.
    • In any DC there is also another relevant relation which is heat vs cost. The major problem being that usually air only is used as a transfer medium, both within the servers and within the DC room. Unfortunately air is a rather poor thermal conductor so the fans in the server must work hard (which btw also translates to them using lots of power, usually much more than e.g. the disks) and within the DC a lot of power is needed to circulate air.
    • Heat is one of the major causes of death in electronics, hence hosters must find their preference and sweet spot. However as hardware costs usually are just a rather small cost factor (over life time) many go for "push it hard" which also happens to help sales (Virtually every customer cares about speed but virtually none cares about server life time)
    • Power consumption increases in a non linear way with increased frequency and power consumption is a major cost factor so usually it makes sense to invest in more cores and to have them running at a reasonable speed (about 80% - 90% of base speed seems to be a good rule of thumb). The additional cores will pay for themselves rather quickly by significantly reduced power consumption.

    TL;DR Forget about "boost" and "turbo" as you usually can not at all count on it being available and also because it's largely just a marketing pseudo-feature (in real life).

    Thanked by 1sureiam
  • ClouviderClouvider Member, Patron Provider

    @ctoom said:
    It is oK to run at 100°C for a server CPU. I'm not a hardware engineer but I have an old server in my offic to mine some cryptocoins, the temperature is always too high even it have 6 fans and I also added an exptra heattsink on the CPU... It will keep mining for weeks and months and everything is OK...!

    That’s not a server CPU.

  • vfusevfuse Member, Host Rep

    We still have a couple of ex41's left that we will replace soon with ax51's. Load on these nodes is between 40-80%. We see much better temps in Helsinki. We are more worried about the nvme temps which can get pretty toasty.

  • @jsg said:
    Some possibly interesting remarks

    • Turbo/boost is largely a marketing trick that strives to make processors look faster. The basic idea is to use the upper part of the thermal envelope to make one or a few cores run faster for short amounts of time with short meaning a small fraction of the heat-time product (meaning e.g. one longer burst (per second) of say 10 ms or 100 longer bursts of 90 us). Note that the available "window" is strongly depending on the total load that is, not insignificant boost capacity is available only when the load is low. So, if you have compute intense jobs running (like e.g. video transcoding) there is no or only very little boost available.
    • Heat is one of the major problems for many semiconductors, one of the reasons being that the heat density is very high because the area is very small.
    • Heat is closely related to frequency and unfortunately that relation is not linear, so running a chip say 20% faster is not increasing heat by 20% but much more.
    • What temperature are we talking about anyway? On-die, case-junction or ambient? The sensor is almost always on-die but that does not mean that the temperature you are told by the system is the on-die temperature. 100°C on-die is not a problem, at the junction it may be, and ambient it definitely is.
    • In any DC there is also another relevant relation which is heat vs cost. The major problem being that usually air only is used as a transfer medium, both within the servers and within the DC room. Unfortunately air is a rather poor thermal conductor so the fans in the server must work hard (which btw also translates to them using lots of power, usually much more than e.g. the disks) and within the DC a lot of power is needed to circulate air.
    • Heat is one of the major causes of death in electronics, hence hosters must find their preference and sweet spot. However as hardware costs usually are just a rather small cost factor (over life time) many go for "push it hard" which also happens to help sales (Virtually every customer cares about speed but virtually none cares about server life time)
    • Power consumption increases in a non linear way with increased frequency and power consumption is a major cost factor so usually it makes sense to invest in more cores and to have them running at a reasonable speed (about 80% - 90% of base speed seems to be a good rule of thumb). The additional cores will pay for themselves rather quickly by significantly reduced power consumption.

    TL;DR Forget about "boost" and "turbo" as you usually can not at all count on it being available and also because it's largely just a marketing pseudo-feature (in real life).

    As you said, it's just a heat buildup limitation. Well, you're really just talking about budget and typical CPU coolers. Only money and space prevent you from using CO2 and getting Turbo all day long.

    The short answer is, this guy isn't paying for an expensive cooling system and gets what he gets.

  • redvi4redvi4 Member

    In my House computer, to render 3D, i use Liquid refrigeration, cos its easy to burn the CPU

  • ctoomctoom Member

    @Clouvider said: That’s not a server CPU.

    not I7, but E5 2650v2,

  • bitcoin mining ?

Sign In or Register to comment.