[Solved]Alternative CPU in my current server


krone6

Recommended Posts

After figuring out everything goes down when the array is stopped I decided to build a virtualization server for my needs. This leads me to figure out what cpu to replace what's in my current unraid server while still not being overkill (I try to be efficient with my money for my tech).

 

Requirements:

 

1: Must use ECC ram as there's already 8GB of ECC ram in there

2: Needs to be as efficient on energy. Electricity is expensive.

3: Must be able to run many dockers. If there are some demanding ones let me know. Right now only Plex and MineOS come to mind though I don't know how demanding those can be yet without more testing.

 

Current Specs:

 

Model: Custom Norco 4224 Server

M/B: Supermicro - X10SLM-F

CPU: Intel® Xeon® CPU E3-1230 v3 @ 3.30GHz

Cache: 256 kB, 1024 kB, 8192 kB

Memory: 8192 MB (max. installable capacity 32 GB)

 

Thanks for the help and if anyone spots something that can be improved let me know, though as far as things such as RAM goes I feel comfortable sitting at 8GB since virtualization isn't going to be a big use anymore.

Link to comment

Not sure I understand the question - are you pulling the Xeon E3-1230 v3 off the Supermicro - X10SLM-F to place in another server?  Or upgrading this server to run ESXi or KVM and think you need a beefier CPU?

 

I'm using the cpu that's in there now for something else and will need an alternative cpu. I'm looking for opinions on what could go in that motherboard.

Link to comment

That motherboard supports  Intel® Xeon® E3-1200 v3, 4th gen. Core i3, Pentium, and Celeron processors.  I believe they all support ECC.

 

This is just my opinion... I'd only use a Celeron for basic NAS duties, no Dockers. A Pentium should be capable of some light Docker work and maybe single stream Plex transcoding, and I'd go up to a Core i3 if you want to run a bunch of Dockers or a couple of Plex streams.  If you want to run many Dockers or ones that tax the CPU (multiple Plex streams), I'd just stay with the Xeon honestly... You mention that you were planning to stay at 8GB, but also that you were planning to run many Dockers.  You can run a bunch of Dockers in 8GB, but keep an eye on it - those Dockers do require RAM...

Link to comment

That motherboard supports  Intel® Xeon® E3-1200 v3, 4th gen. Core i3, Pentium, and Celeron processors.  I believe they all support ECC.

 

This is just my opinion... I'd only use a Celeron for basic NAS duties, no Dockers. A Pentium should be capable of some light Docker work and maybe single stream Plex transcoding, and I'd go up to a Core i3 if you want to run a bunch of Dockers or a couple of Plex streams.  If you want to run many Dockers or ones that tax the CPU (multiple Plex streams), I'd just stay with the Xeon honestly... You mention that you were planning to stay at 8GB, but also that you were planning to run many Dockers.  You can run a bunch of Dockers in 8GB, but keep an eye on it - those Dockers do require RAM...

 

Thanks, that helps me a bunch on which cpu to look into. Currently I am only using about 11% RAM with my dockers that I've tried though if I need more I can always increase it later.

 

Though to anyone else reading this: If you have a specific cpu that'd work best that'd help a lot. After reading tdallen's post I am looking into the pentium family as long as it supports ECC. I don't plan to do much more than a single plex stream if I decide to use plex, so it should be fine.

Link to comment

First you should look at the watt usage of the cpus when they idle down. Nearly all of them idle at the same very low power draw, so you might not gain any advantage at all by switching only cpus. To gain real advantage you may also need to switch motherboards and cpu. You may save more by using a higher efficiency power supply.

Link to comment

First you should look at the watt usage of the cpus when they idle down. Nearly all of them idle at the same very low power draw, so you might not gain any advantage at all by switching only cpus. To gain real advantage you may also need to switch motherboards and cpu. You may save more by using a higher efficiency power supply.

 

If you have a good source to view idle wattage please let me know though that was an extremely difficult spec to find when I looked. As far as my power supply goes going from gold to platinum would cost me instead of save me money. Assuming my server used 500 watts 24/7 for an entire year it'd be about $1-$2 savings instead of gold which I am at now for way more than $1-$2 increase in price for that power supply. For my home case the cpu's TDP and idle temp would have a greater impact than going from gold to platinum for a power supply for now.

 

Now if I am very wrong in these numbers please let me know. I don't want to sound like I know everything, just what I've found when I looked that info up at the time.

Link to comment

I think the current CPU is/was a good choice,  4 cores, 8 threads, good amount of cache for virtualization.

 

if i really were looking to save heat and money (over time), I might choose a low voltage CPU.

 

E3-1220L like I have in my hp micro server.

You can consider E3-1260L (which I plan to add to my microserver).

They do have an E3-1230L, however I'm not sure of the availability nor cost.

 

Keep in mind these are specialty CPU's. I get them on eBay used.

There's a trade off, you pay extra for these CPU's. It will take time before you realize any savings.

However, savings in heat when you have to pay for electricity for air conditioning matter.

 

In years gone by, I saved allot of heat and thus cooling costs by using the LV XEONS of the time.

 

However, with what I've seen in usage patterns and recent needs, I think the current 1230 CPU is a good choice. 

I have an E3-1241 in my larger server.  E3-1220L's in my HP microservers.

 

Why the larger CPU?  Virtualization and file hash sums.

 

For a regular NAS, a medium powered CPU works fine, but when you want to virtualize, context switch allot and/or do file hash summing at a decent speed, the extra cache and power helps.

 

Take it from experience, the hash summing on my N54L's takes days whereas on the XEONs it finishes in less then a day.

It matters when you want to check the integrity of your system over time.

 

If cost of the CPU is a concern, grab a used one on eBay or from a member here.

 

My personal point of reference, if I am going to virtualize, go for the lowest CPU with the most cores, hyper-threading and 8M cache.

If you need to run virtual OS environments at near bare metal speeds, then use faster process with larger cache.

Same, if you have to hash hundreds of thousands of files to insure integrity.

 

FWIW, You may be able to idle the CPU through use of the kernel cpu governor.  Use to work well on the N54L AMD chips.

Having decent power available when you need it provides a cushion you may depend on later.

Link to comment

I think the current CPU is/was a good choice,  4 cores, 8 threads, good amount of cache for virtualization.

 

if i really were looking to save heat and money (over time), I might choose a low voltage CPU.

 

E3-1220L like I have in my hp micro server.

You can consider E3-1260L (which I plan to add to my microserver).

They do have an E3-1230L, however I'm not sure of the availability nor cost.

 

Keep in mind these are specialty CPU's. I get them on eBay used.

There's a trade off, you pay extra for these CPU's. It will take time before you realize any savings.

However, savings in heat when you have to pay for electricity for air conditioning matter.

 

In years gone by, I saved allot of heat and thus cooling costs by using the LV XEONS of the time.

 

However, with what I've seen in usage patterns and recent needs, I think the current 1230 CPU is a good choice. 

I have an E3-1241 in my larger server.  E3-1220L's in my HP microservers.

 

Why the larger CPU?  Virtualization and file hash sums.

 

For a regular NAS, a medium powered CPU works fine, but when you want to virtualize, context switch allot and/or do file hash summing at a decent speed, the extra cache and power helps.

 

Take it from experience, the hash summing on my N54L's takes days whereas on the XEONs it finishes in less then a day.

It matters when you want to check the integrity of your system over time.

 

If cost of the CPU is a concern, grab a used one on eBay or from a member here.

 

My personal point of reference, if I am going to virtualize, go for the lowest CPU with the most cores, hyper-threading and 8M cache.

If you need to run virtual OS environments at near bare metal speeds, then use faster process with larger cache.

Same, if you have to hash hundreds of thousands of files to insure integrity.

 

FWIW, You may be able to idle the CPU through use of the kernel cpu governor.  Use to work well on the N54L AMD chips.

Having decent power available when you need it provides a cushion you may depend on later.

 

Thanks for the help. Since I will eventually fill this chassis up with 21 active 4TB drives does that mean for every drive I add in it'll increase the time for parity and a faster cpu would help speed it up? Also due to my VMs and dockers going off when I take the array down I have to offload virturalization on to a specific server meant for that that won't suffer from that problem. Since the NAS won't virtualize much at all once the main server's built it'll mostly be dockers and maybe a virtual machine for some reason though I can't think of why I'd run one if I'd have a main server meant for that purpose.

 

Also what would tax the cpu the most besides virtual machines? Is there a way to gauge from the list of popular dockers?

 

I'll also take a look in the L family of xeons. I forgot about those.

Link to comment

Thanks for the help. Since I will eventually fill this chassis up with 21 active 4TB drives does that mean for every drive I add in it'll increase the time for parity and a faster cpu would help speed it up? Also due to my VMs and dockers going off when I take the array down I have to offload virturalization on to a specific server meant for that that won't suffer from that problem. Since the NAS won't virtualize much at all once the main server's built it'll mostly be dockers and maybe a virtual machine for some reason though I can't think of why I'd run one if I'd have a main server meant for that purpose.

 

Also what would tax the cpu the most besides virtual machines? Is there a way to gauge from the list of popular dockers?

 

I'll also take a look in the L family of xeons. I forgot about those.

 

21 4TB drives... If you plan to do hash sum checks on these drives, the faster CPU will help.

 

You can possibly do one whole drive's validation a day for each day of the month, Parity on the 27th day and you will be golden.

Parity will not be any faster with the CPU.  parity checking is IO bound.

 

I can't tell you about CPU requirements for Dockers.

Only my experience with running Windows VM's in an ESX virtual environment. Fast cpu and large cache help.

 

Other then that the faster cpu with cache would aid in doing the hashing.

 

Also if you set a cpu governor in the kernel, the CPU drops to low idle wattage when doing nothing.

 

I would suggest getting a kill-o-watt meter, kill-o-watt power strip or a UPS that shows how much power is being utilized via software or LCD screen.

 

While you may be painstakingly selecting a CPU, 24 idle drives is probably going to pull a fair amount of power as well.

In this case, consolidation onto 6TB and/or 8TB in the future may help lower electricity and heat dissipation requirements.

 

With today's CPU's throttling so well, I'm not sure how much a LV XEON is going to help other then reducing maximum output.

 

The 1220L's in my HP microservers are passively cooled, but they are only 2 cores at a max of 2.3. 

Certainly enough for NAS and/or file hashing. I'm not sure about the transcoding.

 

With transcoding, you need good speed and cache to provide a continuous stream.  Garycase may have more input on that one.

Link to comment

Thanks for the help. Since I will eventually fill this chassis up with 21 active 4TB drives does that mean for every drive I add in it'll increase the time for parity and a faster cpu would help speed it up? Also due to my VMs and dockers going off when I take the array down I have to offload virturalization on to a specific server meant for that that won't suffer from that problem. Since the NAS won't virtualize much at all once the main server's built it'll mostly be dockers and maybe a virtual machine for some reason though I can't think of why I'd run one if I'd have a main server meant for that purpose.

 

Also what would tax the cpu the most besides virtual machines? Is there a way to gauge from the list of popular dockers?

 

I'll also take a look in the L family of xeons. I forgot about those.

 

21 4TB drives... If you plan to do hash sum checks on these drives, the faster CPU will help.

 

You can possibly do one whole drive's validation a day for each day of the month, Parity on the 27th day and you will be golden.

Parity will not be any faster with the CPU.  parity checking is IO bound.

 

I can't tell you about CPU requirements for Dockers.

Only my experience with running Windows VM's in an ESX virtual environment. Fast cpu and large cache help.

 

Other then that the faster cpu with cache would aid in doing the hashing.

 

Also if you set a cpu governor in the kernel, the CPU drops to low idle wattage when doing nothing.

 

I would suggest getting a kill-o-watt meter, kill-o-watt power strip or a UPS that shows how much power is being utilized via software or LCD screen.

 

While you may be painstakingly selecting a CPU, 24 idle drives is probably going to pull a fair amount of power as well.

In this case, consolidation onto 6TB and/or 8TB in the future may help lower electricity and heat dissipation requirements.

 

With today's CPU's throttling so well, I'm not sure how much a LV XEON is going to help other then reducing maximum output.

 

The 1220L's in my HP microservers are passively cooled, but they are only 2 cores at a max of 2.3. 

Certainly enough for NAS and/or file hashing. I'm not sure about the transcoding.

 

With transcoding, you need good speed and cache to provide a continuous stream.  Garycase may have more input on that one.

 

EDIT: Looks like I'm going to keep the cpu in the server. I just ran bunker to calculate SHA256 hashes on my main share and it is using 36-40% of my cpu from 2-3%. I can only imagine what a pentium would be like on a near full 4TB hard drive or worse, a 50TB+ share. Since this isn't the place to talk much about this I'll end the thread here.

 

I'm surprised an idle hdd would take much more than a watt or two. I didn't think it'd matter much when it's not running at all.

 

At this point I guess I'll just keep it simple and either keep the cpu in there (expensive one at $206) or go with a similar, cheaper one. I may do something similar with he 500GB SSD for a cache. Instead of using that internally in the virtual machine server to just keep it in this one and buy another 500GB or just use it over the network. I'm sure .5-1ms more isn't going to matter too much for home use.

Link to comment

EDIT: Looks like I'm going to keep the cpu in the server. I just ran bunker to calculate SHA256 hashes on my main share and it is using 36-40% of my cpu from 2-3%. I can only imagine what a pentium would be like on a near full 4TB hard drive or worse, a 50TB+ share. Since this isn't the place to talk much about this I'll end the thread here.

 

I'm surprised an idle hdd would take much more than a watt or two. I didn't think it'd matter much when it's not running at all.

 

At this point I guess I'll just keep it simple and either keep the cpu in there (expensive one at $206) or go with a similar, cheaper one. I may do something similar with he 500GB SSD for a cache. Instead of using that internally in the virtual machine server to just keep it in this one and buy another 500GB or just use it over the network. I'm sure .5-1ms more isn't going to matter too much for home use.

 

I'll remind you of this, you would be surprised at all the lil incidentals. fans, etc, etc.

I would suggest getting a kill-o-watt meter, kill-o-watt power strip or a UPS that shows how much power is being utilized via software or LCD screen.

At 24 drives, it starts to add up. Frankly, I couldn't tell you how much it would be for your system.

My older 20 drive server at 'near' idle would pull almost 100watts.

 

Different architecture, still it added up. The kill-a-watt or UPS would be helpful. (you do have a UPS right?).

I believe some APC models report how much power is being utilized.

 

At this point I guess I'll just keep it simple and either keep the cpu in there (expensive one at $206)

I don't find that CPU all that expensive for what it's capable of doing. When you add in the cost of cleaning the CPU to change it, then the arctic silver, or whatever product is utilized, you start to see the tiny savings of just leaving it in and/or getting a used CPU for the other task.

 

I did a study once of all incidentals, and started to see how things climb when you need to replenish supplies and/or order them without free shipping.

 

For such a large server, the CPU will come in handy for the file hashes.  If you get your hands on a low voltage CPU, then I would say go for it.

Link to comment

EDIT: Looks like I'm going to keep the cpu in the server. I just ran bunker to calculate SHA256 hashes on my main share and it is using 36-40% of my cpu from 2-3%. I can only imagine what a pentium would be like on a near full 4TB hard drive or worse, a 50TB+ share. Since this isn't the place to talk much about this I'll end the thread here.

 

I'm surprised an idle hdd would take much more than a watt or two. I didn't think it'd matter much when it's not running at all.

 

At this point I guess I'll just keep it simple and either keep the cpu in there (expensive one at $206) or go with a similar, cheaper one. I may do something similar with he 500GB SSD for a cache. Instead of using that internally in the virtual machine server to just keep it in this one and buy another 500GB or just use it over the network. I'm sure .5-1ms more isn't going to matter too much for home use.

 

I'll remind you of this, you would be surprised at all the lil incidentals. fans, etc, etc.

I would suggest getting a kill-o-watt meter, kill-o-watt power strip or a UPS that shows how much power is being utilized via software or LCD screen.

At 24 drives, it starts to add up. Frankly, I couldn't tell you how much it would be for your system.

My older 20 drive server at 'near' idle would pull almost 100watts.

 

Different architecture, still it added up. The kill-a-watt or UPS would be helpful. (you do have a UPS right?).

I believe some APC models report how much power is being utilized.

 

At this point I guess I'll just keep it simple and either keep the cpu in there (expensive one at $206)

I don't find that CPU all that expensive for what it's capable of doing. When you add in the cost of cleaning the CPU to change it, then the arctic silver, or whatever product is utilized, you start to see the tiny savings of just leaving it in and/or getting a used CPU for the other task.

 

I did a study once of all incidentals, and started to see how things climb when you need to replenish supplies and/or order them without free shipping.

 

For such a large server, the CPU will come in handy for the file hashes.  If you get your hands on a low voltage CPU, then I would say go for it.

 

I don't have a UPS yet though that's near the top now. Main thing is heat since summer's here. My junk hard drives I'm learning with at around 40F idle and I need to install an AC unit and block the sun from coming in to keep the room at 70F. After that I'll buy a UPS. I'd rather not buy multiple which is why I'm holding off. I want one main one than many smaller ones to keep it simple.

Link to comment

You need to consider that an AC unit will create noise on the electrical line if they are in the same room.

This is where a UPS will help as well.

 

 

Also you might be able to use the Kernel CPU idle governor to save some heat and/or adjust the bios to limit max speed or number of cores.

Link to comment

You need to consider that an AC unit will create noise on the electrical line if they are in the same room.

This is where a UPS will help as well.

 

 

Also you might be able to use the Kernel CPU idle governor to save some heat and/or adjust the bios to limit max speed or number of cores.

 

I have two separate clean lines running to my room for such purposes. I wanted to make sure I only ran my rack off of its own line.

Link to comment

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.

Guest
Reply to this topic...

×   Pasted as rich text.   Restore formatting

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.