suggestions for a low power system to replace that old dual xeon macpro

[joeinaz]

Is Intel enough of a source for you?

https://www.intel.com.au/content/www/au/en/support/articles/000031072/processors/intel-core-processors.html
Thermal Design Power (TDP) is the maximum amount of heat that a processor can produce when running real life applications. It is used mostly to match up processors with an adequate heat sink that is capable of cooling down that processor effectively.

TDP is NOT performance per watt, which is what you are looking for to compare.

In the end if you think the experts here are wrong, more power to you.

Buy what you want. Being wrong doesn't mean it won't function, only that it will have a higher power usage per work done.

"L" CPUs are for lower TDP cases/enclosures, not that they use less actual power per work. They tend to be higher in price due Intel not making as much volume of these parts.

Intel® Xeon® Processor E5-2630L v4 (25M Cache, 1.80 GHz) Product Specifications

Intel® Xeon® Processor E5-2630L v4 (25M Cache, 1.80 GHz) quick reference guide including specifications, features, pricing, compatibility, design documentation, ordering codes, spec codes and more.

ark.intel.com

From the ? around the TDP in the attached document is a more formal description of TDP:

TDP
Thermal Design Power (TDP) represents the average power, in watts, the processor dissipates when operating at Base Frequency with all cores active under an Intel-defined, high-complexity workload. Refer to Datasheet for thermal solution requirements.

So Intel determines their TDP based on a Intel defined workload running at BASE frequency which uses all the cores;

"TDP is NOT performance per watt, which is what you are looking for to compare."
I never said it was. It seems to be how Intel has chosen to describe the thermal efficiency of their processors. For comparison, if you divide the benchmark results by the Intel PUBLISHED TDP you CAN arrive at a work/watt estimate.

"In the end if you think the experts here are wrong, more power to you."
What experts?

"In such an environment, you should find that -- given the same generation silicon and all the other expected qualifiers -- the L CPU's tend to be a little less efficient on watt burn per unit work. Intel's got some application note about it somewhere."
I don't think you've shown this with any numbers. On the other hand the benchmarks and the TDP numbers point to a different conclusion.

E5-2630 V4: Passmark score/TDP : 13902/85 = 163.55 units/w
E5-2630L V4: Passmark score/TDP : 12847/55 = 233.58 units/w

This discussion on the low power CPUs has gone astray. Obviously, if power and cooling are not a concern, the low power CPU is not the best choice because the regular CPU will provide more computes. My own system has an E5-2630-V4 (ES) because the 2630L was too expensive for me right now.

The key to any solution is the customer need. In this case you simply have to look to the title of this thread to see power IS a concern and the interest is in finding a LOW POWER SYSTEM. There was also an interest in running some virtualization along side the FreeNAS. Given this, I set out to recommend a LOW POWER SYSTEM which may meet his needs. I hope you can help do the same.

 

[John Doe]

I remember the same, TDP is an indication (which not necessarily be correct!) to design your cooling solution.

in devices, where you cannot independent scale up cooling systems, you need to limit the heat (cpu clock). that is typically the mobile or L cpu.
So the L cpu will have a lower TDP and lower clock but takes longer to do the work compared to the faster brother.

it has been figured out, that the "run to idle" is the current way to go in order to safe energy.
Have higher clocks to do the operations and get to back to idle clock. Basically every modern smartphone is doing that.

now it depends on the scenario. if you have an application what would not be in line with the "run to idle" you cannot benefit from it and it that case the L version might be more suitable.
From what I understood of the requesting thread starter, i doubt that this will be the case. honestly I have not seen any application in real life.

 

[hotdog]

3. It turns out for this customer, power consumption is everything; in their case, even the 60w 2630L V2 may be too "power hungry"..

Don't get me wrong. I don't insist on the 50W. Let me shortly draw a clearer picture:
At the moment I have some spare time so I decided to look into our infrastructure, as our yearly bills are quite high. So I asked my partner for "permission" to replace the old MacPro with a less power hungy thing. He's first reaction was: why? I answered something about energy consumtion. He then made some quick calculations himself but the numbers did not speak for themselves.
Anyway, I finally "got" him telling him I will invest about 2000 bucks and build a machine drawing a FRACTION like a forth or a fifth of the old one and that the HDDs are likely to fail sometime and our current system drive is too small for updating freenas to a version where the encrypted backup is possible. Shortly ALL I WANT is hold my word :)
Unfortunately I really can't remember the wall measuring we've done a year ago.. I will do it again and report back. What I remember is: I first looked up the internet for the estimated power consumtion of our Mac, was shocked and expected something around 700W, decided to verify and it looked like "false alarm" (promised, I WILL measure again as soon as I am near machine again)..

BTW: About that micro windows VM I am talking: at the moment there is a old workstation running only for distributing the floating licences. This machine will be powered off when the new server will be life, so again, much improvement here.
This also could be done by let's say a RaspberryPi but I don't think this is necessary while having a freenas box standing around.

 

[hotdog]

I am surely not a fanatic about energy consumtion:

This was my last workstation. 6 of the 7 gpus (2 are sitting directly on the MB) where basically idleing a good portion of the time.

I transformed it into that one a year ago, capable of holding 4 "airy" GPUs. Only the i9 9920X from silicon lottery is water cooled and slightely overclocked. You can't really hear the machine in an office environment:

 

[jgreco]

That is not how the world works.

Flat wrong.

Industry benchmarks measure how much work was done when CPU was stressed with a given test.

Correct. That is one part of the ratio (the ratio being operations per joule or other equivalent work per power unit ratios). Measuring the actual power consumed is the other half. Not this B.S.:

You can then claim a certain amount of work was obtained with a CPU that has a maximum TDP of x.

You can claim whatever the hell you want. There are people who claim all sorts of wrong things, and unfortunately in many cases, the actual truth is too complex or too abstract for them. This shouldn't be one of those things, though. This is pretty simple and is measurable with the proper gear.

Ok, I will ask you the same thing you asked me: cite;

I'm not interested in doing so. I do this stuff professionally. I read tons of stuff and I don't keep track of all the things I read so that I can create citations for random forum participants who are wrong about basic electrical engineering and computer science. There's no value to me "proving" something that is obviously correct; I've provided an explanation of why you need to use energy consumption in the ratio.

 

[Newfoundland.Republic]

Outside of the low power/TDP discussion... ;)

Have you considered something like an Intel Atom C3758? That is what iX uses in their FreeNAS Mini XL+.

You can get a board/CPU combo on eBay (https://www.ebay.ca/itm/Supermicro-...963480?hash=item3b3b2e5858:g:VEcAAOSwNopdrfEQ) for US $183.00...

 

[hotdog]

FWIW, I would not boot from a USB stick for a production server. Use something built for the purpose, i.e. a SSD or a SATADOM. I've read too many stories of woe here related to failed USB sticks to ever want to use one, even in a mirrored boot pool, on a production server. Using SSDs for the data pool is also pretty prohibitive at this point unless the work flow justifies it. If I needed low power and was OK with the related performance tradeoff, I'd stick to the 2.5" spinners the OP mentioned.

That said, the random write performance of that 2019 DIY NAS linked to above is nothing to be particularly happy about. Even with a 2-VDEV system, you'd be looking at maybe 60MB/s per that test. I wonder how applicable Brians testing/setup is to the OP use case.

SATADOM it will be. The VMs can sit there, too, i guess? Or would you consider a separate ssd?
About that random write performance: you think this is coming from the 2.5 drives Brian is using? Or where is the bottleneck there?

 

[hotdog]

Outside of the low power/TDP discussion... ;)

Have you considered something like an Intel Atom C3758? That is what iX uses in their FreeNAS Mini XL+.

You can get a board/CPU combo on eBay (https://www.ebay.ca/itm/Supermicro-...963480?hash=item3b3b2e5858:g:VEcAAOSwNopdrfEQ) for US $183.00...

Your link is pointing to a C2758, isn't that one affected by the C2000 bug? I had no time yet to read much about it, but I am trying to avoid such stuff by simply not risking it.

 

[hotdog]

Depends in part on the cost of power, peak demand pricing, cooling considerations, and so on.

I needed a new server because I lost faith in AsRock after my second replacement failed. With 8 spinning disks, a SLOG, and a L2ARC, about 90W of power consumption is decent.

I can't see the problem:
90W - 8x7W (your disks) + 6x1.1W (my future 2.5 Barracudas idleing) = 41W :)

or did I miss something?

 

[joeinaz]

I can't see the problem:
90W - 8x7W (your disks) + 6x1.1W (my future 2.5 Barracudas idleing) = 41W :)

or did I miss something?

The Intel Atom based solution is great especially for a FreeNAS appliance that does just FreeNAS. The challenge maybe adding applications and future expansion. Is there enough horsepower (and DIMM slots) within the Atom motherboard to allow for virtualization and growth? The benefit of a discrete motherboard is the ability to have a wide range of choice in terms of CPU and memory capacity. If I outgrow a CPU, I can swap it with one that has more capability? How much memory can you add?

If you go the Atom route for FreeNAS, there is no reason you could not also deploy a secondary small server for virtualization and other applications and still come in much lower in power than your 11 year old Mac. Two small servers may also offer the benefit of a simpler deployment and would remove the FreeNAS system as a single point of failure. It's something to consider.

 

[Newfoundland.Republic]

Your link is pointing to a C2758, isn't that one affected by the C2000 bug?

I think previous stepping were affected:

The official errata says the B0 stepping of C2xxx Atoms are vulnerable to failure, and these parts began shipping in 2013.

See here.

 

[Newfoundland.Republic]

If you go the Atom route for FreeNAS, there is no reason you could not also deploy a secondary small server for virtualization and other applications and still come in much lower in power than your 11 year old Mac. Two small servers may also offer the benefit of a simpler deployment and would remove the FreeNAS system as a single point of failure. It's something to consider.

Good point! Some light plugins for the FreeNAS box and another one for, say, virtual machines? Use NFS or iSCSI to mount the storage on FreeNAS

 

[hotdog]

Hmm. Now options are growing. Not good for decisions... ;)

 

[joeinaz]

Hmm. Now options are growing. Not good for decisions... ;)

I just wanted to point out in an Atom based solution, the CPU potential growth path is limited. In your case it's simple decision point with as to deploying one system or two and then proceeding from there. What we want to avoid is the deployment of a solution that is too small and then having to go back to management for additional funding. I have been in those customer meetings before and it's no fun...

 

[hotdog]

I just wanted to point out in an Atom based solution, the CPU potential growth path is limited. In your case it's simple decision point with as to deploying one system or two and then proceeding from there. What we want to avoid is the deployment of a solution that is too small and then having to go back to management for additional funding. I have been in those customer meetings before and it's no fun...

In this case it's basically me in half managment position :)
basically I like the X10SDV-2C-7TP4F which Constantin was talking about for it's features. A socket based cpu would have been nice on that board...

 

[joeinaz]

Flat wrong.



Correct. That is one part of the ratio (the ratio being operations per joule or other equivalent work per power unit ratios). Measuring the actual power consumed is the other half. Not this B.S.:



You can claim whatever the hell you want. There are people who claim all sorts of wrong things, and unfortunately in many cases, the actual truth is too complex or too abstract for them. This shouldn't be one of those things, though. This is pretty simple and is measurable with the proper gear.



I'm not interested in doing so. I do this stuff professionally. I read tons of stuff and I don't keep track of all the things I read so that I can create citations for random forum participants who are wrong about basic electrical engineering and computer science. There's no value to me "proving" something that is obviously correct; I've provided an explanation of why you need to use energy consumption in the ratio.

"I do this stuff professionally."
I do too; I have been for almost 40 years. My installations have ranged from VAX systems all the way to latest hardware and software. I have installed and maintained complex solutions such as clustered 8 way servers in a SAP deployment to IBM Power8 installations to enterprise storage solutions. Numerous time in my career, I have receive a call saying a customer's server is down how soon can you be on site?

Today, I am currently in my lab and I manage about 30 running servers and some network attached storage.

"I read tons of stuff and I don't keep track of all the things I read so that I can create citations for random forum participants who are wrong about basic electrical engineering and computer science. "

I do more than "read tons of stuff"; I hold numerous industry technical certifications including: IBM Power9, HPE Hybrid solutions, HPE and IBM enterprise storage, Cisco, VMware, Lenovo Servers and other smaller vendors.

I feel I am a bit more that a "random forum participant"...

Now back to today, challenge, setting up a tagged VLANs on a layer 3 switch to separate phone traffic on the companies new phone system and updating firmware on some of my servers...

 

[joeinaz]

In this case it's basically me in half managment position :)
basically I like the X10SDV-2C-7TP4F which Constantin was talking about for it's features. A socket based cpu would have been nice on that board...

Well that's half the battle won! :)

 

[hotdog]

Well that's half the battle won! :)

:)

BTW: Are boards with RJ45 10Gb copper ports fully backward compatible?

 

[joeinaz]

No; 10GBaseT has some backward compatibility with caveats.

1. A minimum of Cat6A cable. A quality Cat6 cable should work at a shorter distance but its better to go with Cat 6A. Cat5 is not recommended.
2. Check your end device. While gigabit networks are often compatible with 10/100 devices not so with 10GbaseT. 10/100/1000 devices should be fine. The problem may be with a 10/100 device.

 

[jgreco]

"I do this stuff professionally."
I do too; I have been for almost 40 years.

Great. Me too.

My installations have ranged from VAX systems all the way to latest hardware and software. I have installed and maintained complex solutions such as clustered 8 way servers in a SAP deployment to IBM Power8 installations to enterprise storage solutions. Numerous time in my career, I have receive a call saying a customer's server is down how soon can you be on site?

That's nice. Spent more than two decades as "the buck stops here" guy, including being on call as the ultimate problem resolver. Works out to being constantly on-call. Pager, phone, two laptops, always within reach, always reachable. BTDT. Yawn.

Today, I am currently in my lab and I manage about 30 running servers and some network attached storage.

Great.

"I read tons of stuff and I don't keep track of all the things I read so that I can create citations for random forum participants who are wrong about basic electrical engineering and computer science. "

I do more than "read tons of stuff"; I hold numerous industry technical certifications including: IBM Power9, HPE Hybrid solutions, HPE and IBM enterprise storage, Cisco, VMware, Lenovo Servers and other smaller vendors.

Well, I thought it'd be clear from context that I do more than just read.

I feel I am a bit more that a "random forum participant"...

You have a few hundred posts and what appears to be a software guy's misconceptions about TDP.

I came into this through the electrical engineering side of things, and one of my companies still maintains an electronics shop, even of most of what goes on here these days tends to be data center server engineering.

I read technical documentation and application notes out of necessity. And I know what TDP means out of necessity.

Since you seem to find what I've written to be inaccessible, I did look and watched several minutes of this to convince myself that this information was headed in the general right direction:


I have not watched the whole thing since it is BORING. To me. But if you can find that guy saying that "TDP is an accurate representation of power consumed for unit work purposes", by all means let me know the time index.

Now back to today, challenge, setting up a tagged VLANs on a layer 3 switch to separate phone traffic on the companies new phone system and updating firmware on some of my servers...

Well that sounds complicated, but all vlans are tagged - definition of the beast, the concept of "virtual lan" is implemented in silicon as tag filtering and switching - it's just a matter of how you ingress and egress frames. Layer 3 switching? Yawn. Some of us run hundreds of vlans in production.

Updating firmware, okay, well, that can be teh crap.