Server Duty: cMP vs. Mac mini 2018

[fhturner]

I've made no secret of my opinion about the (over)pricing on the new Mac mini in other threads, and my own (personal + consulting company) tally of Mac Pros is now up to 6 (1 x 2006, 1 x 2007, 1 x 2008, 2 x 2009 (4,1>5,1), 1 x 2010)...so clearly my thought process is always leaning toward the big, beautiful, incredibly flexible towers. In addition to my own personal uses for video, utility, and serving, I have also deployed nearly a dozen to various clients for server duty as well.

As I started working on a proposal for a client's expansion from single user to new office, I was trying to balance my cMP affection w/ practical considerations, including cost efficiency, expandability, and future OS support. When I was looking at the components I'd like to have, I realized I needed/wanted 5 PCIe slots' worth w/ only 4 available slots:

• NVMe [for boot]
• GPU [GT120 would be fine...just need display occasionally]
• RAID [probably Areca-based internal SAS setup from MaxUpgrades]
• 10Gb Ethernet
• USB3 [for backup drives and other file transfer]

Oops, 1 too many. NVMe boot is not critical, just nice to have, esp since we can do it officially now! But I'd just have to do w/o that. So then I got to thinking about the opposite end of the spectrum, the new Mac mini. It pales in expandability, right? Except...it needs no slots (or external TB chassis, etc.) to provide:

• GPU [and the weakness of that doesn't matter in *this* case]
• USB3
• 10GbE (when optionally configured for $100)

And it's already got NVMe boot. So that just leaves RAID, which could pretty effectively be handled by something like a ThunderBay or Pegasus. Hmmm.

For further background, I'd likely be building up a 2009 cMP with 6-core X5680/90 and 32GB RAM; the Mac mini I'd consider would be the 6-core as well w/ same RAM.

Of course, the main problem w/ the Mac Pro now in 2018 is that we have to "waste" several PCIe slots for modern functionality that's included in current systems. Don't get me wrong, it's fantastic we can still add these technologies 8-9-10 years later, but they unfortunately limit further improvement unless you start looking at a PCIe expansion chassis— which I have done, but would be unnecessarily complex and overkill for this job.

Cost-wise, I can get to roughly equal capability for less money w/ the cMP, but since I'm reselling this to a client and the refurbishment and prep involves my time, I need to include a reasonable markup in the sale. I'll try not to make this long post overly longer, but in a nutshell, I'm comparing these approx. prices to the client:

• cMP + X5680/32GB/250GB 860 EVO/X520 DA2 10GbE/MaxUpgrades Areca RAID/KT-4004/SoftRAID: $2002
• Mac mini 2018 + i7 6cx/32GB/256GB SSD/10GbE + ThunderBay w/ SoftRAID: $2278

So, only around $300 apart... Again, the cMP is w/ my refurbishment, cleaning, testing, CPU upgrade, X520 flashing, and some assembly included. Could probably cut some corners and bring that down a little, and if it were for my own uses, my time would be "free", of course. Comparing that way (my cost alone), the cMP would be around 35-40% less.

Anyway, this has been kinda eye-opening for me, given this use case. I'm still leaning toward the cMP, as it still has more configurability and internal capacity, but the mini will have a better-performing CPU, quicker boot SSD, and longer support from Apple. Not really looking for a definitive answer here, but would love to hear other arguments, counters, and input here. What are your pros/cons for each? How would you adjust the config of either to make it better suit this type of use or to be more cost effective?

Thanks!
Fred

 

[eksu]

I’ve decided rather than getting a 10GB Ethernet NIC on my Mac Pro, I’m going with Apple RAID, Thunder Bolt add-in card, Highpoint Adapter & a GPU.

I have a router that can do link aggregation between the two 1GB Ethernet ports, but I also wonder what sort of bandwidth/performance a 10GBE to Thunderbolt adapter would have on Slot 3 with the Apple Raid in Slot 4.

 

[orph]

i dont know much about it but did you see the video on the macmin server farm

we know where the mac mins go now

 

[fhturner]

i dont know much about it but did you see the video on the macmin server farm

we know where the mac mins go now

Yeah, I actually have a 2010 mini in MacStadium's ATL facility. Used to have a 2009, but they made me swap it out so it wasn't the old form factor. Needed a 2010 so I could still run Mac OS X 10.6.8 for mail server duty...anything newer wouldn't run 10.6.8 reliably.

 

[orph]

i relay like the look of the mac min, if it was 10-20% cheaper for the 6core unit (or sold with no ram? with price cut) id think of it as a amazing deal.

if your providing long on going support and repairs the macpro can be DIY fixed with parts that dont cost 'to much' while if the macmin's id gess is like imacs for repairs ie same price as a new one almost so maybe apple care is worth it?
if you are able to swap the clients unit with a spare or new unit while old one is fixed you can then use "repaired" unit for next client?

space and power use is a plus, but some times client's are easily impressed by big things

one problem may be the macmin is so small&light it may be more easy for "client's" to brake it not shore what skill level your client's are

is it worth re selling the macmin ram? may be able to get some cost back

 

[AidenShaw]

As I started working on a proposal for a client's expansion from single user to new office,...

What will the system be doing?

If it's a file server, six cores and 32 GiB of RAM might be an unnecessary expense.

Does it need 10 GbE? Since few Apple systems have 10 GbE ports, you'll sink a bunch of bucks on a 10 GbE switch and 10 GbE dongles for all of the clients.

 

[chrfr]

What will the system be doing?

If it's a file server, six cores and 32 GiB of RAM might be an unnecessary expense.

Does it need 10 GbE? Since few Apple systems have 10 GbE ports, you'll sink a bunch of bucks on a 10 GbE switch and 10 GbE dongles for all of the clients.

Just going 10GbE from the server to the switch and then 1Gb to all the clients can offer a pretty good performance increase itself. Making everything 10Gb isn't a necessity.
If I were putting just a file server together, I definitely wouldn't spend the money on 32GB of RAM or even an i7 CPU.
I also wouldn't be putting in any new servers based on old Mac Pros at this point. They're old and among the ones I still have in service at work, I am starting to see an increase in failures.

 

[fhturner]

What will the system be doing?

If it's a file server, six cores and 32 GiB of RAM might be an unnecessary expense.

Does it need 10 GbE? Since few Apple systems have 10 GbE ports, you'll sink a bunch of bucks on a 10 GbE switch and 10 GbE dongles for all of the clients.

Will be primarily file serving and backup of itself and client stations via Retrospect. A 4-stick 32GB kit is easy enough to acquire and implement for the Mac Pro, so while 16GB would likely be sufficient, I like to give myself a fair bit of headroom when possible.

Same with 6 cores. While 4 would probably suffice, I like being able to "use" (the system will obviously dictate) a couple for basic system tasks, a couple when server duties and file transfers kick up, and another couple for Retrospect threads.

Can probably make do with built-in pair of 1Gbps Ethernet ports, but since I have some luxury in this case to "overbuild", I'd like to have a nice, fat pipe into the server for when simultaneous file transfers and client backups are occurring. Not so much hitting 1GB/s or something silly-fast to one station, but handling a multitude of client traffic needs at once.

In short, I could get by w/ less of either configuration, but I'd rather overbuild and have plenty of headroom. Make sense?
[doublepost=1541275809][/doublepost]

Just going 10GbE from the server to the switch and then 1Gb to all the clients can offer a pretty good performance increase itself. Making everything 10Gb isn't a necessity.

Yes, I've done this in a couple of setups and really like it. 10GbE to most stations is not necessary, but allowing several to max out a 1Gbps link at once w/o choking the server is ideal.

I also wouldn't be putting in any new servers based on old Mac Pros at this point. They're old and among the ones I still have in service at work, I am starting to see an increase in failures.

But. I. Like. Them. Too. Much!

 

[weckart]

I’ve decided rather than getting a 10GB Ethernet NIC on my Mac Pro, I’m going with Apple RAID, Thunder Bolt add-in card, Highpoint Adapter & a GPU.

For the cMP? what is this card?

 

[AidenShaw]

Yes, I've done this in a couple of setups and really like it. 10GbE to most stations is not necessary, but allowing several to max out a 1Gbps link at once w/o choking the server is ideal.

This reasoning assumes that the pipeline to the disk volume is fatter than 1 Gbps. Parity SoftRAID to a set of spinners will probably be slower than a single 1 GbE link - so the 10 GbE infrastructure (even if just a 1 GbE switch with a 10 GbE uplink) could be wasted expense.

It really depends on the price sensitivity of your client. Would they rather spend a bit more and have some headroom, or keep the initial cost low and upgrade as needed?

I also would 100% go with the MiniMac option. The cMP towers are big, noisy, power-hungry, and getting into the end-of-life scene where OS support and hardware failures are becoming more and more of an issue.

 

[fhturner]

This reasoning assumes that the pipeline to the disk volume is fatter than 1 Gbps. Parity SoftRAID to a set of spinners will probably be slower than a single 1 GbE link - so the 10 GbE infrastructure (even if just a 1 GbE switch with a 10 GbE uplink) could be wasted expense.

It really depends on the price sensitivity of your client. Would they rather spend a bit more and have some headroom, or keep the initial cost low and upgrade as needed?

I also would 100% go with the MiniMac option. The cMP towers are big, noisy, power-hungry, and getting into the end-of-life scene where OS support and hardware failures are becoming more and more of an issue.

I am mainly concerned w/ where the cMP and Mac mini differ. Decision for or against 10-gig Ethernet will be made independently of which machine I use.

As far as saturating the link, I'm not terribly concerned w/ that, but I'm sure I can saturate a ONE Gbps Ethernet link, even with a single HDD. And as I said above, I'm looking to support "simultaneous file transfers and client backups"...this will include file sharing likely from a SATA SSD RAID, as well as backups to USB3 HDDs. Let's just assume for now that I will implement a 10GbE solution in both a cMP and a Mac mini 2018.

Client won't be overly price sensitive. No penny-wise/pound-foolish here...just nothing outrageous. This is part of a larger office build-out, so minimizing the initial cost isn't critical. Again, just within reason (which I think both cMP and Mac mini options I'm comparing are).

You're absolutely right about the point on the "lifetime arc" for the cMP...still hard to tear myself away tho, because they can do so much and continue to improve! The size and power draw, I don't think will make that much difference in this wiring closet location. I've never found my cMPs to be noisy, unless under sustained full load. I've always thought they did a nice job w/ fan speed and layout, as well as thermal zones. The Power Mac G5s were a different story...

Thanks for your input!

 

[PianoPro]

If I were the client and you offered me the choice of a 9-year old Mac versus a 9-day old Mac solution I wouldn't have to think very long. If you told me the 9-year old Mac may never be able to run any future Mac OS's I wouldn't have to think at all.

I've been limping along trying to do only necessary upgrades to my Mac Pro waiting for the 7,1 MP. But with it still a year (or more) out I'm seriously considering just switching at least temporarily to the Mac mini. I'm just waiting for 2 pieces of information - how much is the i7 CPU throttled in its cooling-challenged case, and how much do I have to tear that case apart to upgrade the RAM?

For my design work I would just buy the i7/8GB with the minimum 128 GB SSD ($1099) for a boot disk, 32 GB OWC RAM ($330 now), and add a $400 ThunderBay TB3 4-Bay enclosure or equivalent, and perhaps an additional 4-slot NVMe TB3 enclosure ($320). Total = $2150 (w/NVMe enclosure). I'm guessing the MP 7,1 will start at $5000 minimum with a 6-core or 8-core uP, and a lot less internal storage space than I just described. (I'd expect a fully loaded MP 7,1 to go north of $15,000 based on the iMac Pro pricing.) It doesn't apply to your situation, but when I upgrade to a new MP I'll move over the external storage and just add a cheap external TB3 SSD to the Mini and use it for another application.

 

[zipgs]

Unless they severely crippled the CPU in the mini, even the i3 would outperform the 6-core X5680, it shouldn't even be close. RAM is also much faster. For your case, I'd go with the i3 and 16GB of RAM, the RAM will drop in price, if they need it upgrade it later.

Don't get me wrong, I love the Mac Pro, but I think this is something the mini is certainly more than capable, and is the obvious choice, plus it looks so awesome, why wouldn't you want to get your hands on one now!


But since you like the Mac Pro idea, I'll give you a few reasons to go with the Mac Pro. In your case, some advantages of the Mac Pro are ECC RAM, and hardware RAID, making this more server grade for what it's worth. I think the older hardware probably negates this advantage however.

With noone even using the new Mini yet, if this server is critical to the business, there may be a little wisdom in sticking with what you know, and can support easier, since I'm assuming you're going to support it, it sounds like you have some of the parts available if something goes down, will you still have those parts a year from now? That's a big advantage when people just need things to work and are going to depend on one machine as it sounds like they are. To get the Mac Mini back up, for the next year or so, I'm sure you could walk into an Apple store and grab a new one and be back up in the same amount of time or as repairing the Mac Pro, but warranty repair may take longer. At some point you may need to have a spare around. While I think it's very unlikely, the Mini may have major issues and be unreliable, noone knows yet.

I'm sure there's other cases, but in my opinion, the only times I'd recommend a Mac Pro over the new Mac Mini would be if you need a more powerful GPU, or if for some reason you needed 64GB or more RAM (but didn't care it was slower), since that's very expensive on the Mac Mini until 32GB SODIMMS drop in price, and 128 is currently impossible, but not too expensive on a Mac Pro if you have the dual tray.

 

[eksu]

For the cMP? what is this card?

https://forums.macrumors.com/threads/testing-tb3-aic-with-mp-5-1.2143042/page-4#post-26748425

 

[b.lam]

I'd rather go for a current Synology or Qnap NAS if it is meant to provide mainly file serving and backup purposes. Even if the MacPro is in good shape, I wouldn't sell 9 year old Hardware to a client.

 

[orph]

on the cost side if there both close then it's what the client gets unless you want to cut costs or get a bigger margin it's not going to matter to much. (and time spent on refurbishing a CMP is part of the cost and relay can add up depending on how much extra work it needs)

i assume he is using macs for a feature he can get from using the mac over a windows/linux box

if it was not relent on some osx feature then something with a AMD ryzen 2400G cpu in a box with all the drives will cost close to half as much (maybe less) with windowz or linux, thats what id use for a home/work server setup ^^
ryzen 2600 is maybe better but then you need to get a cheep GPU so it's mixed there but an option.

it sounds like it may be past the level of a normal NAS box.

i was pushing apple care (or something of the like) as it's the first macmin update in ages and we wont know for a year or so if there's going to be any odd problems.

 

[deconstruct60]

https://forums.macrumors.com/threads/testing-tb3-aic-with-mp-5-1.2143042/page-4#post-26748425

That is an entertaining kludge, not a solution for a production critical server. It "happens to work" after warm booting from Windows ( and loading a static Thunderbolt configuration ) snapshot.

If the power goes out and the server has to power cycle how do you execute that kludge remotely or by automation? If need to attach a rotating backup drive ...oops have to reboot (twice!) to connect another drive.

Only just a bit higher than the kludge of selling a system that is just about to hit the "Obsolete" list in terms of hardware support. (operating on a 'fumes' extension ).

 

[weckart]

https://forums.macrumors.com/threads/testing-tb3-aic-with-mp-5-1.2143042/page-4#post-26748425

Thanks. Interesting thread. Didn't think any functionality would be possible without firmware support but there you are.

 

[pl1984]

With noone even using the new Mini yet, if this server is critical to the business, there may be a little wisdom in sticking with what you know, and can support easier, since I'm assuming you're going to support it, it sounds like you have some of the parts available if something goes down, will you still have those parts a year from now? That's a big advantage when people just need things to work and are going to depend on one machine as it sounds like they are. To get the Mac Mini back up, for the next year or so, I'm sure you could walk into an Apple store and grab a new one and be back up in the same amount of time or as repairing the Mac Pro, but warranty repair may take longer. At some point you may need to have a spare around. While I think it's very unlikely, the Mini may have major issues and be unreliable, noone knows yet.

I can't speak for the new Mini but I have had a couple of 2012 i5 models running 24 x 7 for over 2 1/2 years without a single problem.

 

[orph]

not shore if TB3 hack is stable for client use, may give problems

 

[tu2thepoo]

Just for reference, I'm benchmarking a Terra Master 5-bay USB-C (5 gbps) enclosure, and two drives in USB3 RAID0 easily match the same two drives in SATAII RAID0. A proper thunderbolt enclosure in RAID will probably outperform SATAII, and I can't think of many situations where the as-delivered performance of a cMP would be better than a 2018 Mac Mini plus a thunderbolt enclosure. The main attraction of a cMP is the hackability and being able to putz around with the parts - what customer wants to do that for a line of business system?

Like zipgs said, if a vendor offered me a 9 year old system full of custom hacks vs an almost plug-and-play solution, I would choose the latter every time. Every add-in card just barely brings the Mac Pro to parity to a modern solution, and that much more risk that a software or system update will bring down the house of cards.

Plus, any after-purchase support to your customer should be justified on speed of repair and replacement. If the server poops the bed friday evening you can buy a new Mac mini and probably have it running by the end of Saturday. Good luck achieving the same on a secondhand Mac Pro.

 

[fhturner]

Very good points and considerations, everyone. Again, I'm not necessarily looking for a definitive answer here, but I sure do appreciate the input and discussion. I'm still kinda out in the middle between the two. While the Mac mini seems to be a wiser option simply because it is new and will be supported longer, there are other factors to consider specific to this setup and my involvement in it...

If I were the client and you offered me the choice of a 9-year old Mac versus a 9-day old Mac solution I wouldn't have to think very long. If you told me the 9-year old Mac may never be able to run any future Mac OS's I wouldn't have to think at all.

Well, when you put it that way... Totally agree that a 9-yo system doesn't sound like the smart way to go on the outside. OTOH, these systems are so robust and were so well-designed and well-built that I think that negates much of that downside. And the 9-yo ones are essentially the same as the 5-yo ones (MP2012s last sold in late 2013)...so not really THAT long in the tooth.

But since you like the Mac Pro idea, I'll give you a few reasons to go with the Mac Pro. In your case, some advantages of the Mac Pro are ECC RAM, and hardware RAID, making this more server grade for what it's worth. I think the older hardware probably negates this advantage however.

Good point about the ECC RAM. I also do like the ability (and always have) of being able to easily add storage internally. If I went the MaxUpgrades/Areca RAID route, that would actually be housing four 2.5" SSDs in the lower optical bay, which, depending on what I decide to do for boot drive, still leaves three or four 3.5" HDD bays available for other utility disk use (as well as the upper optical bay, for that matter, I guess).

With noone even using the new Mini yet, if this server is critical to the business, there may be a little wisdom in sticking with what you know, and can support easier, since I'm assuming you're going to support it, it sounds like you have some of the parts available if something goes down, will you still have those parts a year from now?

Yes, precisely. That's probably the main point that many of you may not be considering here (since I haven't really emphasized it)— I'm not so much selling the client a standalone server and saying "Here ya go!" as I am outfitting his office w/ a service that I will be administering and maintaining. It will be fairly abstracted from him. As such, you're exactly right— there's a big "going with what I know" factor as well. Regarding parts, yes, I can be a source or find a source for spare parts when needed. But I'll do you one better: I have enough Mac Pros on hand myself that I can quickly deploy one as a temporary replacement in the event of a failure.

I'd rather go for a current Synology or Qnap NAS if it is meant to provide mainly file serving and backup purposes. Even if the MacPro is in good shape, I wouldn't sell 9 year old Hardware to a client.

if it was not relent on some osx feature then something with a AMD ryzen 2400G cpu in a box with all the drives will cost close to half as much (maybe less) with windowz or linux, thats what id use for a home/work server setup ^^

it sounds like it may be past the level of a normal NAS box.

Yeah, despite the main function being file serving and backups, I also get great utility out of other macOS Server functions like NetBoot, DNS, SUS/Caching Server, etc. So, beyond what I'd want to try to implement w/ a NAS.

Like zipgs said, if a vendor offered me a 9 year old system full of custom hacks vs an almost plug-and-play solution, I would choose the latter every time. Every add-in card just barely brings the Mac Pro to parity to a modern solution, and that much more risk that a software or system update will bring down the house of cards.

Yes, good points about the age of the Mac Pro and how it must be outfitted to "compete." However, since this will be a server installation, it may sit on the same OS version for long periods of time. I'm not too worried about an update bringing down the "house of cards" (further, I wouldn't consider it a "house of cards"...Mac Pros are fantastically capable machines— that's why they have those expansion slots in the first place!).

Plus, any after-purchase support to your customer should be justified on speed of repair and replacement. If the server poops the bed friday evening you can buy a new Mac mini and probably have it running by the end of Saturday. Good luck achieving the same on a secondhand Mac Pro.

Following on from my reply further up this post, I can have a temporary replacement Mac Pro onsite just as quickly, and since I could actually swap the storage from one tower to the other, could be up and running more quickly than a mini swap.

One further consideration that just occurred to me... The Mac mini is only going to be able to run macOS 10.14 Mojave, nothing older, which means the crippled Server.app w/o many of its prior functions. I'll need to review again what that means for the services I use. That in itself may be a deal breaker for the mini.

Thanks again for all the input and discussion!

 

[msh]

OTOH, these systems are so robust and were so well-designed and well-built that I think that negates much of that downside.

That hasn't been my experience. I am on my third logic board counting the original one - replacement about every four years. And I have never run it particularly hard.

 

[orph]

yes there will be problems, with old systems age/heat will slowly kill caps, PSU/VRM
corrosion from humidity (if your near the sea salt in the air relay helps that)
bad electric quality can relay add risk and loss of stability
9 years of bumps shaking
thermal pads or thermal past aging on the little heat sinks letting parts overheat and so on

depends on how much time it takes to renovate a used macpro and how cheep replacement parts are & how long the support contract is.
one plus with the macpro is you can just pull the drives and swap computer, can you do that on the macmin? if there's any internal data on the mac min it will need backups.

and as i pushed for apple care or something of the like for a macmin, some times with a new model there is a flaw in the product which we tend to only know about 6months or so later like my old macbook pro 17" with a good old GT8600 (think that was it) failed two times & that was just the GPU think i had the internals replaced almost two times lucky i had apple care.

 

[msh]

If I were to consider the MP for a file server I would use the OpenZFS ZFS implementation for mac os; you would have very robust data protection from the ECC Ram and all the check summing and other features of ZFS. You also wouldn't need hardware raid and it is arguably better than hardware raid anyway. In fact, I used to use my MP that way for awhile but the energy usage was too high and that is another vote against it.