The idea of the NAS Killers is that a NAS with enough performance for most people’s needs can be built for cheap, specifically cheaply relative to off the shelf appliance type solutions. That’s one of the reasons it’s 8th generation consumer hardware. Good price point. 12th and 13th is not.
It does seem a bit like you’re striving for a level of extreme performance we don’t usually find is particularly necessary. It’s doable, but it may be beyond the scope of a NAS Killer build.
Thanks for your feedback. I am new to these forums so I am investigating. the problem with the older builds its the high power consumption, something thas has become quite sensible lately due to the fact that power costs have increased drastically in the past year due to the war.
I have a 10 year old hardware, and when I checked the consumption of the processor, I would have spent a tremendous amount of money compared to investing in new hardware and having it running safely with a UPS for 10+ years.
Currently I have a DS1513+ running for 10 years, but no longer receives support, and I am pissed at Synology for their pickyness on the hard drives you can use, the nVME´s limitaitons if they are not their own brand and the really slow and outdated processors they include with their systems.
1 Like
Thanks for the write up. I grabbed an ASR-71605 off ebay which I guess is a dud (status LED’s stay a solid red and cannot detect the card). I guess this is a common issue when people run them without proper cooling? I then grabbed the ASR-7805 from the seller listed in this post which booted right up (and they did a great job packaging btw).
Now I just need to plan out my replacement NAS and move the E3-1270 to a backup.
There are generational gains in both idle power consumption and efficiency in terms of computation / watt.
For the use case of a basic NAS / plex / jellyfin media server the computational requirements are extremely low and you mostly only care about not wasting power while idling. That is mostly going to depend on your motherboard / cpu / psu. Basically your core system.
The case where you need anything even close to a modern level of compute power is when you are running a bunch of other docker containers or VM’s that do something else much more compute intensive. If that is what you are doing, you will likely know.
For instance I am currently running a beefed up Nas Killer 6 with an Xeon E-2276G a 6c/12t beast of a processor. It idles at 15 watts without cache or hard drives attached. It was a 6 year upgrade from my previous build which was basically an NK 5 with a E3-1285L 4c/8t CPU. It idled at 18 watts, so on average 0.5 watts per year idle power draw improvement. (Technically more since the new build has IPMI which does draw a few extra watts even when the server is powered off)
In my experience an older used enterprise motherboard will be both cheaper and more power efficient than a modern consumer motherboard. I’m also in the US where used enterprise hardware can often be found on the cheap.
It does depend on where you are located, where I am in the US I would likely have to run the server for a decade straight to just break even in terms of power cost vs component cost. There is of course always a case to be made for environmental impact.
If you are looking for the latest and greatest in terms of maximum compute power the NK guides are probably not what you are looking for. If you want a well balanced home server with an eye towards value and maybe some neat extras like ECC RAM or IPMI then you are in the right place.
4 Likes
Thank you for sharing your thoughts😉
Congrats on the new house!
The Intel nics are enterprise class and considered best in class in terms of reliability and OS support. That said the Realtek nics work just fine, for sure on any linux based OS like Unraid, True Nas Scale, or any Linux distro. I haven’t actually tried it but I suspect they are also well supported on Free BSD based OSes like Truenas Core.
Especially at 1Gbps you will not notice any difference in real world use. All else equal go for the Intel, but I wouldn’t really pay more or sacrifice on other features for it.
The 12th gen CPU’s are pretty new by the standards on these forums and a motherboard is probably going to be more a bit more pricey. The NK builds are at least partially predicated on the fact that older used enterprise gear tends to be pretty cheap, but new enterprise gear is crazy expensive so you will likely be looking for a consumer board especially if the ram you already is not ECC.
If you can find a board with enough SATA drives out of the box then great, but don’t pay a price premium that is greater than the cost of a used SATA controller unless you are looking for a tiny board without enough PCIe slots or something like that.
2 Likes
Thanks! Ended up finding a B660 board on ebay for $66 so not too bad all things considered - thankfully it has a full x16 PCIe and a second x16 slot that runs at x4 so can have the HAS card from this thread and still the bandwidth to add a 10G NIC down the line
Not bad at all! Sounds like its going to be a pretty sweet home server for the new place!
While a typical ATX power supply is 150mm x 86mm x 140mm, this PSU is 150mm x 97mm x 155mm (depth not being a problem in this case). I’ve attached photos of the measurements as well as showing how the hole pattern would line up. You can see that if I align the bottom holes (which they do line up when this PSU sits in the case), the top holes are way too high.
So again, it just don’t fit. It is too tall for the case.
Did you actually build this, @JDM_WAAAT?
If you did, you may have mixed up your PSUs, as it’s not physically possible to get the DPS-475CB in the N400 case. I’d say the right thing to do would be to adjust your guide before someone else like me follows it and ends up with parts that don’t fit together
1 Like
I’m not very familiar with the Delta DPS-475CB line, but it looks like you may have a proprietary HP version the DPS-475CB-1 A which seems to have been made specifically for HP to run in some specific non-standard HP workstation (Looks like the Z400?). In the image above it has an HP product number.
Delta makes a ton of OEM PSU’s and will customize them for large OEMs. It likely has identical internals but a proprietary form factor (maybe also MB connector who knows).
Can you post a list of links to the parts you bought? This doesn’t look like the part you mentioned.
Could be a mistake on the seller side as well.
It does look slightly taller, and the exact model does show as a possible HP OEM part, which might be non standard. It does, however, look very close to a normal ATX power supply.
Power Savings Difference
Existing Hardware
Case: CSE-846
MBD: SuperMicro X9DRi-F
CPU: (QTY-1) Xeon E5-2637 v2 3.50GHz 4Core/8Thread
RAM: 64GB
Min Power no spinners 130watts ish
New Hardware
Case: CSE-846
MBD: ASROCK Z370 OEM
CPU: Intel i5-8500T 2.10GHz 6Core/6Thread
RAM: 64GB
Min Power no spinners 110watts ish
Cons
I gave up IPMI and ECC RAM
Pros
20-30watts power savings
iGPU hardware transcoding
quicker Unraid Interface
I run UnRaid w/ 14 dockers and 1 VM.
3 Likes
What else is in the system or perhaps common between them? That usage feels very high for that cpu, mb, and ram
I reused everything except the CPU, Motherboard and RAM. The SuperMicro Case has stock fans and power supply (I only have one installed). It has 4PCI cards, quad NIC, HBA, NVME, 10gb Fiber. There are 4 SSD accessing all the time. Power compare was to just show a 48hr (24 old hardware, 24 new hardware) same load.
1 Like
@Ian may be on to something here. Here is a link to the part I bought, which does mention it is for the HP Z400. Although mine does not look like the image you provided, @Ian. As for a standard DPS-475CB vs. an HP Z400 specific version, nearly all of the variants on eBay call out the HP Z400 in their description. So if there is a “large” varient that is too big to fit in the N400 case, it seems to be the more popular varient when you click on the link in this article to navigate to the eBay search page
@Destate You are right about the image, on closer inspection it looks like I pulled an image of an unrelated PSU off Google Images. I will take that down so as not to miss-inform future readers of this thread.
In general OEM parts are something of a hazard in the second hand market for enterprise server parts. HP, Dell, IBM, etc will all order special runs of otherwise standard parts like Motherboards, PSUs, Nics, HBAs, etc that have non-standard physical dimensions or special firmware just so that when their customers need replacement parts they are forced to buy them from the OEM at vastly inflated prices for what are really just super cheap parts.
However because the parts are only usable in whatever specific workstation or server they were designed for they are extra cheap when they hit the second hand market. Often the lowest priced options on Ebay are always the OEM versions. You end up having to scroll down to the Intel branded Nic, or Super Micro branded motherboard or the Delta branded PSU and sellers don’t always put the correct info in the descriptions.
A while back I needed an HBA and ended up buying an IBM version of a standard LSI controller and even though it was physically identical it had a specialized firmware that made it useless outside of an IBM server.
The power draw you are seeing does seem a bit high for your core build, but you also do have quite a lot going on both in terms of docker containers and add-on hardware.
Possibly one or more of your docker containers is actively doing something that requires a fair amount of compute.
It is also likely that one or more of your add-on PCIe cards, besides just using power itself, is preventing your CPU from entering into higher C-states, or low power idle states. Basically they may be interacting with your CPU in a way that prevents it from “sleeping” when it is otherwise idle.
There is often not much to be done about that, you could remove the PCIe cards one at a time to identify the culprit, but even if you figure out which card(s) are the issue it doesn’t really matter because that card is probably doing something important or you wouldn’t have installed it in the first place.
While it is nice to have a power sipping server, there isn’t much point if you have removed the parts that made it useful in the first place. You can try to find a replacement, but you will just be rolling the dice that the new part is any different and the cost of replacing the part may exceed any savings from the reduced power draw. That mostly comes down to the cost of power where you are located.
The rest of the system probably accounts for the rest of the usage.
Well today I learned something! Thanks for the explanation, @Ian. I’ve purchased a different PSU (one that has an actual datasheet) and hopefully I’ll be able to get this thing up and running!
For the sake of future visitors to this article, I’d really think that changing the link from an eBay search that returns incorrect results and claiming it was part of the sample build (which was never actually built) would be the thing to do.
Also, in terms of error correction (upcoming pun slightly intended), I also went with the recommended motherboard (the X11SCQ), which this article suggests “ECC support contingent on the CPU,” but a quick Google shows that the X11SCQ is listed as having “non-ECC memory.” Unless I am misunderstanding something?
I know I shouldn’t be putting this much trust in folks on the internet, but it feels like the least I can do is try to save the next poor NAS builder the trouble of following this erroneous guide. Thoughts, @JDM_WAAAT?