Busting 8 Common Homelab Power Efficiency Myths
HTML-код
- Опубликовано: 28 май 2024
- To try everything Brilliant has to offer-free-for a full 30 days, visit brilliant.org/Wolfgang/ .
The first 200 of you will get 20% off Brilliant’s annual premium subscription.
Power efficient builds spreadsheet from HardwareLuxx: docs.google.com/spreadsheets/...
Follow me:
Mastodon tilde.zone/@notthebee
GitHub github.com/notthebee
Twitch twitch.com/notthebeee
Support the channel:
Patreon / wolfgangschannel
RUclips Members / @wolfgangschannel
Ko-fi (one time donation) ko-fi.com/wolfgangschannel
Videos are edited with Davinci Resolve Studio. I use Affinity Photo for thumbnails and Ableton Live for audio editing.
Video gear:
Camera geni.us/IdnhpP (Amazon)
Main lens geni.us/MF3Ej (Amazon)
Secondary camera geni.us/aQ3l (Amazon)
Secondary lens geni.us/jh0Oy7Y (Amazon)
Microphone geni.us/tgiSqL (Amazon)
Key light geni.us/Gi1zE2 (Amazon)
Softbox geni.us/F86pM (Amazon)
Secondary light geni.us/aciv (Amazon)
Other stuff that I use:
Monitor geni.us/KUzcmcP (Amazon)
Monitor arm geni.us/Ei4x (Amazon)
Laptop stand geni.us/hG7Y39 (Amazon)
Keyboard geni.us/KSR6En (Amazon)
Mouse geni.us/JpQG (Amazon)
Audio interface geni.us/sdhWsC (Amazon)
As an Amazon Associate, I earn from qualifying purchases
This video is sponsored by Brilliant.org
Timestamps:
00:00 Introduction
00:30 Brilliant.org
01:34 Myth #1 - Idle power draw is meaningless
03:59 Myth #2 - The lower the TDP, the lower the idle power draw
05:45 Myth #3 - Undervolting can decrease idle power draw
06:47 Myth #4 - ARM is more efficient at idle
09:37 Myth #5 - Newer = more power efficient
11:39 Myth #6 - Laptops draw less power than desktops
15:24 Myths #7 and #8 - PSU-related myths
18:42 Outro 
Наука
To try everything Brilliant has to offer-free-for a full 30 days, visit brilliant.org/Wolfgang/ .
The first 200 of you will get 20% off Brilliant’s annual premium subscription.
Power efficient builds spreadsheet from HardwareLuxx: docs.google.com/spreadsheets/d/1LHvT2fRp7I6Hf18LcSzsNnjp10VI-odvwZpQZKv_NCI/edit
Useful vid, as usual! Hope you don't mind me putting this useful info here : In case anybody is wondering, 32GB hyperx fury 3200mhz runs at full speed and capacity on the n100m and n100itx.
Do you think CPUs with E-Cores are worth it for good efficiency and performance for not so heavy tasks?
I'd really love if Wolfgang willing to make an example/tutorial using old mainstream/popular hardware of their era and still abundant and optimized it, not just the niche one. It would be nice for me, a common people without IT education, a novice. Using anything available on their home (unknowingly mistake etc), and then, mix-match it, not just buy an used old hardware.
For example: on my area, H61 chipset Mobo for miserable Sandy-Ivy Bridge (2nd-3rd) is still abundant. Along with DDR3. Maybe the pebble-like abundant is similar to Haswell - Kaby in Europe. And, C chipset is good, but B85M is abundant and not everyone using 6 HDDs, 4 at max for their homelab.😊
😂😂😂 why did you do it you didn't have too 😂😂😂 🤣🤣 @ 9:30
12:58 - laptop without screen and battery or how was it compared? 🤔
"Myth#4 ECC..."
"NO. we're no doing that today"
😂😂😂
😂
What is an "ECC" here? ECC Memory?
@@LichtTempleryes ECC memory. There is a big debate on if you need ECC memory for home servers. It's basically if the data is important enough to have backups and or data parity dedicated to protecting it, it's important enough to spend the extra money on ECC. Vs. But ECC memory is expensive and memory has gotten much better and memory errors are much rarer than they used to be. I'm probably glossing over massive points here but I already had ECC so I stopped looking into it at least for now.
@@Zanthum , I see, thanks! I thinks same, I have ECC memory only on servers, where I experiment with AI learning stuff and it takes some time. But for data storing I use "normal" RAM.
@@LichtTemplerYou did it completely backwards. Data storage should consider ECC, because bit flip can potentially corrupt your filesystem. (Despite how small a chance it is).
Machine learning algorithms can usually tolerate some error. Sometimes, errors are part of the algorithm. Worst case the application crashes, it can restart from the last checkpoint.
However, one more thing, server RAM, RDIMM is way cheaper than desktop RAM on the used market. They only work for server platforms tho, which aren't power efficient. So much lower frontal cost, slightly higher running cost. It all depends on your electricity price.
As a low power enthusiast who runs my homeserver 100% off a solar panel, I have to admit myth #5 tripped me up. I knew all the rest though!
Spill the specs!! I'd love to run something like that off solar
I also run my nas of solar power 100%. Same for my router. I have 3 100-watt solar pannels. Hooked up to a 20ah mppt solar charger. I also use 2 20ah lifepo4 batteries.@@BomarJr
I use a rasberry pi4 with 8 4tb HDDs. It sips power. Also, i have a really nice router. Its all hooked up to a UPS, just incase my solar batteries die. Plus, i also use this setup to charge all my drvices. Like batteries, phones, steam deck... etc... 😂. SOLAR POWER !!! @@BomarJr
@@ricardofranco4114solar system total watts?
@@BomarJrI will assume something like a 50W server running from a li ion battery of 2 kWh or so. A single 400W solar panel could keep that charged in most conditions without any problems.
The information efficiency of this video is incredibly high. Seriously this is one of the best myth v reality videos I've seen. I've been working with servers and home servers for quite a while and I picked up some really useful info here. Wolfgang is putting out some great content on this channel
My idle video consumption on Wolfgang’s channel only ever increases each gen
small form factor hp (and other) office machines are great entry points. they are (usually) engineered well for power efficiency and can be picked up cheap.
I agree. I have an optiplex. Been powered on for 5 years now, spends 98% of its time in idle c6.
do you mean like the ultra small form factor ones that have no pcie slots? or the ones that have half-height slots? Or the full-blown desktops?
I've been thinking about getting one if cheap enough, but I'm not really sure about how I'd set up my drives (probably have to find a real PC case), what PSU to use to power up HDDs, or what to do about sata expandibility.
If you have any insight on what you're running and the power draw I'd be curious.
@arsenalfanatic0971 no I think he means Dell optiplexs and lenovo think centres. Sff office PCs...usually matx boards?
HPs are a step above. Z420-z640 run SILENTLY under decent load. They put a lot of care into these details and it shows. Lenovos are good too but you can always count on an HP to run off broken dreams alone lol they take a ruthless beating and still run.
@@arsenalfanatic0971 like the other commentor said, i mean the compact desktops not the ultra tiny thin clients. altho they can be used too and they are efficient they're way more limited in terms of power and expansion.
I picked up a job lot of HP 290 G2 sff boxes and have been blown away. less than 3 watts from the wall at idle (with no sata devices plugged in). spends most of its time in c8 power state
Always a great day when Wolfgang uploads
Especially if it's about power efficiency
i'd like to mention one thing you could mention. When it comes to CPU power efficiency generation does matter for idle loads combined with the motherboard, software and bios features. Turning off audio can save 1W (measured at wall) in bios. When the PC is in bios it consumes a lot more power than when it has booted into an OS.
I tried turbo stat on 3 PCs. 1 a low end amd ryzen dell, 1 a 1U erying PC, and 1 an old haswell in a high end motherboard. Tweaking the erying is a nightmare but the erying has aggressive clocks, so the idle is between 1.6-2+W. the amd ryzen had a consistent 1.6W idle while the haswell showed around 20W. I have the haswell tuned with power efficiency options available and i have a wattmeter running at the wall for my cluster, showing 100+ watts including the switch. I've measured each computer individually.
The biggest issue no one talks about in power draw which your video does not address is the GPU. GPU idle is a huge pain and is why i dont run GPUs in servers. If you have a program idling in the GPU it will for sure not idle at its rated idle, but even then thats still a significant increase because consumer dedicated GPUs do not shut themselves off when unused like with nvidia optimus. Sadly even this option isnt available for linux. I am trying to use openCL and other features on GPUs so the IGP is good but once i start using the IGP in my own software, the idle power consumption will go up for sure.
Many do not configure their bios for power efficiency but not all bios has the options and not all motherboards draw the same power as it depends on their VRM designs and how much hardware is packed on the motherboard. things like optical, multiple NICs, more sata ports even PCIe can actually increase power draw depending on what is added to introduce it. Fans for me can consume 20-30W alone.
Storage drives are also a significant power demand. Even older spinning disks can be set with different sleep profiles from constantly spinning for low latency to fully parking the heads and disk when idle for more than a few seconds. It is a difference of around 8w per disk from parked to a continuous write, so a 4 disk array will take ~10w parked swinging up to 40w when writing.
Storage consumes more than 50% of the total energy used my NAS, which idles at 4w and peaks at 15w with one 5400rpm 3.5" disk. A second disk and it's up to 6w and 20-22w.
It has fairly low computing power and was manufactured specifically for the file serving task. (Single core ARMv7 SOC at 1.2GHz and 128MB ram, just enough to feed files on 1Gb ethernet and do some minor services like DHCP, local DNS and background torrents.)
istg every time this man does a sponsor segue he always catches me off guard, "computer science and math doesn't have to be boring"
and I'm staring at my screen like, "Yeah it isn't, but why does that-"
"Well here's the sponsor of todays video, brilliant!"
"Damn it! He got me again!"
"ECC Is a must have" - Man. I can tell you from experience... Your server is 1000x more likely to screw something up because of a small mistake made by a developer behind some library, used by another library, used by a yet another library used by some small background service running on your OS than a cosmic ray flipping a bit in your memory.
How do i know that? - I am that developer.
Thanks for the insight man!
ECC is not just about stability (which is what I believe you are focusing on). It's also about data integrity. Folks at truenas forums, for example, strongly recommend ECC.
I don't have a view one way or the other about ECC. But food for thought: Why don't we have error correcting L1 L2 L3 cache at the CPU level? I have sort of always wondered.
Kindest regards, friends and neighbours.
And what's 1000x more likely than even that is me doing something stupid and working the system myself 😅
@@-blackcat-4749 Agreed! *Anything* with high memory transfer rates should be ECC, as should a high performance NAS/SAN, but for the average home user probably not necessary (with the exception of a NAS or server). Hell even my workstation at home has ECC. I've heard the non-ECC argument for the better part of 30 years, it rarely holds up to realities. The only real issue is cost.
Well done! Thanks for mentioning #5. Many people won't believe you can build low power systems even from 20+ year old parts. Although #2, #3 and #6 do play a more important role with less sophisticated power saving features in older hardware.
" 550 Watts is not enough for a modern gaming system :) "
So I'm running some old but premium bequiet 550 Watt PSU with a 5800X3D and a 4070. I did undervolt the GPU though but only because I am a sucker for efficiency, works fine on stock settings too. Great video, saving all the watts!!!
seasonic calculator recommended their 450w models for an i3 12100f wit a 4060 8gb !!
Great non typical content Wolfgang! Thanks for sharing your knowledge and expertise
A couple of extra thoughts regarding PSUs:-
1). Personally I found that the more powerful the PSU the higher its idle draw power.
2). A server with lots of hard drives will need a larger number of SATA power connectors, which can exclude some lower power PSUs. I found that SATA/Molex power splitters are dodgy at best and something to be avoided if possible.
The number 1 is just like he said, some higher wattage PSU are more efficient than the lower one, so yeah always do some research
what do you reccomend instead of rm550x 2021?
One thought concering power usage: If I have a beefier CPU, I can do task faster then with a low tier CPU, even if they draw the same amount of power in idle. So it makes sense to go with a mid-range CPU depending on the usual tasks you do.
Great video! I learned a lot for my upcoming home server build. Thanks for doing this one!
i hope this will come in handy when i get to build my home server because one thing i want to achieve is a good power effeciency.
your videos are really good and the content seems to be faithfully researched.
Definitely liked watching this video! It makes me feel glad that I'm not deluding myself into thinking I don't HAVE to use a RasPi or go hunting around for The Most Efficient Hardware to run a home server. I can just use what I've got or build what I need and not worry so much about the power efficiency! 😄
Pretty awesome video! I'm in the process of building a low power consumption system and I've just discovered your channel! Awesome work!
Still using an old 4790k for my server. Having quick sync build in, is really handy. For an update, I would want one with AV1 and H.265 support. Having 2x SSD and 5x 5.25" harddisks, running a handful of dockers and two VM's (one of them being a windows), power draw is 65w-120w, depending on what it is doing.
haswell's idle power efficiency is great for the age of the chip.
Concerning the efficiency of ATX power supplies, would it be possible to do a comparison with some PicoPSU + external power supply substitute?
In my home server I took an SFX PSU from the scrap heap, gutted it, cut a bit of FR4 to watch the PCB and then mounted an 4"×2" 12V 170W PSU which feeds a PicoPSU. By buying a high end 4"×2" PSU the idle power efficiency is over 90%. I also don't need the huge cable bundle of an ATX PSU which improves my air flow which improves my power efficiency too. Then again I did a full custom wiring loom for the system so there is no excess cable like you would find in a server from a tier one vendor. The PSU can in theory run on convection cooling, however it gets ducted exhaust flow from the hard drives which keeps it under 45 Celcius worst case which is well below its rating which is more like 60 Celcius from memory.
@@jonathanbuzzard1376 what company did you buy the high-end 4"x2" PSU from?
@@arsenalfanatic0971 it's an XP power PSU from memory. I got it from RS Components, but Farnell, Digikey and Mouser all stock the same or similar. It was the middle of the pandemic so I even ended up getting the semi enclosed version which I then removed just down to actually being able to get one. I can look the exact model up if you want. It was about £90 so not cheap but I expect it last for a long time and in my experience picoPSUs last at least a decade.
What often gets overlooked about ARM CPUs is that they're usually more efficient because they're more customizable. What contributes way more to power usage than the CPU itself are the features and I/O (nowadays I/O is the most significant factor in power consumption. Moving data is typically way more expensive than calculating it).
ARM allows their customers to custom build CPUs. AMD and intel on the other hand have one architecture (they do have multiple) that must serve multiple applications. ARM systems are often more purpose built which make them way more efficient. Good examples are the A64FX (designed for the Fugaku supercomputer) and the Ampere Altra line. They're both based on basically the same CPU cores (the A64FX's became N1) but have different feature sets. The A64FX is designed for maximum 64bit floating point compute (and it still outperforms most modern CPUs at that) and maximum memory bandwidth (it's basically a more general purpose GPU) and doesn't need much memory or many I/O Lanes. The Altra on the other hand doesn't have HBM Memory support, lacks the SVE (vector compute for FP compute) extensions but has way more I/O and great integer performance.
Unfortunately, most reviewers don't see that as an advantage and tend to point out that certain ARM systems are not so great at every single task (even though they perform well at what they're designed to do).
And i guess in M series chips, it is very power efficient due to a lot of dedicated hardwarw acceleration? Also they can cram anything to a single package since they are usually an SoC
ARM also comes from a background of low power devices and Intel normally makes high power devices
@@thewhitefalcon8539 that doesn’t matter as much. Most modern ARM architectures are not low power and the "efficiency" cores (little cores) are actually less efficient than the big cores
@@siontheodorus1501 that’s another factor. ARM is built around the fact that accelerators can be added if necessary.
But I don’t think that that is what makes Apple Silicon so efficient (it’s actually not that efficient just low power). It’s more about their integrated design, all the things they are criticized for like soldering down the RAM and Flash allows the higher efficiency and lower power consumption. Soldering down RAM allows to drop the voltage and get it closer the cpu (physically closer, in apple silicon it’s right next to the Chip) which reduces the power consumption by a lot (each RAM DIMM consumes like 3-4W even at idle). Then, their SSD controller is integrated into the CPU which allows to share memory and reduces power consumption for transferring data between controller and CPU.
In general the more flexible you want your I/O to be, the higher your power consumption will be. That’s why those tightly integrated systems, laptops and SBCs achieve much lower consumption. Apple just adds an efficient CPU and GPU to it
@@lukas_ls i see, how do you know that it is actually as efficient as x86 systems?
I have two N100 machines, a Chuwi Larkbox X 2023, and an N100DC-ITX.
Both idle at 4.2 and 6.2w respectively. Im rather quite fond of them.
These N100 systems are awesome! I'm just consolidating multiple pi deployments into 1 N100 based mini-pc
@@OliverHamilton yep. Pretty much. Go ahead.
20W ish under load.
What PSU's would you recommend for that? Just ordered one of those boards from Ali for a 304 build.
@@TomasIgnatavicius The Chuwi is whatever 12V2A+ is lying around. The N100DC-ITX was whatever old laptop chargers I had.
@@LeminskiTankscorThe efficiency of the power supply can matter a lot at those numbers. I have an N100 box too, and I tried all the 12V PSUs in the house and picked the one that drew the least from the wall. Also, some old 10 year old power supplies will draw a fair amount even if nothing is connected to it. Thisch is a factor for PCs that are off a lot.
great video as always. My file backkup server is based on Intel G2030 2c/2t 55w tdp cpu from 2013 ($35 total from ebay, except for hdd), and is pretty much all I need. It powers on automatically only when needed and turns itself off when it finishes making backup job, so I don't worry too much about power consumption.
As a person just starting to consider building some home network/server stuff, this information is very helpful.
another thing against arm systems in homelab is software compatibility. I am running an dell R730 with xcp-ng for gameservers and you cant easily get a ready to use virtualisation platform for arm. also many gameservers have no arm build available
Great video. I’ve been down this road in multiple ways with multiple projects. The Asrock J5040 ITX is a great board for low idle and still having decent enough performance for some applications. But 12 or 19 volt DC powered boards can be an even better option when paired with a high quality power factor corrected “brick” PSU. The Micro PSU adapters are not great as you’re triple converting the power from the wall to the CPU.
It’s also sad that idle currents for desktop CPU’s have generally gone up with each new generation. 12th gen Intel idles at about double its 6th gen equivalent. And the peak “turbo” power is also way up and well beyond the normal TDP unless your bios lets you restrict it. That means an inefficient at idle oversized power supply. The benchmark wars have been mostly waged by Intel and AMD cranking up turbo speeds.
The older i3 Intel NUCs are impressively low idle power if they have enough I/O for an application but they often don’t. It’s unfortunate not even Asrock makes fanless boards with lots of expansion interfaces that use similar power to a NUC or laptop.
I also want to point out the idle power of Apple’s M chips crushes anything in the x86 world of similar performance. You imply they do not have an advantage at idle but they do. Now the crappy Raspberry Pi indeed has horrible idle power consumption in excess of a Macbook M1 Air. I think it’s funny how many use Pi’s for 7x24 applications when they’re not well suited besides being cheap. They don’t even have a SATA interface and trash their SD cards. And they use so much power they need fans or aftermarket heatsinks to survive inside an enclosure.
Excellent explanation of the subject. Thanks.
Greatly explained. Thanks!
Awesome video! This will be great to post around to people as I commonly see a lot of those myths said out there... Also have to say it is quite amusing to discover this channel just as I'm about to deploy a Threadripper as my new home server... :D
I’m loving these low power videos.
Very informative, Thank You!
Nice video. With the older vs newer model CPU comparison, I would have liked if you compered modern CPUs to even older ones, like Sandy Bridge, Ivy Bridge, Broadwell, Haswell. Those are still very usable CPUs for home server use, and there is still negligible difference in idle power draw.
You're right but actually idle power consumption or rather low load power consumption improved a lot over time. Ivy bridge and older will draw significantly more power than more modern cpus
I run an Ivy Bridge Xeon Server and stepped down from a 12 core high end model to a 10 core sort of low TDP version. In this generation that actually made a big difference even in idle. While these chips are still very usable I would only recommend them if electricity cost is an absolute afterthought. Here in Germany, it is, sadly, not.
@@catriona_drummond Couple of thoughts. I’m not disagreeing (or agreeing) with you, just want to add to the discussion. First, _higher core_ count will always lead to higher idle power usage, regardless of architecture. So, if you compare older and newer CPUs with equal core counts, I still don’t believe there is a big difference in idle power draw. (I might be wrong.) Now, it’s true that older CPUs with 4-6 cores won’t be as performant, but that’s part of the deal when choosing an older CPU. If you need performance, an older CPU is definitely not a good choice anyway, regardless.
Second, lower TDP CPUs may have lower idle power usage, but the reason is not the low average TDP rating. If all other factors are equal, then what matters is the _base frequency._ If you choose a CPU with a low base frequency, but high turbo frequency, it’ll have high TDP rating, but the idle power usage will still be low. TDP is not a good indicator of idle power usage.
A third thing I want to mention is that many people see older servers using a lot of power, and they attribute it to the CPU, while this is a false assumption. Old servers are cheap. So, people buy higher specced rackmount machines with a lot of components. Like SAS controllers, hot-swap backplanes with 6-12 bays, 8-1000W dual PSUs, a huge number of system fans, hundreds of Gigs of memory etc. And then they say, “oh it’s inefficient, because it’s old”. No. It’s inefficient, because there is a lot of stuff in it that they don’t need. The same architecture, in a tower case, with 1-2 fans, a SATA controller, and 16-32GB of RAM would use maybe 10-15W more than a similar, newer config. Or not even that. Whether that’s a lot, I don’t know. The ROI on buying newer can be 3-5 years or even more. That's a long time.
@@tamask Cores can be shutdown in firmware.
Anyway... is 15w a lot? My server has a peak of 15w while writing to a spinning hard drive (which is 8w alone). It is 2012 vintage (32b ARM, but that is the sandy/ivy era.)
Testing is important, manufacturers will pull all sorts of tricks like including sleep states as idle power specifications. However sleep states are generally not suitable for a server that needed to be on standby listening for requests. (There are use cases for server sleep in a large commercial setup with WOL and load shifting to fewer machines during off peak hours. Not relevent to a solitary home server.)
Which reminds me Haswell [IIRC] had such low current in one of its sleep states that it would cause voltage instability in some early motherboards and they would self reset.
A video on energy efficiency with very interesting information, thank you very much for taking your time to provide all this data.
I leave you some ideas for future videos: how to carry out measurements on equipment that we already have in use? What BIOS or kernel parameters can be useful to improve energy efficiency?
Guten Rutsch :) Danke für so viele Videos dieses Jahr!
Looking forward to that Asrock N100 DC ITX review. I've got mine down to 7w which I am happy with even though I could never get the system to actually reach C states lower than C3 even though my NVME drive has ASPM enabled. With a SATA drive I think it can go down to 6w. Oh well...
Thx for focusing on efficiency, as an amateur user this is essential.
I have no clue how to get that low idle power consumption. I had a 4770K and a 4790T and the whole system consumed 60-70W in idle with 6 HDD + 2 SSD, one pcie network card with 2 ports and 4 fans in the case. I changed to a 5800x and now server is consuming about the same, even a little more. Running ESXI (older version 6.0) with 5 VM (FreeNAS is one of them) that are idle most of the time. I tried every power saving setting in the BIOS but no luck.
Ryzen is not very efficient at idle. Are your hdds spinning down when not in use?
Put your drives into standby mode. My Haswell (Xeon E3 1231v3, which is pretty much the same to your i7's) is idling @14W from the wall with 5 HDD + 2 SSD and both NICs (2.5G, 1G) connected to different VLANs.
@@AuftragschilIer sadly my drives won't go into standby mode for some reason. Even If I set up standby to 5-10 min with/without spindown they don't. I'm not the only one with issues with FreeNAS. Saw other people not being able to put their HDD in standby.
@@AuftragschilIer well I finally managed to figure out why freenas was not spinning down my drives. Because of the jails that for some weird reason are on the pool. I really thought they were on the system drive but that doesn't even seem to be an option to move to. Once I stopped all jails the drives spun down (25W power reduction). Power usage is still around 50-60W but it's a step in the right direction.
Have to put an extra small ssd for the jails now and figure out how to move them away from the pool.
Amazing research man great work
Thanks for all the videos in 2023
Great content! One question: is there a good source for power related commands and tweaks? Or as a suggestion: could you make a video about the subject? :)
Awesome video!
I can't wait for Asrock N100 test!
Hi Wolfgang, when can we expect video about intel n100 nas setup? I'm super hyped for this video as I propably going to build something with asrock n100dc or some of the aliexpress boards.
Great video as usual :) Maybe you mentioned before, but which devices do you use to measure and log the power consumption ? I only know the ones with a display, didn't realise they could be connected to the home network as well. Thanks !
edit : found it actually by going back to your b-roll : looks like Shelly Plugs.
Thanks for the video again, I was going to buy an N100 mini PC to upgrade my Pi4 based NAS with video decoding, but will probably build a proper rig with an i3 instead for expandability.
There are a number of 'smart' plugs out there that can be connected to home assistant to monitor power draw from either the wall or a power brick but there's a lot of discussion as to whether or not they could handle the load from power hungry servers over time. A decent UPS will also let you monitor power draw via NUT which can also be monitored via home assistant. There are probably other options but those are the 2 I'm most familiar with.
A lot of people use the old reliable Kill-A-Watt, which just gives you a LED display readout of the power consumed at the socket.
Ultra dense and useful. Thank you.
I run my daily containers (Home Assistant, Frigate, RTL433, etc.) off a Lenovo laptop with a Sandy Bridge i7. It draws about 8 watts at idle and works well for what I need it for. It was $13 at a thrift store (plus $10 for a caddy to add a second SSD) and it would be hard to justify replacing it unless I was running much more substantial loads.
16:30 "30W of 750W is less than 3%" - is this one of the errors that are introduced on purpose to increase engagement in the comments?
Just commenting to make that N100 video happening since I'm thinking about building a server with it.
N100 = nvme epeen opportunity. Nice fanless desktop for web and basic office etc, some esports at low settings. Previous gen the 4125, is just as nippy and uses much less power
As somebody beginning to investigate homelabbing, what is missing in terms of info to me is actual build recommendations with modern and widely available components. People are usually referencing old stuff they picked up on ebay that is not being produced anymore etc. Some people don't have time to scour ebay-like sites hoping to pick something up from 5y ago or some obscure power supply to save 100$/€, and just want to know ok these components you can order anywhere and are good for low power draw and these typical use cases, etc. combined with these hdd's for this reason.
Thanks for the content!
It's part of the learning process N-Man. Very few people have all of the answers you require because it doesn't make good content. AND your requirements will be unique: your budget/ skill set/ willingness to overcome setbacks/ tasks you need your server to do/ risk aversion/ etc. Keep looking through your favourite content creator's back catalogue for hints. Try Wolfgang's links
The thing is, it's more like saving $1,000-10,000. New server and workstation stuff, even with tech that's a few years old, is expensive, *and* has gone up quite a bit over the last few years. The new ASRock Rack and Supermicro equivalents to some of these 3-5 year old servers are commonly $2500+, and can often be $700-1500 for the motherboards alone. I don't even mean the latest and greatest, either, but like Skylake and Kaby Lake based Xeons, and Goldmont Atoms, that Intel has kept pumping out. On top of that, there's always new stuff coming out every year, and older parts finally crashing in price.
If you don't care about IPMI or ECC, 4th-7th gen Intel Pentium, Core, and Xeon E3, on basic motherboards, are good bets, with 8th to 10th gen exemptions for the Pentiums and i3s. 8th gen and up Intel can be OK, with a mini PC type business computer, or good luck. But, since they just kept making Skylake bigger and badder, to compete with Ryzen, results can vary a lot, with common desktop parts. FI, a lot of those later Skylake++ motherboards will have VRMs made to push many amps, that are awfully inefficient at idle, or with low-power CPUs. But, home desktops will tend to be off or in standby, so it was typically only a minor issue for the intended use case[2].
For newer stuff, the monolithic Ryzen 4000G(E) and 5000G(E) chips (note that only the Pro xx50G(E) ones support ECC), on newer B450[1], A520, or B550, are pretty good (also 6000 and 7000, in mini PCs and such). ASRock has good support for ECC, and x16 bifurcation, across their 500-series lineup, too (you can get x8/x8 risers, x4/x4/x4/x4 risers, or 4xm.2 cards, and use them on $75 boards). AMDs before Zen 2 (1xxx, 2xxx, and 3xxxG(E) CPUs), Ryzens with IO dies/chiplets (almost all 3000 and 5000 non-G models, with a few exceptions, like the 5500), and the X570 chipset (which *is* a 3000/5000 IO die), not so much, as the IO die will typically use 10+W, just sitting there. AMD's main weaknesses are firmware and software compatibility with popular virtualization feature sets, quality of FreeBSD hardware support, and not always knowing if you can run a given motherboard headless on a pre-7000 series non-G/H/U CPU.
The more features, the harder it will be to save power, and many server-centric parts also make it harder (FI, some Intel NIC drivers, and the LSI HBA drvers, can disable ASPM, by default).
Also, try to avoid most 2.5GbE NICs. The i226-v is supposed to be OK, but its predecessor, the i225-v, was hot garbage, and Realtek still has some chips with buggy Linux drivers. IMO, leave 2.5Gb stuff to Windows, if possible, and just take the plunge into 10GbE w/ SFP+, if you want more speed.
If 16-64GB (but, some can take 96GB, now) max RAM, and little to no expansion, is OK, just get the tiny/micro form factor business PCs, with up to 10th gen Core CPUs. If you want real server features to play with, though, or a lot of add-on hardware, there is no easy solution on a budget.
[1] But *generally* steer clear, as most were craptastic respins of craptastic B350/A320 designs, as the board makers didn't want to put real effort into early Ryzen boards. With a few exceptions, like the ASRock B550(M) Pro4s and Steel Legends (same basic boards, but the latter has more RGB), only the B450 boards that came out well after the Zen 2 (3500X and up) release are really worth it, just in general. And, it can take a little effort to tease that info out, whereas you can spend $10-20 more and get a 500 series board, instead.
[2] One reason business PCs are better about this is that they don't usually get turned off, or go to sleep. Between Windows Update, and multiple levels of security software, trying to use a computer within the first half-hour of booting up will be painfully slow - yes, with nice SSDs in them all. Figure that you'll also have 20+ in any decent sized office, but maybe 100+, and that idle really matters.
@@laurelsporter4569 not talking about 10k setups for starting a homelab. But I don't think it would be a stretch to say this is what a 500$/€ budget gets you, this for 1k and this for 1,5k etc. These can be updated yearly based on what is available and still create a lot of content opportunity aside from that (e.g. buying off the shelf prebuilt system vs same budget but with 2nd hand components or same budget when you build it yourself with new components, etc.). Basically like many channels do with gaming pc's.
Asrock N100DC ITX is a good, cheap and energy-efficient board (6-7w). Add 16gb of RAM, a cheap laptop power supply and the drives and case are determined by your use case and how much storage you need. Install your favourite Linux distro, run a few containers of your favourite apps and you're gtg.
I've been running my entire home network from a device with four i225-v (rev. 03) NICs for more than a year, not a single hiccup. YMMV.
I could get my old Haswell 4-bangers down to 30W or so. A modern full-build uATX motherboard w/APU is about the same. 30W. On the other-hand, mobile APU chipsets do much much better. I've had very good luck with NUC style form factors with both Intel and AMD. The AMD 3550H is a real workhorse and idled below 15W. And I have an old Intel NUC style machine (5+ years old) that runs 7W, and a modern Raspberry Pi 4 that also runs around 7W.
However, at 7W the performance is really bad no matter even for basic tasks. The CPUs just can't burst long enough and you can feel it in the response latencies. Still, the RPI4 is able to Home Assistant and even though it can't really do a whole lot beyond simple data copying and automation, that is all Home Assistant really needs. For example, it can't transcode or archive the 10 security camera feeds, but it has no problem doing a streaming fan-out of the feeds to wall-mounted clients. It is also limited by a meager 4G of ram.
Modern mobile APUs from both Intel and AMD at 15W do much, much better because they idle below 15W (10W is possible if you really try), and yet can burst to 30W which makes them very responsive, and you can hang a significant amount of ram and storage off of them all inside a small NUC style case.
I finally settled on the 3550H APU in a NUC style form factor for non-compiling server-type stuff. I can fit 64GB of ram, one NVMe drive (1TB), one M.SATA drive (slot unused), and one 2.5" SATA drive (4TB) all inside, all SSDs. The whole thing runs roughly 15W when idle or under low loads and never gets above 25W. It can burst significant cpu power for a short period of time which makes it very responsive, and it has two ethernet ports. It can do very basic transcoding (repackaging mostly, though certainly not on 10 concurrent video feeds).
I have three of them. One runs primary home functions (dns, mail, web serving, home media archive, routing, firewall, ntp/time, NFS/Samba exports, NAT... almost everything), the second handles repackaging and archiving feeds from 10 security cameras. And a third is the backup box which backs up all the filesystems for all the machines and handles staging to the off-site backup. And I still have the RPI4 handling the home alarm system (it fits in the alarm cabinet with its own backup battery).
Bulk compiles are painful on these sorts of small boxes, but that's what workstations and larger pieces of metal are for... the threadripper, for example, burns 150W when idle and something like 300W going full out. But it doesn't need to be turned on 24x7.
My office workstation with its dual 4K display is also running on a normal uATX mobo combo and eats around 30W, not to mention the monitors that eat 10W each when not blanked. I have tried many times to downsize it to NUC style or smaller and have failed every time. The mobile form factor is just too slow when you have to 4K screens and half a dozen chromium windows open with many tabs in each, plus video playback in at least one going (sometimes more than one). So my workstation needs to be able to burn 100W+ at times and no NUC form factor can handle that reliably.
Basically any particular need has certain minimum resource requirements to be convenient and there are form factor and idle/nominal power consumption requirements which fit that need no matter how you twist it.
-Matt
Me watching this with my 10yo server idling at 110W be like: 👀
Id be happy with 110W mine draws 270W😂
Chopped about 20watts off my 12400 system, changing a SF600 to an Intertech 120watt pico psu, you can do power supplies that are efficient at high and low load loads in the +94 range but it does cost to get to that place, the space industry has done that for years, but it comes at a cost on the price.
Great vid 👍
Based on NYC-area marginal use rates, each incremental 100W of at-the-wall power draw adds roughly US$150.00 per year more in residential electrical cost if left powered on 24/7/365.
Or approx. US$375.00 over, say, the five-year life of a single, 50W-average device and inclusive of a fistful of 3.5" drives (as of December 2023).
Kindest regards, neighbours and friends.
If you stack the power consumption graphs of the gaming workstation and the server on top of each other, you actually get the idea where to go next - buy a desktop system with additional cores/memory, let it run 24/7. Game when you want to but at the same time, have a bunch of VMs/containers run at the same time doing your server stuff. you hardly edit/encode videos while gaming, so heavy workloads usually do not interfere. if that still happens, use even more cores/memory. having just one big system is simpler, and frees you of having to supply a decent network infrastructure, like switches and the like.
Unfortunately, modern GPUs don't have the best power management at idle.
Both NVIDIA and AMD GPUs have had issues with high idle power consumption, and in general, adding a GPU to your system will increase your idle power consumption by 15-50W.
Having to deal with VMs, IOMMU, GPU passthrough and all the associated quirks and bugs also outweighs the advantages of having one machine for me. I prefer having two separate machines for different use cases
@@WolfgangsChannel Perfectly valid points. And just for the sake of completeness, I also run a dedicated gaming machine. Mainly for two reasons: First, I need tons of USB peripherals for gaming (air combat sim, so headtracking, flight stick, throttle unit, pedals and whatnot) and second, my main (work) machine does not have enough slots to add another GFX card - I already have GFX card and an AI accelerator as well as several PCIe nvme devices.
But I did run such a setup (prior to the AI accelerator) to decent success and stability. It takes time to figure out what to do, but sometimes getting there is fun in itself. Granted, you should have recent hardware - but it is an option for new builds.
I can confirm the rather lackluster idle power efficiency in the envelope you named: AMD W6800 idles at 34W, a smaller RX6600XT still does 20-ish watts. However, as I said, you might save on additional network equipment - modern switches with 1Gbit or higher usually operate in a similar ballpark in terms of power consumption.
I will do this one: you can usually scoop a better deal on large memory lots (64 G+) of ECC than non ECC.
So why not?
I'll paste my reply to another comment
Costs as in ECC vs. non ECC RAM sticks? Sure. Registered ECC memory is actually even a lot cheaper than non-ECC, because it's only supported on HEDT/server platforms.
Costs as in actually building a system that supports ECC? That's a bit more complicated. In a homelab environment, you often have to balance power consumption, the overall size of your system, and other features that you might not need in an enterprise/DC environment (e.g. hardware video transcoding). Building just any kind of system with ECC is easy. Building a compact, power efficient system with QuickSync and ECC that doesn't cost an arm and a leg is harder.
So yes, ECC is great to have, but it's not indispensable in a homelab environment. Most of us don't run mission critical services for thousands of people which lose us money the minute they're down.
Reason i went to newer processor because it's capable to handling 4k transcode and supports av-1
Your videos are great! Thanks!
I would like to also see similar comparisons with older xeon processors that are cheap on ebay.
Would defo want a video regarding the most efficient powerful system (like 16-20 cores)
Thoughts on running pi for idle loads and doing something like wake on lan to wake up bigger server for larger loads?
I am building a setup like that. Really effective.
Sounds like a lot of unnecessary complexity and latency
If you don't need a whole bunch of power for things like sata drives, I've had good results regarding efficiency just getting an industrial 12V PSU and strapping a PicoPSU to it. Only an option if you're willing to either work with mains wiring or have a 12V laptop brick lying around, but those usually have pretty good efficiency at lower wattages due to stricter regulations than desktop PSUs.
If all you're doing is wiring a power cord onto an industrial PSU's screw terminals it's not even that scary.
@@thewhitefalcon8539 True, it was more of an obligatory warning than a comment about serious mains wiring project, but it's still mains wiring for which you need to ensure everything is correct and you know what you're doing before you plug it in or touch it afterwards.
You can get such PSUs that take wider voltage ranges, too, making 16V and 19V/20V bricks viable.
@@laurelsporter4569 True, but those usually have lower wattages and make it harder to hard-wire extra connectors to devices if you need more power than the plug supports.
my low power home server solution was 2 RPI 4's (back when they could be had for 40USD)
a few years later still running my home automation stuff with 100% uptime out of a pair of usb ports
they also handle the DNS server load of a 10g network
for my nas i currently have a single bay unit, although i may have to upgrade soon as its not enough to solve the external hdd problem so that may become a home server
but most of the power consumption in my rack is the rest of the network equipment which totals up to 175w, the energy monitor calculates a 2.158kwh daily consunption
other things are my gaming and website + service servers (not power monitored) which are all x86 pcs, they usually do 30w idle except for a few, but power is cheap where I live so I collect data just to put it in a chart
I didn’t know about turbostat, very useful.
Also does the power usage at 3:00 belong to the server shown at 7:41?
So a setup with a small cloud instance that is almost free, that gives the service 24/7 and spins up a worker machine when needed, would have great value. Any scripts for this type of setup?
Well, ARM isn't always more efficient.
Take ampere altra max for example, the 128c model while although it draws 250w, 110w less than the Bergamo 9754, Bergamo is also THREE times faster than it is
Another great video! You explain things really well and taught me a few new things too, and yes birds are real 😂
What pc case are you using for the 8 drives in your home server?
Thank you for a great video!
Thank you for the video. Are there any low power homelab forum community? I am just getting into it and would like to learn more.
This Unraid forums thread is pretty good, and has lots of tips (not just for Unraid) forums.unraid.net/topic/98070-reduce-power-consumption-with-powertop/
hi i want to get too into server but i cant find cheap rack rig , what are u using? do you recoment it?
great video, still trying to decide if i should build or buy.
Wow, this video is what exactly i need to improve my homelab power consumption in next day: change hardware newer the old one like 10Gbps Lan card, remove Raid card (change from ESXI to Promox), use only 1 channel DDR4, change 3.5 Inch HDD to 2.5 inch, remove some unnecessary things like wifi card, led cable (included on motherboard)... I hope power consumption will decrease from 100W to less than 60W
I'm not sure if your serious or not, but you should probably skip the 2.5" HDD and one channel RAM thing.
Most 2.5" HDDs are SMR, not CMR, which is not recommended for NAS use.
Reducing your RAM to single channel will save you 1W at most, but will drastically reduce memory performance
@@WolfgangsChannel Well, thanks for your advice but after 2 years' experience with my HomeLab, i see that my data only use for backup and read only so changing from 3,5 Inch HDD WD Red Plus (CMR) to 2.5 HDD WD Blue (SMR) are not affect with any main services that will be stored at SSD pool (MLC NAND) beside i change from 2 ram stick to 1 because I want to add another Ram stick which is higher capacity later.
2.5" drives are also often less reliable than the 3.5", and buying all new drives even though your old drives are perfectly fine, is probably not gonna reimburse itself.
What I would do instead is just spin the drives down and only spin them up when you're doing the backups.
Really enjoyed this video! Production Quality as always on point! My system idle's around 40w. I have an i7-4790 that was gifted to me. Do you have any advice for me in the same Generation of cpu or should I go with a friend's advice and just buy a 11gen Intel Chip. ( I really really want AV1 Decoding )
Hope you have a great day!
You really should know first about what are you going to use that for, if the 4790 is enough then use it. But if you really need new features from new hardware, then you can upgrade.
4790 should be strong enough for general computing and gaming, let alone a home server. Tho i guess some undervolting is good since it can draw some power under load
I only consider the power consumption of my devices in the summer and power is cheap and cheerful then.
They heat my technical room in the winter with idle capacity being donated to ALS and parkinson research.
Just want to say how excited I am to see the ITX W680 video
Sorry if you've answered this elsewhere before, but how are you getting the data that you use for your really nice power consumption graphs? I'd be super interested in gathering some usage data for my homelab.
Looks like HomeAssistant with a powerplug connected to it
Grafana, Prometheus, Home Assistant and Shelly Plug S
What’s the MSI motherboard model shown at 11:23?
B150i
Hope you end up testing the Minisforum AR900i, however figure out a reliable way to get them to respond to questions... sent them two separate emails with zero response which is concerning from a potential support perspective. Great vid!
Hello there,
2:58 i would like to know if the shown consumption is just for one device or for more than one.
I have my Router, a Palo Alto Firewall, 2 x small 8 Port Switches from HP, 2 x UniFi APs, 2 x Lenovo 1Litre PCs and one DIY Backupstore with an total idle consumption of 110-120w.
Tbh i would like to lower it even further to make it a little bit more cheaper than it is now.
Also thank you for your content. I really enjoy it!
C u
good stuff, appreciate it!
What server chassis is that at the bottom of your rack? I'm about to embark on a building a rack-mountable workstation in a 2U chassis.
Thats my plan too, please can you make a Video about this 2U/2HE Server case
Supermicro SCC833. It's 3U though
I just f*ing love how grounded and thoughtful your work is every time
Doesn't a load of .12 represent an average of .12 tasks in queue? So, roughly 12% of a thread? (far more than 'less than 1% of a thread'?)
That is correct, sorry, I can't do math :(
Yes it does
This is a weird question but what is the font of your terminal at 2:19?
my server idles around 50w and I'm not sure how I can reduce it. I'm running a kabylake i5 (what I had laying around) and a couple of hardrives on debian. nothing seems to get it to go below c3 (no pcie cards are plugged in) (it's probably the motherboard
Stab in the dark, but do your drives spin down? If you're running unraid, make sure that you have a cache drive and your HHDs spin down. That way I get sub 30w with an old Haswell cpu and full atx motherboard.
the 7600k draws about 40w idle so the extra 10w might be your hdd drives
Try a Windows installation (maybe just to get a reference on what is possible). From my experience, Windows (Server) is hard to beat when it comes to power managing the board and devices. It can cut the power consumption of your system in half. Unfortunately, Linux drivers don't seem to be very good when it comes to low power (I guess). Also some Linux distributions might not have installed the correct GPU drivers for the system, in this case the GPU won't clock down or blank / switch of video ports when running "headless". I tested some Fujitsu Futro Thin-clients in this regard, Windows idle at 3.8W vs. Linux / Proxmox / Debian idle at around 7W e.g. This is quite a difference, if anybody has some tips on how to get Linux power consumption down, please post them. o)
Correct GPU drivers are really important from my experience, a Windows 2019 Hyper-V Server, which does not accept the regular GPU drivers, will also eat significantly more power when idle, compared to a Windows (Server) with correct GPU drivers installed. There is HDMI Audio and maybe some other subsystem connected to the GPU as well, if these things don't get properly initialized and powered down, consumption is higher than needs to be.
Unrelated question, which macbook are you using in the background in the desk setup?
Sir Wolfgang,
Thank you for this video!
I have a question what do you mean with scanning documents?
Because i use a snapscan ix1600, could u elaborate?
Thank you and happy new year!
Ajay
I use Paperless to consolidate my scanned documents into a searchable database
What would you reccomend for a psu as an alternative to the corsair rm550x?
Today, TDP means absolutely nothing. Base clocks are rarely the frequency at which a CPU runs, unless the user blocks it from turboing.
Preach it, brother. People always make big eyes at the thought of having an extra box in the corner as it will eat 500 watts 24/7, when in reality it won't be more expensive than having a freezer or a porch lighting. The difference in monthly fees is not even in double digits.
what is the motherboard, shown at 14:45 ? I couldn't hear well or didn't understand well 😢
ok yea TDP tells not much abut the cpu but you have to watch about the TDP if your cooling solution cant handle more and you do sometimes heavy load. i do some of this sometimes and my server is completely fanless and can handle only 45w and at burst load 60W but this only for 20sec then the whole pc case heats up to 65°C and plus. i have chosen the 5700g and set the TDP to 45w. the problem with the case is that the single HDD i have in is connected to the case and its heat gets also to a part in the case and if the case gets over 70°C the drive shuts down and if the temperature rises even more the data on the drive can be damaged. so i have to use a lower TDP because of that and my cpu use. for the most part my system idles like yours, but like you i have some really heavy tasks that max out the cpu and the cpu goes 100% on all cores. i do analytic stuff with weather data and do abit of keras magic and rendering and this needs power and those runs running for about 10h or more and every other task is in this time in pause so i have the full machine power. now i have a r630 for this becuse that runs are now 80h and need massive ram and storage. i do this particular task for my maker space i am in but i have private project that need some power and having my main server packed for hours is not ideal. so i start up my r630 only i have to because 130W idle and 570W at full load with all 44 cores 100% and all 8 ssd fully packed with writing and reading. i load much as i can in ram but i have only 768gb in this machine but i can only load 200gb each time because the rest of the ram i need for the program itself but with this machine i am able to do this in 3h instead of 80h. my 45w system needs in this time 3.6kwh and my r630 only 1.7kwh so it is way efficient as the smaller system but idle power is massive compared.
so yea different use cases needs different solutions for the best efficency
Hi, re: laptop power draw being more than desktop power draw, was the screen on during the test?
Measured it with the screen physically disconnected as well, the power draw remained the same
What about running on a DC power supply (like a laptop brick), I've no idea if it's possible, especially if there are a dozen hard drives, but would it be more efficient?
I've ran psus on laptop chargers and even some USB 20v phone chargers.
They can be really efficient, just once you're within range
It's possible. Not sure what the point is
Possible and doable with a PicoPSU, but definitely not with a dozen of hard drives. With that many HDDs, the spinup spike will probably be too much to handle, even for some ATX PSUs.
That being said, PicoPSUs are usually more efficient than standard ATX PSUs, but also don't come with a lot of SATA/Molex plugs. So you'll either have to re-wire them yourself or use SATA splitters.
@@WolfgangsChannel PicoPSU is just for 12V input to fulfill all the ATX motherboard voltages. All the other 12V lines should be connected to whatever is powering the PicoPSU. Since SATA also has 5V lines, maybe you can leech them off the PicoPSU or somewhere else because the big power draw should be 12V.
Grüße, Wolfgang. Ich möchte dich gern auf das Board N100M hinweisen, welches einen N100 verlötet hat und dennoch 2 PCIe Slots bietet. Ja, diese rennen nicht x16 fach, sind aber prima für 6x Sata Karten geeignet. Weiterhin hat es den normalen Stromanschluss, welcher ein Netzteil erlaubt, das genug SATA Anschlüsse hat (mein Versuch mit 5x HDD + Pico PSU ist... schwierig).
Kleiner Hinweis: Alle N100 Consumerboards (es gibt nur 3) haben jeweils einen Ramslot, der jedoch gern auch 32gb frisst, auch wenn es anders dokumentiert ist.
Dennoch rennt auf dem Ding ein Unraid mit NVME Cache, 5 HDD, 23 Docker Containern und das härteste was die CPU stemmt ist Plex. Transcoding auf dem Teil läuft prima. Ich bin völlig zufrieden!
Wo finde ich dieses Board und wie ist der Idle Watt Verbrauch ?
Great content. Really answering a lot of questions in this niche.
Mythos Nummer Fünf!
Ich frage mich bei jedem Wolfgang Video, ob er nicht mal ein Video auf deutsch machen kann ;)
Hi Wolfgang.
Many people probably don't need a NAS for their setup, but probably instead need a DAS. Can you recommend a DAS solution to provide redundancy and protect data?
Very nice and informative video, especially for someone who prepares to build his first NAS. Great to see the quality of your content improving, and I'm really glad I've found your channel. You are a constant source of inspiration and entertainment. Thank you for that!
С наступающим Новым Годом, Вольфганг!
8:55 Depends on the software used. Look up the benches.
But say that is fact 100% of the time: the apple you (ajnd half the world) seems overly fond of, is not just an M1 chip. It is a cpu, gpu, _and_ operating system sold as a package. If Intel sold a chip that has their own os perfectly suited to their hardware only, I bet you can see insane gains. Google can do the same (and is working on it). Then MS has to come up with their own and then you have the situation where you must buy into one of the ecosystems with their own payment system, their own app store, their own everything.
Seems pretty bad to me.
2:18 anyone, can you show me any article, videos or tutorial to achieve this S3 management?
You can spare me with assuming I'm a novice.
I've experimented with 2nd Gen, 7th gen, a xeon server, but still can't get what I want. the 7100T still draw 30W before and after. The same with 2395T