TVHeadend build - PSU fan as air mover for CPU

[DonP]

I'm looking to build a very quiet TVheadend system.
It'll be on 24/7 so I'm not going to go completely fanless.

I might also be running a handful of other services on the box (Zoneminder perhaps), so I'd rather have a little oomph in the CPU.

I'm going for a uATX board, so I have some slots to play with. For DVB cards, and perhaps an nVidia graphics card, so it can do transcoding to x265 (yes, I know Intel QuickSync can also do hardware x265 encode on the IGP. but I'd like the nVidia option too).

I'm considering a 35W processor, perhaps the mythical i9-9900T, if that becomes a thing in the near future, or just an i7-8700T. But 35W almost for sure

Mobo will probably be a SuperMicro X11SCQ, if I can source one, or a X11SCZ-Q, worst case a X11SCZ-F. I prefer to have more, smaller slots, because most cards will be 1x, maybe 4x PCI-E.

If I could I'd like the smallest possible case, and only zero or one case fans, and a PSU fan.

I'm considering the Coolcube Maxi V4 case.

That case has a CPU cooler clearance of 120mm.

Conveniently, the Nofan CR-80EH is 113mm tall, and the Noctua NH-C14S is 115mm tall, so they should both juuuust fit.

That would put some or most of the heatsink near the PSU fan (I'm hoping they line up, at least a little bit).

So here's my question:
If the PSU I choose is *not* a semi-"fanless" design (ie it's not one that only spins up above a certain temp/load), so actually runs at low RPMs even on low load, then would it be enough to keep things running along in such a case? And without any other case fans?

Has anyone tried using the PSU fan to suck air through the CPU cooler?

TIA

[CA_Steve]

Since the PSU is top mounted in this case and there isn't top ventilation, by default the PSU has to be mounted intake fan side down (facing the CPU). How well it could pull air through the CPU cooler is a fun science project. I'd lean toward the Noctua (if it'll fit) as the top of the cooler would be within 5mm or so of the PSU. Worst case, you could mount it's fan underneath. That said, this is a big assed cooler for a nominal 35W CPU.

Some design considerations:
- Are you going to be running Windows or linux or? Have you looked at the Supermicro mobo's fan controls to see if they meet your needs? When I think of Super Micro I think servers..and not something tuned for low noise operation.

- I detest these nerfed 35W CPUs. It's like putting a performance engine in a cheap car with bald 13" tires. Yeah, I might make some noise, but it isn't going to get there any faster than stock. Figure out your performance needs (Zoneminder faq talks about using an old Pentium 3 ) and scale accordingly.

- I'm also looking at building a video head end...but I want it to be more of an appliance than PC...I need less operating systems to manage. I'm leaning toward getting a prebuilt NAS box and putting Plex Server on it. Using HDHomeRun or similar for network attached tuners and I'm done. My storage needs are pretty minimal as I don't hoard video. So, I might go SSD based as well.

[Klusu]

Yes to both questions. The downside - if you put in much effort eliminating all the gaps, it's hard to change components afterwards.
Don't be so sure about 35W.

[nonhinged]

Did that on an really old mini-ITX system. The psu fan was the only fan in the system.

On some modern PSUs it's possible to set fan curves and what sensor it reads. For example on my current psu(corsair hx1000i), I can use a software called iCUE to set the psu fan speed dependent on CPU temp or other sensors.

[DonP]

Did that on an really old mini-ITX system. The psu fan was the only fan in the system.
On some modern PSUs it's possible to set fan curves and what sensor it reads. For example on my current psu(corsair hx1000i), I can use a software called iCUE to set the psu fan speed dependent on CPU temp or other sensors.

Thank you for the reply!
That's a great idea re tweaking the fan control. I'm a little old school so I prefer closed-loop hardware control circuits to uC controlled software, but it's a great idea.
My bigger concern with the PSU is that the fan speed may be a function of PSU load rather than PSU temperature. Maybe I'm misreading the specs but looking at some of the Silverstone PSUs they plot load, and not internal temp, against fan speed. I'm looking for a "low-power" modular (for cleaner cabling) PSU.. but because its modular that'll mean 500W. I think I'd be fine with 350W, if such a thing existed.
Thanks again!

Yes to both questions. The downside - if you put in much effort eliminating all the gaps, it's hard to change components afterwards.
Don't be so sure about 35W.

Yes, I stupidly didn't think about needing space for construction.
re the 35W - I don't really believe that it's 35W - ever since it became a marketing tool for Intel. This article is also concerning.
So what would you suggest? Get a 65W i7-8700 (non-T) and underclock? Not that I put that much stock in Passmark benchmarks, but the 8700 at 65W isn't that much faster than the "35W" 8700T: 15148 vs 12692.. which is why I favoured the 8700T.
Thank you for replying.

[CA_Steve]

Nearly all PSU fan profiles are temp based, because it's cheap (thermistor)..the graphs shown are load vs rpm estimates.

[DonP]

Since the PSU is top mounted in this case and there isn't top ventilation, by default the PSU has to be mounted intake fan side down (facing the CPU). How well it could pull air through the CPU cooler is a fun science project. I'd lean toward the Noctua (if it'll fit) as the top of the cooler would be within 5mm or so of the PSU. Worst case, you could mount it's fan underneath. That said, this is a big assed cooler for a nominal 35W CPU.

Actually, the PSU is side mounted - that picture is with the top lid removed and looking down
In the distance you can see the grille at the bottom of the case. Sweeet.

Some design considerations:
- Are you going to be running Windows or linux or? Have you looked at the Supermicro mobo's fan controls to see if they meet your needs? When I think of Super Micro I think servers..and not something tuned for low noise operation.

Linux, all the way. And either docker, or VMs. I'll use PCI passthrough for the cards if VMs. I'm just not sure whether Intel QuickSync will work on virtualised CPUs. Probably won't.
I've built a fair few Supermicro servers and never looked at the fan controls TBH - they're servers Having said that, the PSU fan is controlled by the PSU, not the mobo. I think.

- I detest these nerfed 35W CPUs. It's like putting a performance engine in a cheap car with bald 13" tires. Yeah, I might make some noise, but it isn't going to get there any faster than stock. Figure out your performance needs (Zoneminder faq talks about using an old Pentium 3 ) and scale accordingly.
- I'm also looking at building a video head end...but I want it to be more of an appliance than PC...I need less operating systems to manage. I'm leaning toward getting a prebuilt NAS box and putting Plex Server on it. Using HDHomeRun or similar for network attached tuners and I'm done. My storage needs are pretty minimal as I don't hoard video. So, I might go SSD based as well.

When I used to be a regular here the big thing was underclocking by 10%, so you could undervolt by 15% (I hope I'm remembering those figure correctly, but you get what I mean). With the inverse square law temps would get pretty good. Back then a 35W CPU was a 35W CPU.. today.. I haven't a clue. I like the idea of the hard limit (even though it might not be 35W).
Because of the tight fit and possible construction complications I'm considering a shorter cooler and a shroud.. but I'm not sure I'd want to do the whole DIY thing nowadays.
Are the 35W variants so bad? I posted the Passmark benchmarks above - they look rather good IMO.

re appliance and pre-built.. perhaps, but I like the power and flexibility of a build like I describe. Using someone else's docker images simplifies the software upgrades.. but then again TVH have their own apt servers, so updates are trivial under Ubuntu.

Thank you very much for your replies.

EDIT:

Nearly all PSU fan profiles are temp based, because it's cheap (thermistor)..the graphs shown are load vs rpm estimates.

That would have been my guess too.Ok, so that's great for my use case.

[CA_Steve]

Intel doesn't cherry pick parts for 35W...all the CPU variants follow the same to-get-this-freq-apply-this-voltage curve. Yeah, turbo can be more aggressive for the lower TDP parts...but Intel is speccing TDP for the Base CPU freq..not at turbo. So, if you need the horsepower, just get the std TDP parts. You can always nerf them yourself to fit into a power budget.

[Klusu]

I'm hoping they line up, at least a little bit

I achieved perfect line up by putting a 9cm fan in the right place in the PSU (instead of the 12cm fan). I use i3 and IGP (to keep power draw low).
35W CPUs are bad if they cost more. A 95W CPU will spend most of it's life drawing less than 35W; some workloads will be processed faster, and the CPU will return to low power sooner. Not so much difference in total kWh.

[DonP]

So I've given this some more thought.

i7-8700T is hard or impossible to source, realistically outputs more than 35W, and is locked. i7-8700 is locked. i7-9600K is more powerful than I need. i5-9600K should be ok. It's also unlocked, but 95W TDP.

So now comes the question - how does one undervolt nowadays?
Back in my day you just set the CPU voltage in the BIOS, tested for stability, and be done with it.
Reading around it seems there's a different voltage depending on how many cores are in use, and it'll also bump-up the voltage if AVX instructions are being used. Rather than setting absolute voltages there's also this concept of voltage offsets.

Looking at motherboards I'm now thinking the Asus Prime Z390M-PLUS because it'll have more BIOS tweaking options than a Supermicro board, and it's cheaper, easier to source. Maybe the GIGABYTE Z390 M GAMING.

But back in the day a motherboard manual would list all of the entries in the BIOS, as well as all the possible options.
Today, I guess there are too many to list. I note it's also possible to undervolt in "software" with Intel XTU or something, but I want a BIOS solution - I need the undervolt to be in place in case of an OS malfunction.

I'm willing to try my chances if someone just says "Don't worry, with an unlocked CPU an ASUS mobo/BIOS will let you undervolt and get the TDP to something manageable"

Is this the way to go?

Cooling set up as before - Noctua NH-C14S (without dedicated fan) aligned to the PSU fan.

Thank you all for your invaluable help.

[CA_Steve]

A few years back (ok, might have been 5-7 yrs ago ) when Intel implemented the more involved and much faster voltage and freq stepping based on load need, a new option was added to many mobo's - Core voltage offset. Adjusting this didn't affect the intricate freq vs load dance, it just offset the voltage applied while under load. Add positive offset for overclocking, or go negative for undervolting.

You will need to peruse the mobo manual screenshots to see if this option exists. It's more likely in the enthusiast SKUs than the really cheap, base level models. Also, there was a time where the mobo might permit positive offsets to please the overclockers..but didn't have the ability to do negative offsets (I ran across this with my MSI Z87 gaming mobo). So, if in doubt - ping the mfgr's tech support to see if the mobo you like has the option.

[DonP]

So, if in doubt - ping the mfgr's tech support to see if the mobo you like has the option.

So, I bounced a few e-mails with ASUS technical support. The general response was to refer me to other ASUS motherboard models and that I should have a look at the manual for that motherboard or search for reviews because the BIOS options will be similar (similar.. is not identical). They also repeatedly referred to overclocking and overvolting even though I corrected them a few times.

Anyway, I'm not willing to take the risk.
According to Seasonic's Wattage Calculator the i5-9600K (95W, 6C6T) defaults to 3.7 GHz and 1.2V, while the i7-8700T (35W, 6C12T) defaults to 1.0V and 2.4GHz, and the i5-7600T (35W, 4C4T) 2.8GHz and 1.0V.
Doing some rough math (power scales as f*v^2) undervolting and underclocking an i5-9600K to 2.8GHz/1.0V would put me in the 50W ballpark, if I [theoretically] disabled two cores, that would put me at 33W.

But that's too many unknowns and I'm not willing to take the risk if it turns out the motherboard can't undervolt at all, or by as much as I need it to.

For now I think I'll pick up a NUC8I5BEH - not quiet, and it doesn't have the PCIe slots I need, obviously, but it'll be fine for the next few months until I come across a proper review or blog post by someone revealing the undervolting options of a good LGA1151V2 mATX motherboard. It's also around half the price of the 9600K build. In time perhaps the new R0 stepping Intel CPUs might turn out to be good.

Thank you all for the help and advice. I greatly appreciate it.

When/if I do end up doing an undervolting build I'll be sure to report back here.

Cheers!

[CA_Steve]

Have fun!