hmm - are you sure the Cooljag you mentioned would fit the new Denverton CPU? Intel Atom® Processor C3758 (16M Cache, up to 2.20 GHz) Produktspezifikationen. My understanding is the solution is very similar to the solution I have on my D525 Atom which you can see here HFX micro line, i.e. only a thick copper plate replacing the heatsink currently used on the none "+" version of the board plus heat pipes coming into play.
Atom C3000 / Proxmox VE / SOHO - 4 or 8 cores?
- Thread starter TKH
- Start date
Notice: Page may contain affiliate links for which we may earn a small commission through services like Amazon Affiliates or Skimlinks.
I think the only "real" mITX standard is the size of the board - there's certainly so standard on socket placement not the size/spacing of the mounting holes. As noted by a lot of the people on this forum, most of the HSFs for the Atom and Xeon-D models have been a not-very-common 51x51mm spacing. If you want to utilise one of these you're almost certainly going to need a custom mount, particularly if the stock heatpipes aren't in the right place. As with my streacom, you need to be very careful about mobo arrangement as well.
IIRC the HSF Supermicro use on some of their Xeon-D boards is the same P/N as that of the C3000 boards so the Cooljag should fit, but you'd want to check properly first. StH themselves did an article on aftermarket cooling solutions showing a couple of the other HSFs than fit the 51mm spec;
https://www.servethehome.com/aftermarket-cooling-for-the-supermicro-x10sdv-xeon-d-motherboards/
This does all feel a lot like square peg in a round hole or the tail wagging the dog though - if you were looking for an extremely low TDP unit for an existing case you should have said so
For the stated purpose though I'd be wary of using the C3000 for what you describe but your most important criteria seems to be being able to use your existing case.
Quick'n'dirty benchmark creating a tar.xz (not using pxz and defaulting to single thread) on the same dir (containing 238MB of FLAC files) on both machines;
C3758 is approx. half the speed of the Xeon in CPU-intensive single-threaded tasks like this. Switching to pxz to attempt to use all available cores;
4 cores vs. 8 narrows the gap considerably although the Atom is still performing at ~66% of the E3.
Edit: performing the same tests on an old Ivy Bridge 1265Lv2 (bought long ago for a TDP-limited project with base 2.5GHz and 3.5GHz boost clock);
...and it's still faster than the C3000 and pretty much the same speed as the E3 v3 (no real surprise there). Long story short, I'd plump for a Xeon if I were you even if it means forking out for an expensive L model, especially if it means you won't need to fork out moolah on a custom-fitted cooling solution. But fitting any sort of desktop virtualisation server into a 35W TDP is a tall order IMHO.
IIRC the HSF Supermicro use on some of their Xeon-D boards is the same P/N as that of the C3000 boards so the Cooljag should fit, but you'd want to check properly first. StH themselves did an article on aftermarket cooling solutions showing a couple of the other HSFs than fit the 51mm spec;
https://www.servethehome.com/aftermarket-cooling-for-the-supermicro-x10sdv-xeon-d-motherboards/
This does all feel a lot like square peg in a round hole or the tail wagging the dog though - if you were looking for an extremely low TDP unit for an existing case you should have said so
For the stated purpose though I'd be wary of using the C3000 for what you describe but your most important criteria seems to be being able to use your existing case.Quick'n'dirty benchmark creating a tar.xz (not using pxz and defaulting to single thread) on the same dir (containing 238MB of FLAC files) on both machines;
Code:
effrafax@E3-1230v3:~$ time tar -c /path/to/flac/dir/|xz -c > /dev/null
tar: Removing leading `/' from member names
real 1m24.014s
user 1m23.780s
sys 0m0.484s
effrafax@C3758:~$ time tar -c /path/to/flac/dir/|xz -c > /dev/null
tar: Removing leading `/' from member names
real 2m49.212s
user 2m46.010s
sys 0m2.469s
Code:
effrafax@E3-1230v3:~$ time tar -c /path/to/flac/dir/|pxz -c > /dev/null
tar: Removing leading `/' from member names
real 0m20.390s
user 1m44.752s
sys 0m1.100s
effrafax@C3758:~$ time tar -c /path/to/flac/dir/|pxz -c > /dev/null
tar: Removing leading `/' from member names
real 0m36.750s
user 2m52.431s
sys 0m2.236sEdit: performing the same tests on an old Ivy Bridge 1265Lv2 (bought long ago for a TDP-limited project with base 2.5GHz and 3.5GHz boost clock);
Code:
effrafax@E3-1265L-v2:~$ time tar -c /path/to/flac/dir/|xz -c > /dev/null
tar: Removing leading `/' from member names
real 1m25.946s
user 1m25.840s
sys 0m0.736s
effrafax@E3-1265L-v2:~$ time tar -c /path/to/flac/dir/|pxz -c > /dev/null
tar: Removing leading `/' from member names
real 0m20.989s
user 1m51.668s
sys 0m1.364sThanks a lot for running the tests comparing the CPUs for me!
In my initial post I was set for getting the ATOM C3000 (which I would have attempted to put into the case and only if that would not have worked would I have considered to use fans or even invest into a current fanless case) - my main question there was only do I need 4 or 8 cores....
Based on the feedback and advise received here I am now going to go for an E3 1240L V5 (or even 1260L) as you also suggested. Impactics confirmed they would also be able to deliver a cooling solution for this set-up. Sent them pictures of the case today and in case of no bad surprises hopefully decision for the exact CPU and mainboard to best be used in this specific case can be made next week, parts ordered afterwards and finally case & parts sent over to them for assembly. Curious to see if (and how) the desired result will be achieved. Will share how this progresses.
In my initial post I was set for getting the ATOM C3000 (which I would have attempted to put into the case and only if that would not have worked would I have considered to use fans or even invest into a current fanless case) - my main question there was only do I need 4 or 8 cores....
Based on the feedback and advise received here I am now going to go for an E3 1240L V5 (or even 1260L) as you also suggested. Impactics confirmed they would also be able to deliver a cooling solution for this set-up. Sent them pictures of the case today and in case of no bad surprises hopefully decision for the exact CPU and mainboard to best be used in this specific case can be made next week, parts ordered afterwards and finally case & parts sent over to them for assembly. Curious to see if (and how) the desired result will be achieved. Will share how this progresses.
Well don't just trust one completely arbitrary benchmark, but glad you found it useful at any rate. In general I think an E3 Xeon's the best option, although I'm slightly miffed by Supermicro's lack of mITX choices for the 1151 socket, so I think the ASRock E3C236D2I is the only game in town in that regard. The C236 WSI's is of course an option although then again you're going to be using a custom HSF block and limited to using SODIMMs rather than regular UDIMMs.
Best of luck with your build and, given the discussion here and the fact you're going to be using a somewhat exotic cooling solution, you should chuck up a wee build thread when things are underway
Best of luck with your build and, given the discussion here and the fact you're going to be using a somewhat exotic cooling solution, you should chuck up a wee build thread when things are underway
[QUOTE="so I think the ASRock E3C236D2I is the only game in town in that regard. The C236 WSI's is of course an option"[/QUOTE]
The ASUS P10S-I I had mentioned above is a lot less expensive than these two ASROCK boards - is there anything I am missing why this is not an option? It is 150 EUR versus ~200 EUR and I can't see any difference that seems relevant and am asking myself why the ASUS costs a lot less and if no real difference why to spend more on ASROCK.
The ASUS P10S-I I had mentioned above is a lot less expensive than these two ASROCK boards - is there anything I am missing why this is not an option? It is 150 EUR versus ~200 EUR and I can't see any difference that seems relevant and am asking myself why the ASUS costs a lot less and if no real difference why to spend more on ASROCK.
Last edited:
This. A low RPM 150mm fan just doesn't make a lot of noise. If you have some air moving through a reasonably sized case, a CPU fan will also throttle down, and a good one of those isn't very loud either when running at low speed. With that sort of configuration the only time the fans are going to spin up is if you're doing the kind of processing that the low TDP chips can't even do. For some applications I go passive (usually things like firewalls, where I want to eliminate moving parts). For anything computationally intensive, it almost always makes more sense to just use a big slow fan.
Depends whether or not you want IPMI - on the Asus board it appears to be an optional extra (i.e. extra £$€ and likely supportability issues), and personally I had enough run-ins with Asus BIOSes under linux some years back to relegate them to "only if I absolutely have to". Supermicro would be my first choice, ASRock second.
@mstone: thanks for sharing advice and experience - this is the feedback I am looking after. However in my specific case the chassis (which was expensive at the time and which I still like a lot) is very small and does not allow a fan (or fans) being installed into the case itself. The only fan I could install though is indeed a fan sitting on top of the CPU direct - whether that would lead to enough air flowing through the case or not I don't know. Using whatever fan would make the use of a passive case somewhat a mute point.
The company willing to install their passive cooling solution (CPU heatsink and heatpipes to connect to the chassis walls) into my existing case say their solution can handle the heat as long as
a) I buy a CPU that is limited in terms of TDP (i.e. my CPU of choice would be the E3-1260L v5 which is the (only) one specified at 45W and the highest performance L-version CPU) OR
b) I buy a mainboard that allows to limit the TDP of the CPU of a normal E3-12xx v5 (or even v6) CPU. The ASROCK board seems to allow for that - I have not yet received a definitive answer from ASUS support for the cheaper P10S-I.
@EffraxOfWug: as the server will be located under my desk I would think I do not need IPMI. I had checked Supermicro as well, but they do not really offer a similar board - and the one they offer (with an embedded E3-1585 I guess it was) was so incredibly expensive I did not want to consider it.
The company willing to install their passive cooling solution (CPU heatsink and heatpipes to connect to the chassis walls) into my existing case say their solution can handle the heat as long as
a) I buy a CPU that is limited in terms of TDP (i.e. my CPU of choice would be the E3-1260L v5 which is the (only) one specified at 45W and the highest performance L-version CPU) OR
b) I buy a mainboard that allows to limit the TDP of the CPU of a normal E3-12xx v5 (or even v6) CPU. The ASROCK board seems to allow for that - I have not yet received a definitive answer from ASUS support for the cheaper P10S-I.
@EffraxOfWug: as the server will be located under my desk I would think I do not need IPMI. I had checked Supermicro as well, but they do not really offer a similar board - and the one they offer (with an embedded E3-1585 I guess it was) was so incredibly expensive I did not want to consider it.
Last edited: