Thanks luckylinux, for educating me, or trying to...Sure, any day of the Week. Even though I'm not a PFC Expert, I am regularly working with Power Converters. As I always need to stress to everybody: don't believe the Marketing Hype ! An Operating Point and Performance has to be specified at a PRECISELY Defined Operating Point. MANY Parameters are involved in that. Grid Voltage. DC-Link Voltage. Switching Frequency. AC/DC Load. Auxiliary Consumption. Reactive Power (if Active Front-End on AC side). Temperature. Etc. Otherwise it's just Marketing like "up to 95%" Efficiency or "from xxxx $". And in reality at YOUR operating point you might get 3x worse Performances for 3x higher Price
You might want to look into LED Lightbulps for one. Let's have a talk about Power Factor there, because that's probably as bad as it gets (for the Grid or if you have any Contactor/Relay/Breaker/Fuse installed), that's a very nice Capacitive Load that can lead to Contactors Welding or Breaker/Fuse tripping on Inrush (> 10x Nominal Current). Or cheap Mean Well AC/DC PSUs or other "cheap" (reasonably Priced) Manufacturers for 25-300W PSUs. Same for your Phone Charger or Laptop AC Adapter for that Matter, especially at low Load. Tell me about your Power Factor there. And don't mind an Inverter-based Heatpump, those do not even have Active PFC, here in Europe most are just standard 3ph Diode Bridge Rectifiers (although PF for 3ph Systems is already much Better).
Also remember ... Power Factor != cos_phi. Power Factor is cos_phi multiplied by the THD !
For ATX Power Supplies, while there is usually Passive PFC (mostly on old Designs - filters are Expensive and Bulky after all) or Active PFC, the question is: at which AC Voltage you are Operating and at which Load you are Operating.
I just gave it a try right now to prove my Point to you. Seasonic SSR-1000PD Active PFC F3 PRIME ULTRA PLATINUM 1000W PSU. 80+ Platinum. 230 VAC. Load 30-40W measured on AC side (3-4% or Rated Power) when AMD 5950x is idling on low Power, same for the GPU. The Kill-A-Watt meter is fluctuating between 0.37 and 1.00 Power Factor, oscillating quite a lot whenever the Load (slightly) Changes. Simply because it's very difficult to take a sinusoidal input current (and precisely measure / calculate the required Current) at such low load and there is not enough discharge of the High-Voltage side DC-Link / Capacitors in order to do that. Instead it's probably periodically "pumping" some current in order to recharge the Capacitor.
(It would probably be better to use something like an Eastron SDM 120 or even Eastron SDM 630 or just put a scope with Differential Voltage Probe and Current Probe to get a better Measurement though)
There is a PFC Circuit, yes, but the sizing of it is (probably) made on the Assumption that you are at least at 20%+ Load or something like that.
Then, what happens if (your PFC Transistor Switching Frequency being constant), your Input Current is so small (compared to the Design Operating Point) that you go into Discontinuous Conduction mode ?
Take a Look here to see the Waveforms or even more of them. But if now the PFC Ripple becomes large compared to the average Input Current required by the Load, then you hit Discontinuous Conduction Mode (the border case is described in the Figure on the Right side in the 2nd Link). And you obviously do NOT want the PFC to increase the Switching Frequency at low Load, that would thrash the Efficiency completely !
115 VAC is easier than 230 VAC to always boost (you definitively do NOT want to boost too much, otherwise for 230 VAC you go too high) so the Power Factor is usually better in 115 VAC of course (also for passive PFC for that matter).
But at low Load (whatever "low" means depends on the Power Supply and many Design Choices), do NOT expect the Power Factor to be very good. Same for the Efficiency for that Matter.
The Ripple Current is, depending on the design, 10-20% of the Nominal Current (depending on several compromises: size of Inductor/Filter, choice of Switching Frequency, target efficiency, etc), so if you are at "low" Load you are not really "following" a Sinusoid. It's more similar to a passive PFC at that point. Also because the voltage on the High voltage DC Capacitor doesn't "drop" as expected since the load is so small ... so the PFC Controller will not require (or not be able to demand) higher AC current.
Above say 20% Load I'd expect the Power Factor to be relatively good (> 0.90 or even > 0.95).
EDIT 1: the "Energy Star" Initiative for Instance requires: "True power factor of 0.9 or greater at 100% of rated load." That's a funny one, because you are NEVER going to fully load your PSU
EDIT 2: The official 80+ Certification States that:
- PFC ≥ 0.90 for 80+ Gold at >= 50% Load
- PFC ≥ 0.95 for 80+ Platinum at >= 50% Load
80+ Titanium is even more Strict in that Regards: PFC ≥ 0.95 at >= 20% Load.
But yeah, nothing Required below 20% Load ...
1U Supermicro Server 6x 10GBE RJ45 X10SLH-LN6TF LGA 1150 H3 X10SLH-N6-ST031
- Thread starter Bert
- Start date
Notice: Page may contain affiliate links for which we may earn a small commission through services like Amazon Affiliates or Skimlinks.
No worries.
In reality MAYBE the Power Factor it's not *that* bad especially if it's a quality PSU, but it's still something you should check. At least that should be the way to do it: assume the Power Factor is not great as a starting point. Never assume otherwise
. Then if measurement prove that it's better than expected all the better of course.After all, since it's not required to be that good and every manufacturertries to save pennies on whatever they can when designing their PSU, I would NOT expect that they do significantly better than the minimum required.
EDIT 1: my comment was even more true and generic in regards to your statement about "SMPS" (not necessarily ATX for a PC). SMPS essentially just means that it's NOT a Linear Regulator but it's a Switching Converter instead. It does NOT make ANY claim about Power Factor, Efficiency, etc.
Last edited:
Did you ever print this out/check the fit? I'm keen to set this up, especially if it'll fit properly in the 1u chassis that it was built for, though I have a feeling that wouldn't work.
I didn't test it, but it definitely won't fit in a 1U, for that your only options are 40mm fans and the stock passive heatsink. I think my duct would need a 3U at least, because 80mm fans don't really fit above a motherboard in a 2U.
I was thinking that might be the case. I've found some official 1u active heatsink/fan combos for the E3 1150 socket, I'm just concerned that the fan is going to be extremely loud, like the other fans are that I'm planning on swapping out by adding the 40mm fans to the other heatsinks using the 3D printed shrouds.
I suppose there is another option to make a custom centrifugal blower setup, if you're up for it, perhaps two of these each ducted to half of the heatsink (so they don't interfere with each other) could work: Dell Inspiron OptiPlex Vostro Cooling Fan AVC BASB1120R2U 3WY43 03WY43 | eBay
That's kind of what this is, no? Supermicro SNK-P0049A4 1U Active CPU Heat Sink For Intel Socket LGA 1156 | eBay
I mean you could place the blowers in the front of the chassis and duct them onto the included passive 1U heatsink, would probably be quieter since the individual blowers are larger and wouldn't have to spin as fast, plus that 1U heatsink has a lot more capacity than the little one stuck under the blower.
True. I'm not super keen on the official one, because I've had fans like that before and the point is to try and lower the sound levels, not raise them, which that would assuredly do. I might try a total jank method of attaching an 80mm fan to the heatsink directly, though there's no copper to help with the heat dissipation :/.
The stock 1U heatsinks have a spiral of heat pipes in the base, or at least the LGA2011 ones do, pretty sure they're constructed the same. Simply laying a fan on the heatsink doesn't get you great flow though because the fan doesn't move any air at the hub and most of the airflow from the perimeter just hits the outside edges of the heatsink fins and then exits.
Old Thread but I was wondering, given the high Power Consumption Figures ... Did anybody check to see if ASPM is enabled ?
My Experience on the X10SLL-F/X10SLM-F Series at least showed me that you need:
- ACPI Patching of FADT to allow ASPM: GitHub - luckylinux/acpi-linux-patching: Some Scripts to Patch ACPI Issues (mainly FADT for ASPM Support)
- ASPM Tweaking Script: GitHub - luckylinux/aspm-troubleshooting: aspm-troubleshooting (run
- BIOS: enable ASPM for both CPU and PCH Slots, you might need to Force L0L1 instead of Auto etc
Check Result with
It could still be a good small Router with those NICs, even though they only support 1gbps and 10gbps (2.5gbps and 5.0gbps are not supported) and SFP+ RJ45 Modules are both expensive and Power Hungry.
My Experience on the X10SLL-F/X10SLM-F Series at least showed me that you need:
- ACPI Patching of FADT to allow ASPM: GitHub - luckylinux/acpi-linux-patching: Some Scripts to Patch ACPI Issues (mainly FADT for ASPM Support)
- ASPM Tweaking Script: GitHub - luckylinux/aspm-troubleshooting: aspm-troubleshooting (run
python3 aspm.py --auto and, if there are stubborn Devices, force them to enable ASPM with python3 aspm.py -d 02:00.0 for Instance)- BIOS: enable ASPM for both CPU and PCH Slots, you might need to Force L0L1 instead of Auto etc
Check Result with
powertop / turbostat.It could still be a good small Router with those NICs, even though they only support 1gbps and 10gbps (2.5gbps and 5.0gbps are not supported) and SFP+ RJ45 Modules are both expensive and Power Hungry.
Does somebody know of a "Successor" to this Board though ?
Like a Xeon E3 v5/v6 or E-21xx/E-22xx, possibly using a X550 instead of a X540 NIC ?
Like a Xeon E3 v5/v6 or E-21xx/E-22xx, possibly using a X550 instead of a X540 NIC ?
I've definitely seen some, but I don't recall their model numbers. Sorry.
Probably some of these, but I see 4x10gbe as the Maximum and they are mostly Xeon-D or Atom based for that Configuration.
1U Edge Network Systems | Supermicro
Application-optimized embedded servers for space-constrained deployments. Multiple form factors supporting low-power Intel® processors for maximum efficiency
Tons of options out there with that, but you presumably want cheap. That's not happening currently. The motherboard discussed in this thread was a custom SKU built specifically for Softlayer. They likely realized that there were more economical options out there on newer generations of hardware that negated the need for the sizable capital expense of paying an OEM to design another motherboard.
If you can live with a 1U appliance that has no PCIe slots and limited room for disks, then there are options. I have some with a Xeon D-2187NT, 2 x Intel X710, 4 x Intel X722, 6 x Intel I350 (just 1GbE), and 2 x M.2 slots for example.
Better question is, why don't you just use a PCIe NIC?
If you can live with a 1U appliance that has no PCIe slots and limited room for disks, then there are options. I have some with a Xeon D-2187NT, 2 x Intel X710, 4 x Intel X722, 6 x Intel I350 (just 1GbE), and 2 x M.2 slots for example.
Better question is, why don't you just use a PCIe NIC?
Always. Good, Cheap. Pick one
.And the Supermicro X10SLL-F/X10SLM-F was one of my main favorite Platform (technically still is) so the X10SLH-LN6TF is from the same "Family".
That's why I'm surprised at the HUGE Idle Power Consumption of this X10SLH-LN6TF.
I suspect nobody checked ASPM support being Enabled and supported by all the Hardware, at least there was not mention of ASPM at all in this Thread.
That's why I believe that the Power Consumption is so high. The X10SLL-F eats I think around 25W at almost Idle. And the X10SLL-F had a finicky BIOS Bug with FADT Tables, so you need to patch those in a custom initramfs (or a pre-initramfs so to speak) to force ASPM on.
Apparently each X540-T2 can use up to 17W when both ports are in 10gbps mode
. Even when not doing anything ??? Holy ****.What's the Part Number for those ?
Right now I am using the X10SLM-F/X10SLL-F for one Virtualized (Proxmox VE) OPNsense Router, the X11SSL-F for the other Virtualized (Proxmox VE) OPNSense Router
.In Terms of NIC, the Intel X710-DA2 seems to be the best. The Intel XXV710-DA2 is another Option.
The Mellanox ConnectX-4 LX ONLY if in PCH connected Slot (NOT CPU connected PCIe Slot) for ASPM to work correctly, otherwise no
.I was just wondering if it was worth to grab one of these from the US since I am already ordering some stuff from there shipped to Europe via Jetcarrier, so they could come along as well
.But I could also grab some X710-DA2, XXV710-DA2 or Mellanox ConnectX-4 LX if I really wanted, like I have been doing for some other Builds (X10SLL-F, X11SSL-F, Fujitsu TX1320 M3, etc).
I prefer SFP+ since the Switches are so much cheaper (and less Power hungry) though.
Dell/EMC Edge 3800.
Since you are using these as routers, many more options out there. Feel free to start another thread on the topic and I can provide some recommendations.
I don't want to derail this thread, but this Edge 3800 box seems interesting - is there any information out there on storage options, OOB management, power consumption, noise, etc for this platform?