I'm planning on operating a couple of 1U servers "on their sides" (i.e., tall, deep and VERY skinny). Of course, this is normal to their expected operating orientation. I figure the disks won't mind being operated in that orientation as I have shelfs with (the same) drives oriented that way. 1U servers tend to have a bank of redundant fans located behind the drive array to produce a flow through the drives and across the CPUs on its way out the vents in the rear. The fins of the CPU heatsinks are oriented to exploit this flow. But, the heatsinks also were designed with the expectation (?) that the fins would be upright (not sideways). Likewise, the CPU die expect the heatsinks to be "above". And, all the RAM to be in the same horizontal plane, etc. Of course, you can operate a CPU (or RAM, etc.) in any orientation. There are plenty of towers, etc. that are designed that way. And, machines that can be operated in either orientation. But, those are likely designed with that in mind! My question is: how tight is the thermal margin likely to be in these "already cramped" designs? Or, will my cold aisle likely more than compensate for any stresses? I suppose I could look at the temperatures that the two CPUs report and see if there is a noticeable difference...
Equipment orientation
Started by ●March 27, 2023
Reply by ●March 27, 20232023-03-27
On 3/26/2023 8:51 PM, Don Y wrote:> I'm planning on operating a couple of 1U servers "on their sides" > (i.e., tall, deep and VERY skinny). > > Of course, this is normal to their expected operating orientation. > > I figure the disks won't mind being operated in that orientation > as I have shelfs with (the same) drives oriented that way. > > 1U servers tend to have a bank of redundant fans located behind > the drive array to produce a flow through the drives and across > the CPUs on its way out the vents in the rear. The fins of > the CPU heatsinks are oriented to exploit this flow. > > But, the heatsinks also were designed with the expectation (?) > that the fins would be upright (not sideways). Likewise, the > CPU die expect the heatsinks to be "above". And, all the RAM > to be in the same horizontal plane, etc. > > Of course, you can operate a CPU (or RAM, etc.) in any orientation. > There are plenty of towers, etc. that are designed that way. > And, machines that can be operated in either orientation. > > But, those are likely designed with that in mind! > > My question is: how tight is the thermal margin likely to > be in these "already cramped" designs? Or, will my cold aisle > likely more than compensate for any stresses? > > I suppose I could look at the temperatures that the two CPUs > report and see if there is a noticeable difference...Verdict (friends who run data centers) is that I'm not going to hammer on this hard enough to make a difference. Apparently, they also make mounts for these in different orientations. So, we'll see...
Reply by ●March 27, 20232023-03-27
On Sunday, March 26, 2023 at 8:51:59 PM UTC-7, Don Y wrote:> I'm planning on operating a couple of 1U servers "on their sides" > (i.e., tall, deep and VERY skinny). > > Of course, this is normal to their expected operating orientation. > > I figure the disks won't mind being operated in that orientation > as I have shelfs with (the same) drives oriented that way. > > 1U servers tend to have a bank of redundant fans located behind > the drive array to produce a flow through the drives and across > the CPUs on its way out the vents in the rear. The fins of > the CPU heatsinks are oriented to exploit this flow. > > But, the heatsinks also were designed with the expectation (?) > that the fins would be upright (not sideways).There's no need for thermal-gradient-driven convection in most server designs; they're noisy because the fans are intended to deliver turbulent air for effective cooling. Just, don't let the fans stop, or block the vents, and observe the altitude specification.
Reply by ●March 27, 20232023-03-27
On 27/03/23 14:51, Don Y wrote:> I'm planning on operating a couple of 1U servers "on their sides" > (i.e., tall, deep and VERY skinny). > > Of course, this is normal to their expected operating orientation. > > I figure the disks won't mind being operated in that orientation > as I have shelfs with (the same) drives oriented that way. > > 1U servers tend to have a bank of redundant fans located behind > the drive array to produce a flow through the drives and across > the CPUs on its way out the vents in the rear. The fins of > the CPU heatsinks are oriented to exploit this flow. > > But, the heatsinks also were designed with the expectation (?) > that the fins would be upright (not sideways). Likewise, the > CPU die expect the heatsinks to be "above". And, all the RAM > to be in the same horizontal plane, etc. > > Of course, you can operate a CPU (or RAM, etc.) in any orientation. > There are plenty of towers, etc. that are designed that way. > And, machines that can be operated in either orientation. > > But, those are likely designed with that in mind! > > My question is: how tight is the thermal margin likely to > be in these "already cramped" designs? Or, will my cold aisle > likely more than compensate for any stresses? > > I suppose I could look at the temperatures that the two CPUs > report and see if there is a noticeable difference... >Heat doesn't recognise up and down. The forced-air is fast enough to void any effect of convection.
Reply by ●March 27, 20232023-03-27
On 3/27/2023 4:41 PM, whit3rd wrote:> On Sunday, March 26, 2023 at 8:51:59 PM UTC-7, Don Y wrote: >> I'm planning on operating a couple of 1U servers "on their sides" >> (i.e., tall, deep and VERY skinny). >> >> Of course, this is normal to their expected operating orientation. >> >> I figure the disks won't mind being operated in that orientation >> as I have shelfs with (the same) drives oriented that way. >> >> 1U servers tend to have a bank of redundant fans located behind >> the drive array to produce a flow through the drives and across >> the CPUs on its way out the vents in the rear. The fins of >> the CPU heatsinks are oriented to exploit this flow. >> >> But, the heatsinks also were designed with the expectation (?) >> that the fins would be upright (not sideways). > > There's no need for thermal-gradient-driven convection in > most server designs; they're noisy because the fans are > intended to deliver turbulent air for effective cooling.Fans are all individually and collectively redundant. It would be REALLY hard for them all to fail (without other more noticeable things also failing).> Just, don't let the fans stop, or block the vents, and > observe the altitude specification.No danger wrt altitude; we're only 2k' above sea level. As a 1U device, the only real vents are front (cold aisle) and rear (hot). Given the need to access the various connectors, power supplies, etc. at the rear, it's not really possible to push it back up against anything (the wall?). And, I'm not likely to drop a *pillow* back there! :> Front has a protective face plate that ensures some airflow even if you stood against it (this is at floor level so not gonna happen). And, I need to be able to access the disks, controls, diagnostic panel, etc. so I think it's safe. Thankfully, this one doesn't bitch (audibly) when a redundant supply "fails" (i.e. isn't plugged in!). I've one machine that screams bloody murder until it sees output from both supplies! (no, I don't want to have to dedicate TWO outlets to each box!)
Reply by ●March 27, 20232023-03-27
On 3/27/2023 4:49 PM, Clifford Heath wrote:> On 27/03/23 14:51, Don Y wrote: >> I'm planning on operating a couple of 1U servers "on their sides" >> (i.e., tall, deep and VERY skinny). >> >> Of course, this is normal to their expected operating orientation. >> >> I figure the disks won't mind being operated in that orientation >> as I have shelfs with (the same) drives oriented that way. >> >> 1U servers tend to have a bank of redundant fans located behind >> the drive array to produce a flow through the drives and across >> the CPUs on its way out the vents in the rear. The fins of >> the CPU heatsinks are oriented to exploit this flow. >> >> But, the heatsinks also were designed with the expectation (?) >> that the fins would be upright (not sideways). Likewise, the >> CPU die expect the heatsinks to be "above". And, all the RAM >> to be in the same horizontal plane, etc. >> >> Of course, you can operate a CPU (or RAM, etc.) in any orientation. >> There are plenty of towers, etc. that are designed that way. >> And, machines that can be operated in either orientation. >> >> But, those are likely designed with that in mind! >> >> My question is: how tight is the thermal margin likely to >> be in these "already cramped" designs? Or, will my cold aisle >> likely more than compensate for any stresses? >> >> I suppose I could look at the temperatures that the two CPUs >> report and see if there is a noticeable difference... >> > > Heat doesn't recognise up and down. The forced-air is fast enough to void any > effect of convection.I was surprised that they actually sell mounts that let you mount the server "front (intake) up". Of course, this is much more user-friendly than "back (exhaust) up". I will still check the temperatures of the two CPUs to see if any repeatable difference between the left & right devices.
Reply by ●March 28, 20232023-03-28
On Mon, 27 Mar 2023 15:11:07 -0700, Don Y <blockedofcourse@foo.invalid> wrote:>On 3/26/2023 8:51 PM, Don Y wrote: >> I'm planning on operating a couple of 1U servers "on their sides" >> (i.e., tall, deep and VERY skinny). >> >> Of course, this is normal to their expected operating orientation. >> >> I figure the disks won't mind being operated in that orientation >> as I have shelfs with (the same) drives oriented that way. >> >> 1U servers tend to have a bank of redundant fans located behind >> the drive array to produce a flow through the drives and across >> the CPUs on its way out the vents in the rear.� The fins of >> the CPU heatsinks are oriented to exploit this flow. >> >> But, the heatsinks also were designed with the expectation (?) >> that the fins would be upright (not sideways).� Likewise, the >> CPU die expect the heatsinks to be "above".� And, all the RAM >> to be in the same horizontal plane, etc. >> >> Of course, you can operate a CPU (or RAM, etc.) in any orientation. >> There are plenty of towers, etc. that are designed that way. >> And, machines that can be operated in either orientation. >> >> But, those are likely designed with that in mind! >> >> My question is:� how tight is the thermal margin likely to >> be in these "already cramped" designs?� Or, will my cold aisle >> likely more than compensate for any stresses? >> >> I suppose I could look at the temperatures that the two CPUs >> report and see if there is a noticeable difference... > >Verdict (friends who run data centers) is that I'm not going to >hammer on this hard enough to make a difference. Apparently, >they also make mounts for these in different orientations. > >So, we'll see... >You should follow bottom to top airflow, in a tall structure. If fans are unevenly locarted, use bottom to top as assist. RL
Reply by ●March 28, 20232023-03-28
On 3/28/2023 6:34 AM, legg wrote:> On Mon, 27 Mar 2023 15:11:07 -0700, Don Y > <blockedofcourse@foo.invalid> wrote: > >> On 3/26/2023 8:51 PM, Don Y wrote: >>> I'm planning on operating a couple of 1U servers "on their sides" >>> (i.e., tall, deep and VERY skinny). >>> >>> Of course, this is normal to their expected operating orientation. >>> >>> I figure the disks won't mind being operated in that orientation >>> as I have shelfs with (the same) drives oriented that way. >>> >>> 1U servers tend to have a bank of redundant fans located behind >>> the drive array to produce a flow through the drives and across >>> the CPUs on its way out the vents in the rear. The fins of >>> the CPU heatsinks are oriented to exploit this flow. >>> >>> But, the heatsinks also were designed with the expectation (?) >>> that the fins would be upright (not sideways). Likewise, the >>> CPU die expect the heatsinks to be "above". And, all the RAM >>> to be in the same horizontal plane, etc. >>> >>> Of course, you can operate a CPU (or RAM, etc.) in any orientation. >>> There are plenty of towers, etc. that are designed that way. >>> And, machines that can be operated in either orientation. >>> >>> But, those are likely designed with that in mind! >>> >>> My question is: how tight is the thermal margin likely to >>> be in these "already cramped" designs? Or, will my cold aisle >>> likely more than compensate for any stresses? >>> >>> I suppose I could look at the temperatures that the two CPUs >>> report and see if there is a noticeable difference... >> >> Verdict (friends who run data centers) is that I'm not going to >> hammer on this hard enough to make a difference. Apparently, >> they also make mounts for these in different orientations. >> >> So, we'll see... >> > > You should follow bottom to top airflow, in a tall structure. > > If fans are unevenly locarted, use bottom to top as assist.That was what I would have assumed. But, the mounts that are designed for vertical orientation (bottom-top, top-bottom... as opposed to my "sideways" orientation) seem all to favor the front of the server (cold aisle) at the top -- likely for ease of human interaction (imagine bending down and craning to look UP at the front panel; trying to insert an optical medium into a tray that is opening *downward*, etc.) I assume that this is because the volume and rate of airflow through the enclosure is >> what natural convection can counter. And, if that was the case, then (in my sideways orientation) the same argument would apply to the CPU located "farther from the floor" vs. the CPU *closest* to the floor. I.e., there may be some temperature differences but nothing that would let one CPU be in its SOA while the other was borderline. The fact that such servers are typically (by an overwhelming percentage) deployed in the "flat" orientation led me to wonder if this assumption was baked into the design... [I don't want to site a rack in the office]
Reply by ●March 28, 20232023-03-28
On 2023-03-28 10:14, Don Y wrote:> On 3/28/2023 6:34 AM, legg wrote: >> On Mon, 27 Mar 2023 15:11:07 -0700, Don Y >> <blockedofcourse@foo.invalid> wrote: >> >>> On 3/26/2023 8:51 PM, Don Y wrote: >>>> I'm planning on operating a couple of 1U servers "on their >>>> sides" (i.e., tall, deep and VERY skinny). >>>> >>>> Of course, this is normal to their expected operating >>>> orientation. >>>> >>>> I figure the disks won't mind being operated in that >>>> orientation as I have shelfs with (the same) drives oriented >>>> that way. >>>> >>>> 1U servers tend to have a bank of redundant fans located >>>> behind the drive array to produce a flow through the drives and >>>> across the CPUs on its way out the vents in the rear. The fins >>>> of the CPU heatsinks are oriented to exploit this flow. >>>> >>>> But, the heatsinks also were designed with the expectation (?) >>>> that the fins would be upright (not sideways). Likewise, the >>>> CPU die expect the heatsinks to be "above". And, all the RAM >>>> to be in the same horizontal plane, etc. >>>> >>>> Of course, you can operate a CPU (or RAM, etc.) in any >>>> orientation. There are plenty of towers, etc. that are designed >>>> that way. And, machines that can be operated in either >>>> orientation. >>>> >>>> But, those are likely designed with that in mind! >>>> >>>> My question is: how tight is the thermal margin likely to be >>>> in these "already cramped" designs? Or, will my cold aisle >>>> likely more than compensate for any stresses? >>>> >>>> I suppose I could look at the temperatures that the two CPUs >>>> report and see if there is a noticeable difference... >>> >>> Verdict (friends who run data centers) is that I'm not going to >>> hammer on this hard enough to make a difference. Apparently, >>> they also make mounts for these in different orientations. >>> >>> So, we'll see... >>> >> >> You should follow bottom to top airflow, in a tall structure. >> >> If fans are unevenly locarted, use bottom to top as assist. > > That was what I would have assumed. But, the mounts that are > designed for vertical orientation (bottom-top, top-bottom... as > opposed to my "sideways" orientation) seem all to favor the front of > the server (cold aisle) at the top -- likely for ease of human > interaction (imagine bending down and craning to look UP at the front > panel; trying to insert an optical medium into a tray that is opening > *downward*, etc.) > > I assume that this is because the volume and rate of airflow through > the enclosure is >> what natural convection can counter. > > And, if that was the case, then (in my sideways orientation) the same > argument would apply to the CPU located "farther from the floor" vs. > the CPU *closest* to the floor. I.e., there may be some temperature > differences but nothing that would let one CPU be in its SOA while > the other was borderline. > > The fact that such servers are typically (by an overwhelming > percentage) deployed in the "flat" orientation led me to wonder if > this assumption was baked into the design... > > [I don't want to site a rack in the office] >The natural-convection performance of the pizza box format is terrible, which (along with the small diameter fans) is why they sound like jet engines. Oriented vertically, you could at least get a bit of chimney effect along the flat sides. Cheers Phil Hobbs -- Dr Philip C D Hobbs Principal Consultant ElectroOptical Innovations LLC / Hobbs ElectroOptics Optics, Electro-optics, Photonics, Analog Electronics Briarcliff Manor NY 10510 http://electrooptical.net http://hobbs-eo.com
Reply by ●March 28, 20232023-03-28
On 3/28/2023 8:29 AM, Phil Hobbs wrote:> The natural-convection performance of the pizza box format is terrible, > which (along with the small diameter fans) is why they sound like jet > engines.Yes. So, you have to wonder why they haven't been able to quiet things down. Do they figure server rooms are, already, noisey so "what's the problem with a little more?" Does quieting the fans (only some of which can be done by mechanical redesign of the fans) end up having too large an impact on airflow/cooling capacity? Or, are they just playing things safe (more air is better than less air)? Then, you have to look at the margins. On the one hand, budgets for data centers tend to be flusher than home/personal machines. So, they could possibly afford to build in more thermal margin (than, for example, a cheap "home PC") OTOH, they might reason that data center staff will provide more scheduled maintenance than would typically be seen in a home product (how often do folks clean their filters, heat sinks, etc.)? Finally, how much does the mechanical configuration and orientation *actually* play into the thermal performance of the device. <shrug> No $$ involved and I can always replace with other servers so this will be a learning experience...> Oriented vertically, you could at least get a bit of chimney effect along the > flat sides.
