Many computers have followed the design pattern of a backplane with expansion cards, including arguably the three most influential microcomputers ever built: the Altair 8800, the Apple II and the IBM PC.

All three of those oriented the case horizontally and the cards vertically. The shape of the Apple II case was chosen by Steve Jobs for aesthetic reasons (whether or not one agrees with his taste, it's a documented historical fact that he was motivated by aesthetics). I haven't seen anything written about the criteria by which the shape and orientation of the Altair case were chosen.

By contrast, modern tower cases are vertical and orient the cards horizontally. The latter doesn't have to be so. A tower could be vertical and also orient the cards vertically, just placing the long axis vertical as opposed to the earlier machines that placed the cards short axis vertical. I would expect that to be better for heat dissipation, because it would allow natural convection to work better: hot air could rise between the cards.

On the other hand, modern tower cases came in after all PCs had started adding cooling fans to cope with the increased heat dissipation of high-frequency chips and hard disks. Maybe once you have fans, there isn't a reason to care as much about natural convection? The trend is suggestive, but doesn't prove anything. It's possible the orientations were chosen for other reasons.

Was natural convection ever a consideration in the orientation of the expansion cards, in the design of any computer following the backplane plus card pattern? Is there anything written down anywhere about the criteria that were used?

  • I guess what you're asking for is not convection but air flow. – Raffzahn Nov 8 '20 at 10:50
  • 2
    @Raffzahn: Are you sure you're not confusing convection with conduction? Convection is transfer of heat by air flow (or by some other fluid flow). – Ilmari Karonen Nov 9 '20 at 0:19
  • 1
    @IlmariKaronen Right, convection is heat transfer by air movement. In this case it's important to keep natural convection (driven thru temperature difference) and forced convection - driven by an external force like a ventilator. Usage of the word convection is a bit fuzzy in English. – Raffzahn Nov 9 '20 at 0:28
  • 2
    @Raffzahn: Right. I was just confused by some of your wording above and in your answer below, where you seem to be contrasting convection and flow as if they were two different things (and not mentioning conduction at all). I'm also still not sure what you mean by your claim that convection is "independent of orientation"; for "natural" gravity-driven convection, that's definitely not the case, and indeed it's exactly what the OP seems to be asking about. – Ilmari Karonen Nov 9 '20 at 4:25
  • 1
    @Raffzahn: Convection always happens when there's a temperature difference, a fluid, and a path for the fluid to flow. The more obstructed the path is, in the direction the fluid wants to flow under gravity (i.e. usually vertically), the less efficient convection will be at transporting heat. For cooling a flat plate (like a circuit board) by natural convection, vertical orientation is more efficient than horizontal. – Ilmari Karonen Nov 9 '20 at 14:06

The main means of thermal dissipation are convection and radiation (and flow). Convection is used always and independent of orientation. Board setup of the era have been made in any direction.

Putting boards close and in parallel is, independent of orientation, not a great idea at first, as they heat each other (by radiation). This means for thermal dissipation the whole stack has to be treated as one, independent of orientation or packaging. Similar any closed case will only allow to equalisation within by means of convection/radiation/flow.

This only changes if there's an additional air flow across the stack/between the boards. This can for example be encouraged by an orientation that lets air freely come in from one side and leave (again freely) on the other side. I assume that's what the question is about.

To start with, typical S100 cases, like the Altair, do not really support airflow as the base board is in the way. In addition, there is usually a closed metal sheet direct above the boards.and often solid walls on either side.

In general thermal design wasn't a big issue for early machines as they combined low power consumption (in absolute terms) with huge volume, wo thermal density was quite low. Even loaded S100 machines did usually produce more heat in the power supply than along the boards (*1).

Of course situation was different in minis and mainframes, but they were as well designed for airflow. Here Mounting was usually on a backplane leaving room for a straight airflow across the boards - which usually was enforced by ventilators.

The Apple II did rely only on convection thru the plastic covers, as there was no real airflow design, despite the slots on the side and openings on the back. While this was fine for the low heat production of the basic machine even including a few cards, it only became a problem when really stacked, clogged with cables and operated in an hot environment (like with a CRT stacked on top blocking heat flow or even adding). Here solutions like Kensington System Saver added airflow. In relation to the question it's notable that the System Saver did simply suck out air on one side, square to the card orientation. This worked fine, as the effect of exchanging the air outclasses everything that relies on convection.

Still, it wasn't as much of a problem as Kensignton adds may suggest, as the as the original Apple II PS was rated at 38 Watt (*2). Still keeping over all heat production rather low. Of course, as soon as there was a hard disk to be build in, air flow became a must - in addition to a bigger PS anyway :))

The IBM PC in turn was already designed with that experience in mind. Here the power supply's ventilator sucks air out of the case - replenished via inlets on the opposite side as the PS was located in the back to the right and inlets were in the front to the left - in fact, right were (full length) expansion board would end up.

Still it's important to keep in mind that the IBM-PC was extreme low power from today's POV. The original PSU was rated at 63 Watt. Nop, no zero missing or so, it was that low. Combined with a case that large as the original PC, the design was more than sufficient to keep everything within limits.

Modern PCs with a much higher over all power dissipation have an improved airflow design with the power supplies ventilator forcing air out of the case, thus sucking in cold (room temperature) air via designated ventilating openings or simply any other opening (*3,*4). While focusing on CPU and chipset, this is more than sufficient to get all heat, including additional bards out, despite of board orientation.

The issue with added ventilators to certain chips/boards isn't the over all heat production, but local distribution (*5).

Long story short: Early systems didn't ned much air flow design and board orientation was more due case construction and handling than heat management.

*1 - Yes, that includes the (comparably) inefficient regulators on each board.

*2 - There is a real nice (as usual) analysis of the Apple II PSU as well the IBM-PC's by Ken Shirriff.

*3 - Who doesn't remember some PC clogging up floppy drives with lint as their case was missing dedicated air inlets.

*4 - Related were home build machines were users added ventilators pressing fresh air into the case, which of course increased inside pressure and exhaust true every possible opening, again including floppy drives, depositing even more dirt there.

*5 - Given,there are some high end cards that do have their complete own air flow, sucking in fresh air thru a desicated slot cover.

  • 1
    Your story of the S100 (Altair) is incomplete. The power supply delivered an unregulated overly high voltage that had to be regulated on each addin board. Those regulators were typically attached to huge heat sinks and dissipated the majority of heat on the board. Those heat sinks relied on some airflow to keep them cool. – Mark Ransom Nov 9 '20 at 17:26
  • 1
    @MarkRansom No doubt. But the question is 'was it designed that way' and the answer is no., as the backplane below the boards isn't made to support such a flow. In addition MITS had no say were card manufacturer placed the regulators. – Raffzahn Nov 10 '20 at 10:23
  • 1
    You have it backwards. Card manufacturers put the regulator on the board location that had the best airflow. – Mark Ransom Nov 10 '20 at 17:47

Simply put, for most of the PC's history, either the components were roughly uniform in their power consumption (with the major offenders typically being the 12V motors in disk drives, rather than anything solid-state), or the CPU was the major power consumer and thus source of heat. Only fairly recently has one particular type of expansion card, namely the GPU, become so power hungry that it needed its own thermal management strategy.

When power dissipation was low and uniform, convective airflow was sufficient to cool all of the components regardless of orientation, with cool air filtering in through whichever vents happened to be low on the case, and warm air filtering out through high vents. The main exception might have been the PSU, which often had an exhaust fan purely to cool itself, with little effect on airflow elsewhere in the case.

The move from Baby-AT to the ATX series specification was prompted by 6th and 7th generation x86 CPUs (Pentium II/III/IV, K6, Athlon) routinely dissipating tens of watts, and thus needing a substantial heatsink that was clear of physical interference from expansion cards. The ATX PSU included a relocated fan which explicitly drew air from the vicinity of the CPU, and the ATX case included a space for an exhaust fan from this area as well.

With so much suction towards the CPU area, air movement within the case is towards the CPU in both desktop and tower orientations. This is roughly equivalent, from expansion cards' point of view, to being fitted in a tower case with the cards horizontal. On the other hand, the air movement is generally faster than with natural convection.

When ATX was introduced, graphics card were still relatively limited in their capabilities and power consumption, with only a few outliers such as the last 3dfx Voodoo models including a small fan to bolster the performance of a small heatsink. Many other cards had no heatsink at all, such as most of ATI's cards before the Radeon series. This illustrates that thermal considerations were not a major concern when it came to expansion cards in general.

Modern ATX cases often break from the original ATX spec, in that they put the PSU at the opposite end of the motherboard than the CPU. In these designs, all of the airflow is actively driven and there is no reliance on convection at all.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.