Back in the day, workstations used to get impressively big. Take a look at this — sure, a modern laptop has more computing power, but you don't see personal computers weighing eighty pounds these days.

I'm interested in power consumption and heat dissipation (sadly not listed for the above machine). In particular, did any workstation ever go over a kilowatt of thermal design power? For this purpose, I'm classifying a computer as a workstation if it's quiet enough that you'd want it under your desk, and a server if it's noisy enough that you'd want it in a separate room. So another way of phrasing the question is, did anyone ever manage to dissipate more than a kilowatt with air cooling through a tower form factor, and still keep it quiet?

  • 3
    You definitely wouldn’t want an AlphaServer (the system you linked to) under your desk! It’s not a workstation. It’s worth noting that many workstations were quite noisy back in the eighties and nineties... Dec 17, 2017 at 7:27
  • 2
    Your question wanders into the ambiguous territory of where the distinction between a workstation and a minicomputer lies. A lot of minicomputers consumed more than 1kW (A quick google suggests that various versions of PDP-11 could consume between 1.2kW and 6kW, for instance). And while they were generally designed and used for multi-user applications, some were used for single user purposes, so could plausibly be called workstations... see wolfgang-houben.de/faqpdp11.htm and search for "MicroPDP" for example.
    – Jules
    Dec 17, 2017 at 10:35
  • 3
    What do you mean by "did". Here is a 1.5 kW gaming PSU. Here is a modern 2 kW workstation (Gibberish in Swedish, sorry).
    – pipe
    Dec 17, 2017 at 11:14
  • 2
    @pipe - there's a difference between PSU maximum output and thermal dissipation. It's common to allow relatively large amounts of spare capacity on the basis that PSUs produce better regulated output when they're not approaching their limits. That 2kW workstation, for example, has 2 CPUs @ 135W each + up to 4x GPU boards, which tend to max out at 150W each. Allowing another 100W for other components, that's just short of 1kW TDP.
    – Jules
    Dec 17, 2017 at 12:31
  • 1
    (... that said, I recall that NVidia produce or produced a system intended to allow 8 GPU boards to be installed, which would definitely exceed 1kW; I'm not convinced it was ever intended for use as a workstation, though, but rather as part of a high performance cluster)
    – Jules
    Dec 17, 2017 at 12:41

2 Answers 2


The SGI Onyx and related machines were made in a "deskside" version.

Let us look at the specifications from this manual ...

https://techpubs.jurassic.nl/library/manuals/1000/007-1733-070/pdf/007-1733-070.pdf "POWER Onyx™ and Onyx Deskside Owner’s Guide"

8100BTU/h max Thermal dissipation for the whole system. That is 2.4kW, concurrent with the rated power for the version equipped with the "Infinite Reality" graphics subsystem.

Given these were professional use machines typically used for rendering animated video using the complex graphics hardware, it is unlikely they were not designed to run under full load for extended periods of time. It is also unlikely that the power supplies were overbuilt "just because", since their rating brings the inconvenience of being beyond what a standard 120V wall socket can deliver.

Noise is rated at 60dB.

https://www.ifixit.com/Teardown/SiliconGraphics+Onyx+Teardown/12546 .. these huge boxes were filled with mostly electronics, not air :)

  • 1
    "their rating brings the inconvenience of being beyond what a standard 120V wall socket can deliver" -- yes, but well within the capabilities of a 220/240v supply, which I believe is usually available in commercial premises even in the US? "Noise is rated at 60dB." which is a little quieter than the fan heater I use to warm my office up ... and a little more effective at heating. :)
    – Jules
    Dec 17, 2017 at 19:06
  • Not by much. A computer turns all the power into heat, excepting power used to send communications signals (which will the heat someone else's office), energy consumed to make magnetic changes on media, energy used to power indicator lights or sounds (most of the light end ends up as heat too), energy to move air (will end up as heat somewhere else too). Dec 17, 2017 at 19:14
  • 4
    Was there a 240V socket usually available in an office used for animation work in the 1990s, or would it have meant "call a sparky and maybe have the socket 3 weeks later"? Dec 17, 2017 at 19:16
  • Most US and Canada commercial or industrial buildings are wired for "three-phase" power, in which 120V and 208V, but not 240V, can be easily made available. 208 uses the same outlets as 240, but are only installed where requested. They're common enough in laboratory and mechanical spaces; occasionally they can be spotted in corridors for use by floor polishing machines. Otherwise you'd need to have one installed. But equally easy would be a 20-amp 120-volt circuit, from which I believe 1.9 kW sustained load could be supported.
    – CCTO
    Feb 5, 2018 at 18:36
  • @CCTO my point was, anything that can't be relied on being available in a business building and which will need to be professionally installed.... Feb 5, 2018 at 20:20

This question doesn't make sense in its actual state. 'Thermal design power' is nonsense term when taking about a computer (in fact it's even more when it's about CPUs). Computers are neither designed to do mechanical work, light rooms or alike. There is no physical computing power - the only physical output is heat and a little bit of HF radiation.

So the only usefull question I to be found would be 'have there been desktopcomputers with more than 1 kW power rating'.

And the answer is Yes, quite a lot.

Already the IMSAI PS was rated at 500 Watt primary, and it didn't take long for third party frames to offer PS with 1 kW primry and more. And primary is a key issue here. Before the advent of switching, the power supply itself was usually the most critical part. In case of above IMSAI, up to 2/3rd of the primary power was wasted to heat up the PS. Removing that substantial heat was also needed to protect the PS, not so much the CPU.

Watercooling in computers was also not introduced to keep computing chips cool, but their transformers and power electronics. For example, a SIEMENS 7.760 X4 CPU, an upper end /370ish machine of ~1980 had a watercooled power supply, but air cooled electronics.

Even with a switching PS, like the Apple II pionered, the PS stayed the main heat source within the case. Kensington made good money with their system saver series of add on fans. It took many years until the computing chips itself got miniaturized enough to need even a simple head spreader for the little power they transformed.

  • Given the >70% efficiency typically quoted for PSMPS, and the fact that, as you mentioned, practically all power consumed by a computer is turned into heat, considering the PSU the main heat source (by amount of heat, not heat concentration!) is illogical. And BTW: ELRAD magazine once published a linear PSU design claiming PSMPS-level efficiency for external drives. Dec 20, 2017 at 17:55
  • @rackandboneman Well, efficency at 70% or above was wishful tinking back then. The Apple II PS did manage less than 60% at full load. Without a date and reference he claim about linear vs. switching doesn't make any sense. Then again, specialized linear designs could go well up to 50% when used at peak effiency - and switchign designs of the 70 often did't do much. So yeah, comparing a race donkey with a Clydesdale does make a lot of sense ...
    – Raffzahn
    Dec 20, 2017 at 20:26
  • The Apple II is considered a very atypical SMPS design, no? HP quotes 70% for the PSU in the 2640A terminal, from 1975... 2100A from 1971 is 800W input rated at 500W output, which still makes >60% ... Dec 20, 2017 at 22:11

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .