Manufacturer's measurements for GPUs are infamously bad. Measurements could be from the bottom of the PCI bracket, the bottom of the PCI-E card-edge, the bottom of the PCB excluding the card-edge, bottom of the shroud, to the top of the PCI-E bracket, top of the shroud (with or without shroud manufacturing tolerance), top of the PCB, top of the PCI-E power connectors on-board, etc, or just not correspond to any physical dimension at all. This is obviously a real pain for everyone trying to figure out if a given card will fit in a different case.
I don't think this is out of maliciousness or 'spec padding', but simply because everyone needs to get some sort of dimensions onto their spec sheet. Nobody has any sort of guide or standard to refer to, so every whoever ends up with the job of measuring the card at each manufacturer measures what the last guy who did the measurements told them to measure.
If a guide can be provided that is easy to follow, cheap to implement, and has some tangible reward for doing so, then there's a good change we can persuade manufacturers to all do the same measurements - and provide them in the same format (and nomenclature) - purely out of self-interest.
What dimensions are actually important to measure? What, if any, of these are actually important to know, and which are redundant (inferable from other measurements) or simply too niche to try and make people bother with.
- Bounding box dimensions of entire card with shroud (including or excluding PCI bracket)
- Bounding box dimensions of shroudless PCB with PCI bracket (useful for aftermarket cooling)
- Height of the PCI-E power connectors above the PCI-E card-edge (for determining total clearance needed for cards when the power connectors are plugged in, accommodating cards with PCBs that extend above the level of the power connectors
- Any others?
What nomenclature is preferable? Which part of the card is the 'front', what is the 'top', etc?
How should one go about measuring the card? The three most obvious ones would be:
- Mechanical measurements. Either long-jaw callipers or some sort of sliding-parallel-bar jig. Most accurate, and most GPU manufacturers probably have a set of callipers available. The jig would be preferable to get a good bounding-box measurement (and to ease speed of measurements by having mixed mounting and measurement points), but has the obvious downside of being a material cost,
- Digital measurement from scanned images. The capital outlay for a flatbed scanner is minimal, and known measurements (e.g. PCI-E card-edge length) can be used for calibration. There's also the bonus that aligned and calibrated images of the card are created by default, so these could be used to make a very nicely formatted database of card images AND allow for future measurements by end users. The downside is that some oversized cards may not fit on the platen, getting cards to lie flat with unusually shaped coolers (and for side-on and edge-on images), and cards may not scan in focus due to being too far from the platen.
- Digital measurements from photographed images. This is mostly the same as with scanned images. Setup is more complicated with the need for properly measured distance from camera to card, lighting, known lens parameters, and problems with optical distortion (minimised with a very long shot and a zoom lens, but this takes up a lot of room!). The advantage is accommodating any size of card, and having the process done when the card is photographed for product shots (with most of the equipment needed already being present).
And the big one; what's in it for manufacturers to do this? Having a process for measuring cards may make things marginally easier for them procedurally (you can just point a guy at a document as say "do this"), but it also might end up being more work overall to 'do it right'. A 'stamp of approval' could be conferred making for an extra addition to press releases/product pages/box art, but that might need too much infrastructure to manage validation. The case could also be made that in giving consumers confidence that a given card will fit in their case (confidence they currently do not have) manufacturers may see a slight increase in sales.
I don't think this is out of maliciousness or 'spec padding', but simply because everyone needs to get some sort of dimensions onto their spec sheet. Nobody has any sort of guide or standard to refer to, so every whoever ends up with the job of measuring the card at each manufacturer measures what the last guy who did the measurements told them to measure.
If a guide can be provided that is easy to follow, cheap to implement, and has some tangible reward for doing so, then there's a good change we can persuade manufacturers to all do the same measurements - and provide them in the same format (and nomenclature) - purely out of self-interest.
What dimensions are actually important to measure? What, if any, of these are actually important to know, and which are redundant (inferable from other measurements) or simply too niche to try and make people bother with.
- Bounding box dimensions of entire card with shroud (including or excluding PCI bracket)
- Bounding box dimensions of shroudless PCB with PCI bracket (useful for aftermarket cooling)
- Height of the PCI-E power connectors above the PCI-E card-edge (for determining total clearance needed for cards when the power connectors are plugged in, accommodating cards with PCBs that extend above the level of the power connectors
- Any others?
What nomenclature is preferable? Which part of the card is the 'front', what is the 'top', etc?
How should one go about measuring the card? The three most obvious ones would be:
- Mechanical measurements. Either long-jaw callipers or some sort of sliding-parallel-bar jig. Most accurate, and most GPU manufacturers probably have a set of callipers available. The jig would be preferable to get a good bounding-box measurement (and to ease speed of measurements by having mixed mounting and measurement points), but has the obvious downside of being a material cost,
- Digital measurement from scanned images. The capital outlay for a flatbed scanner is minimal, and known measurements (e.g. PCI-E card-edge length) can be used for calibration. There's also the bonus that aligned and calibrated images of the card are created by default, so these could be used to make a very nicely formatted database of card images AND allow for future measurements by end users. The downside is that some oversized cards may not fit on the platen, getting cards to lie flat with unusually shaped coolers (and for side-on and edge-on images), and cards may not scan in focus due to being too far from the platen.
- Digital measurements from photographed images. This is mostly the same as with scanned images. Setup is more complicated with the need for properly measured distance from camera to card, lighting, known lens parameters, and problems with optical distortion (minimised with a very long shot and a zoom lens, but this takes up a lot of room!). The advantage is accommodating any size of card, and having the process done when the card is photographed for product shots (with most of the equipment needed already being present).
And the big one; what's in it for manufacturers to do this? Having a process for measuring cards may make things marginally easier for them procedurally (you can just point a guy at a document as say "do this"), but it also might end up being more work overall to 'do it right'. A 'stamp of approval' could be conferred making for an extra addition to press releases/product pages/box art, but that might need too much infrastructure to manage validation. The case could also be made that in giving consumers confidence that a given card will fit in their case (confidence they currently do not have) manufacturers may see a slight increase in sales.
