Your Case Publishes One Number. Your GPU Has Three.

I have the published dimensions of 300 graphics cards in one file. Three numbers on every card. Most case listings only ever quote one of them.

Short version for the skimmers. SFF case GPU clearance is three measurements, not one: length, thickness and height. Length is the number everyone prints. Thickness usually arrives as a slot count, which is a rounded marketing figure rather than a measurement. Height is the one almost nobody publishes, and in a small case it is the measurement most likely to stop the build. Check all three against the card’s own vendor page before you buy a case or start a print.

The three numbers, and who publishes which

A graphics card occupies a box. The box has a length along the slot, a height measured up from the slot, and a thickness across the slots. Every card has all three. Every case has a limit for all three.

The problem is that the two sides of that comparison are written by different people using different words.

Case listings almost always publish a maximum GPU length in millimetres. Good. That number is directly comparable. After that it gets loose. Thickness is usually given as a slot count. Height is frequently not given at all, because in a full size tower it never matters, and the people writing the spec sheet learned the format on full size towers.

In a case under about six litres, all three matter, and they do not fail in the order you would expect.

Why a slot count is not a measurement

A slot is a bucket. My 300 cards have 58 distinct published thickness values, ranging from 15 mm to 62.4 mm. Collapsing 58 values into four labels throws away most of the information, and the direction of the rounding is not in your favour.

217 of the 300 cards sit between 38 mm and 45 mm thick. That is a 7 mm spread packed into the region where most two slot coolers land. I should be straight about the limit of my own data here: the dataset records millimetres, not slot labels, so the mapping between the two is convention rather than something I measured. In a case designed with 2 mm of clearance, 7 mm is the whole argument.

This is the same problem I ran into while designing the case, and it is why I built the database in the first place. I went through it at length in the SFF GPU dimensions database post, which has the full table and the per-card sources.

It gets worse at the top end. The thickest card in the set is 62.4 mm. The thinnest is 15 mm. Both have a slot number attached to them, and neither slot number tells you what to cut.

So the rule I use is simple. If a case publishes a thickness limit in millimetres, trust it and compare it to the vendor’s millimetres. If a case publishes only a slot count, treat that as a hint and go find the real number.

Height is the dimension that kills small builds

This is the part that surprised me when I ran the numbers, and it is the reason I wrote this post.

Take a tight envelope: 182 mm of length and 40 mm of thickness. 46 of the 300 cards clear both. Then add a 69 mm height limit, which is what a low profile layout gives you. 11 cards survive. Height removes 35 of the 46 cards that had already passed the other two tests.

The reason is that height is quantised by the bracket, not by the cooler. The PCI Express Card Electromechanical Specification defines a low profile card as no more than 68.90 mm tall and a standard height card as no more than 111.28 mm. Cards cluster hard against those two ceilings. In my set, 14 cards are 69 mm or shorter and 120 of them land between 110 mm and 120 mm. There is almost nothing in between, because there is no bracket in between.

That gap is why height behaves like a switch rather than a dial. You are not choosing a height. You are choosing which of two card populations you are allowed to shop from, and one of them has 14 members.

What each clearance limit actually buys you

Here is the same dataset cut by the limits a case might advertise. Each row is the share of 300 cards that clears that single limit, with everything else ignored.

Read the last two rows together and you can see the shape of the problem. There is no useful height limit between 69 mm and about 110 mm, because no cards live there.

Read the vendor page, not the retailer listing

Retailer listings paraphrase. Vendor pages publish. They also disagree with each other about what to call the three numbers, which is worth seeing once so it stops catching you out.

Gigabyte’s spec page for the GeForce RTX 5060 OC Low Profile 8G gives the card as L=182 W=69 H=36. So Gigabyte’s W is what your case calls height, and Gigabyte’s H is what your case calls thickness. Take those labels at face value and you will compare the wrong pair of numbers.

ASRock’s page for the Radeon RX 6400 Low Profile 4GB gives 150 x 68.9 x 18 mm with no labels at all. That one is unambiguous only because you already know the convention.

Both of those match my dataset rows exactly, which is the point of checking. One more thing to watch: vendors often publish the bare card and the mounting bracket adds roughly 10 to 13 mm to the length. If a card is within 10 mm of your limit, assume it does not fit until you have measured one.

Under4L at a glance

For context, here is the case I designed and print, with the same hard limits laid out the way a case spec should be.

Note the height column. It is the number doing most of the work: a 69 mm height limit is why the case takes 11 of the 300 cards rather than 46. The build log and the print settings are in the Under4L write up, and the files are on Gumroad at $9.98.

Check a card yourself in one command

The dataset is public JSON, so you do not have to take my summary for it. This prints every card that fits an envelope you set:

curl -s https://marcosponce.com/apps/gpu-dimensions.json \
| python3 -c '
import json,sys
L,H,T = 182, 69, 40
for c in json.load(sys.stdin)["records"]:
    if c["length_mm"]<=L and c["height_mm"]<=H and c["thickness_mm"]<=T:
        print(c["gpu"], "|", c["card"], "|",
              c["length_mm"], c["height_mm"], c["thickness_mm"])
'

Change the three numbers on the L, H, T line to your case's limits. If your case only publishes a length, put a generous height and thickness in and see how much the list moves when you tighten them. That tells you which dimension is actually binding before you spend any money.

The field definitions, units and the measurement caveats are documented on the dataset page. Most of the figures are manufacturer stated, and about one in eight rests on a single listing, so treat a 2 mm margin as no margin.

What I would actually do before buying

Four checks, in this order, because they fail in this order.

First, find the card's height. If your case is low profile, that decision is already made for you and the rest of the shopping list is 14 cards long. If it is not low profile, note the height anyway, because a sandwich layout will care.

Second, get the thickness in millimetres from the vendor page. Ignore the slot count entirely. If the vendor only gives a slot count, pick a different card or accept that you are guessing.

Third, check length and add 13 mm for the bracket unless the vendor says the figure already includes it.

Fourth, compare against the case, not against another card. A card that fit someone else's build tells you nothing unless you know their case's three limits, and most build posts do not publish them.

If none of this is going your way, the honest option is to drop the discrete card. For a media box in particular, an integrated GPU does the job, and passing an integrated GPU into a container removes the clearance problem completely. A build with no graphics card has infinite clearance.

FAQ

What does SFF case GPU clearance actually mean?

It means the maximum size of graphics card a case can physically hold, expressed as three separate limits: length along the slot, height measured up from the slot, and thickness across the slots. A case that quotes only one of those has told you about a third of what you need.

Why do case makers use slot counts instead of millimetres?

Because a slot count is how the motherboard sees the card, and it is the older convention. It survived because in a full size tower it is close enough. My set of 300 cards has 58 distinct thickness values. 217 of them land in the 38 mm to 45 mm band that usually gets called two slot, so one label is covering a 7 mm spread.

How tall is a low profile card exactly?

The PCI Express Card Electromechanical Specification caps a low profile card at 68.90 mm measured up from the slot, against 111.28 mm for a standard height card. Vendors usually publish 69 mm. There is effectively nothing on the market between those two ceilings because the bracket is what sets the height.

Does the published card length include the bracket?

Often not. Vendors frequently publish the bare card, and the mounting bracket adds roughly 10 to 13 mm. Two vendors can describe the same physical card with numbers 13 mm apart and both be telling the truth. If you are within 10 mm of your limit, measure rather than trust.

My case says it fits a 250 mm card. Is that enough for most GPUs?

For length, largely yes. 224 of the 300 cards in my dataset are 250 mm or shorter, so three quarters of the market is open to you. Length is usually not the constraint at that point. Thickness and height are what to check next.

Where did these 300 measurements come from?

Manufacturer specification pages, mostly, with a retailer listing used where the vendor published nothing. Each row carries a source quality flag, and 39 of the 300 rest on a single or secondary source. The units, the field definitions and the caveats are all documented on the dataset page rather than being left for you to infer.

The short version

A graphics card is three numbers and a case is three limits, but the two sides are written in different vocabularies. Length is comparable. Thickness arrives rounded into a slot count that hides a 7 mm spread. Height is usually missing, and it is the one that removed 35 of the 46 otherwise viable cards in the tightest envelope I tested.

Get all three from the vendor's own page, add the bracket to the length, and compare millimetres to millimetres. If you want to run the check against real data instead of taking my word for it, the GPU dimensions dataset is free and documented.