Vendor Slot Labels Lie. Here Are 300 GPUs in Millimetres.

Graphics card makers label coolers in slots, and slot labels are rounded marketing numbers. A card sold as “2-slot” can be anywhere from 36 mm to 43 mm thick, which is the difference between fitting a small case and not. So while designing a Mini ITX case I pulled the real millimetre dimensions for 300 graphics cards, one source at a time, and put them in the table below. Length, height, thickness, power draw, connector, and where each number came from.

This is the SFF GPU dimensions table I wanted when I started and could not find. Every card here uses a single 8-pin connector or less, which is the practical ceiling for a Flex ATX power supply. The table covers eight generations: RTX 20, 30, 40 and 50, Radeon RX 5000, 6000, 7000 and 9000, and both Intel Arc families.

Why slot labels are useless for case design

A PCI Express slot is 20.32 mm wide, so “2-slot” ought to mean about 40 mm. In practice vendors round to whatever number their marketing prefers, and the same label covers a 7 mm spread. These are all real examples from the table:

• ASUS Dual RTX 5050, labelled 2-slot, measures 40 mm.
• ASUS Dual EVO RTX 5060, labelled 2.1-slot, measures 42 mm. The standard Dual, same family, is 50 mm.
• Sapphire RX 9060 XT Pulse, labelled 2.3-slot, measures 46.1 mm.
• Onix Arc B580 LUMI, labelled 2.2-slot, measures 52 mm. That label would imply about 45.
• Zotac is the honest one: it labels its RTX 4060 Ti Twin Edge 2.2-slot and publishes 40.1 mm.

If you are designing an enclosure, or checking whether a card clears a heatsink or a side panel, the slot label tells you nothing you can build to. Use the millimetre figure. I later went through which of the three dimensions actually binds first in a small build, using this same table, in the SFF case GPU clearance guide.

The table: 300 graphics cards in millimetres

Sorted by thickness, thinnest first. Click any column heading to re-sort, or filter by maximum length and thickness to see exactly what fits a given space. Thickness is the dimension that runs across the slots, height is measured from the slot upward, and length runs along the card. All figures are in millimetres.

Download the raw data

The whole table is available as machine-readable files, free to use with attribution (CC BY 4.0). If you are building a compatibility checker, a case configurator, or you just want it in a spreadsheet, start here rather than scraping the page.

• gpu-dimensions.json: all 300 records with a metadata header describing units, licence and caveats.
• gpu-dimensions-csv.txt: the same 300 rows as CSV, served as plain text so it opens in a browser. Save it with a .csv extension for Excel or Sheets.

Columns are gpu, generation, card, length_mm, height_mm, thickness_mm, tdp_watts, power_connector, connector_position, availability and source_quality. Thickness is the dimension across the slots, the one vendors call slot width.

A ⚠ marks a row that rests on a single or secondary source, usually a retailer listing or a sister-brand card with an identical cooler. Treat those as provisional. Prices move too fast to publish here, so the table sticks to physical facts.

How many cards fit a given space

This is the table I actually used to pick an envelope. Each cell counts how many of the 300 cards fit within that length and thickness at the same time.

Read across any row and the same pattern shows up: the jump from 40 mm to 42 mm is the steepest single step in the whole grid, and 42 to 45 mm adds comparatively little. Read down a column and length matters much less than you would expect. At 230 mm, spending two more millimetres of thickness buys more cards than extending from 185 mm to 230 mm buys in the first place.

The practical takeaway for anyone designing around a 40 mm limit: 41 of the 82 cards that pass 230 × 40 do so with less than a millimetre to spare, and those are vendor nominal figures, not measurements. A cavity built to exactly 40.0 mm will reject some of them. Whatever number you publish, build in a couple of millimetres of real clearance behind it.

Cooler thickness by generation

Four generations of the same power class, and the trend is a cycle rather than a march in one direction.

Coolers got thicker from Ampere and RDNA2 to Ada and RDNA3, with the share of true 40 mm cards falling from 47% to 24%, then recovered partially with Blackwell and RDNA4. The recovery is real and traceable to specific products: ASUS created the Dual EVO line to hit 42 mm when the standard Dual is 50, PNY’s entire RTX 50 range sits at 39.8 to 39.9 mm, and MSI’s Shadow 2X and Gigabyte’s Windforce put genuine 40 mm coolers back on shelves.

The more useful number is the last column. The share of cards at or under 42 mm has stayed between 64% and 76% across every generation, while the share at or under 40 mm swings from 24% to 47%. If you are designing something that has to keep working in three years, 42 mm rides a stable band and 40 mm rides a volatile one.

The thinnest cards that are also short

Anything under 185 mm long and 40 mm thick, ranked by performance. This is the shortlist for the smallest builds.

Where the numbers came from

Manufacturer spec pages first, then official brochure PDFs (Zotac and Sparkle publish good ones), then retailer spec sheets, then reviews that printed measured figures. TechPowerUp, PCPartPicker and VideoCardz all block automated access, so this data was not scraped from an existing database. It was assembled source by source, and then every row was checked a second time against its original source before publication.

That second pass changed four things and confirmed the rest, which is roughly the error rate you should expect from any dataset like this. Two of the corrections were connector types on Zotac RTX 20-series cards that a European retailer had listed wrong. One was a Yeston card whose vendor page went live between the first pass and the second. One was a PowerColor claim about connector position that traced back to a review quote that turned out not to exist, so the claim was softened rather than deleted.

Known limits, stated plainly. Most figures are the manufacturer’s own, and manufacturers occasionally publish the bare card while the bracket adds 10 to 13 mm of length. EVGA never published millimetre thickness for any card, so EVGA rows are absent rather than guessed. Connector position is documented by almost nobody, so where it matters the table says so instead of pretending. Cards from Colorful, Maxsun, Manli, Yeston and Vastarmor are thin on primary sources, and rows that could not be verified were left out rather than filled in.

What I built with it

This started as background work for a 3D printed Mini ITX case, where the entire design question was how much space to give the graphics card. The data pushed the answer in a direction I did not expect. I had assumed the length of the case was the constraint worth spending volume on, and the coverage table says thickness is worth about twice as much per millimetre in the range that matters.

The finished case ended up at 218 × 240 × 147 mm with a 230 × 45 mm graphics card opening, which supports 134 of the 300 cards here. There is a shorter 185 mm panel option for people building the smallest possible version, and that one supports 47. More on the design itself, and the other small form factor builds, in the project write-ups.

The research itself was run through Claude, which is how most things on this site get built now. Roughly 300 source pages fetched and cross-checked in an afternoon, then a second full verification pass. That is not a job I would have done by hand.

Questions people ask

How thick is a 2-slot graphics card in mm?

A PCI Express slot is 20.32 mm, so a true 2-slot card should be about 40 mm. In practice cards labelled 2-slot in this dataset range from 36 mm to 43 mm. The label is not a specification, so use the manufacturer’s millimetre figure for the specific card.

What is the thinnest current graphics card?

Among cards with a single 8-pin connector or less, the thinnest full-height models are the Gigabyte RTX 5050 and RTX 5060 OC Low Profile at 36 mm, and the Zotac RTX 5050 and RTX 5060 Solo at 36.4 mm. Single-slot low-profile Radeon RX 6400 cards go down to 15 to 18 mm but are far slower.

What is the shortest RTX 5060?

The MSI Inspire ITX at 145 mm long, though it is 45 mm thick. The PNY Single Fan is 146.7 mm at 40 mm thick, and the Zotac Solo is 164.5 mm at 36.4 mm, which makes the Zotac the easiest of the three to fit in a tight enclosure.

Which graphics cards use only one 8-pin connector?

Every card in this table does, or uses a 6-pin or no external power at all. That includes all RTX 5050, 5060 and 5060 Ti models except a handful of 12V-2×6 variants, the whole RX 9060 XT family, and most 150 W to 180 W cards. The exceptions worth knowing: several 175 W RTX 2070 and RX 5600 XT models use 8+6-pin despite the modest power draw, and the MSI RX 5700 MECH and MECH GP look identical but the GP adds a 6-pin.

Do vendor dimensions include the bracket?

Usually not, and vendors rarely say. KFA2 publishes both figures for its RTX 3060, 245 mm for the body and 258 mm with the bracket, a 13 mm difference. PowerColor and ASRock also publish both for some cards. If you are designing to a hard length limit, assume the bracket adds 10 to 13 mm unless the vendor says otherwise.

Take the data, and the case it was built for

All 300 records are yours under CC BY 4.0 as JSON or CSV. The dataset page documents every field, its units, and the caveats worth knowing before you trust a number.

I built this table because I needed it for a case of my own. That case is the Under4L, a 3.73 litre Mini ITX design that takes a 182 x 69 x 40mm card, and the STL files are on Gumroad for $9.98.