Graphics card memory explained: GDDR6, GDDR6X and GDDR7
Memory capacity is the number on the box, but bus width and memory type often matter more. Here is how to read a graphics card's memory specification properly.
Graphics card marketing leads with capacity, because a larger number is easier to sell than a faster one. But two cards with identical capacity can perform very differently, and the reason is usually in the specs sitting underneath it.
The three numbers that matter
Capacity: how much data the card can hold locally. This sets the ceiling on texture quality and resolution before the card has to fetch across the PCIe bus, which is dramatically slower.
Memory type: GDDR6, GDDR6X or GDDR7. Each generation raises the data rate per pin. GDDR7 is the current standard on new high-end cards.
Bus width: how many bits move per clock, typically 128-bit, 192-bit, 256-bit or wider. This is the one most often overlooked and it is a direct multiplier on bandwidth.
Bandwidth, which is what actually governs performance, is roughly data rate × bus width. A 128-bit card with fast memory can easily be outperformed by a 256-bit card with slower memory.
Why capacity alone misleads
A card advertised with 16GB on a 128-bit bus and one with 12GB on a 192-bit bus are not straightforwardly ranked by the capacity figure. The 12GB card may sustain higher frame rates at the resolutions either can handle, because it can move data faster, while the 16GB card only shows its advantage in the narrower set of workloads that genuinely exhaust 12GB.
Capacity is a ceiling. Bandwidth is a rate. Running out of capacity is catastrophic, because performance collapses as data shuttles over PCIe. Being short on bandwidth is a gradual, proportional cost.
So the sensible approach is: get enough capacity for your resolution, then optimise for bandwidth within that.
How much capacity is enough
This shifts over time as games grow, but as a general guide:
- 1080p: 8GB is workable, 12GB comfortable.
- 1440p: 12GB is a sensible floor.
- 4K: 16GB or more, particularly with ray tracing enabled.
Ray tracing and frame generation both increase memory use meaningfully. If you intend to use them, size up rather than down.
What GPU core counts tell you
You will also see a core count: CUDA cores on NVIDIA, stream processors on AMD. These are comparable within a generation and vendor, and misleading across them. An NVIDIA card and an AMD card with similar core counts are not thereby similar in performance; the architectures differ too much for the number to transfer.
Treat core count as a way to rank cards in the same family, not as a cross-brand metric.
Reading a spec table properly
For any card, work through in this order:
- Capacity: is it enough for your resolution? If not, stop; nothing else compensates.
- Bus width: wider is better, and it is where cheaper cards cut corners.
- Memory type: newer generations mean a higher data rate.
- Boost clock: meaningful within a family, less so across them.
- TDP: determines your PSU and cooling requirements.
Every graphics card on VantaSpec lists Memory, Memory Bus, GPU Cores, Boost Clock and TDP, so all five are visible on one page rather than scattered across a manufacturer's site.