GeForce RTX 3060 12GB
A 12 GiB discrete GPU that leaves room for many small and medium quantized models, subject to system RAM and context requirements.
- Vendor
- NVIDIA
- Type
- Discrete
- Dedicated VRAM
- 12 GiB
- Architecture
- Ampere
Curated GPU catalog
Browse graphics hardware
Compare the catalog’s dedicated memory capacities and documented hardware details. Open a GPU for its source and planning notes.
GeForce RTX 4060 8GB
An 8 GiB discrete GPU for smaller and medium quantized models; larger models may require partial offload or CPU execution.
- Vendor
- NVIDIA
- Type
- Discrete
- Dedicated VRAM
- 8 GiB
- Architecture
- Ada Lovelace
GeForce RTX 4070 SUPER 12GB
A 12 GiB discrete GPU with the same capacity class as other 12 GiB cards; the calculator intentionally does not infer speed from the GPU name.
- Vendor
- NVIDIA
- Type
- Discrete
- Dedicated VRAM
- 12 GiB
- Architecture
- Ada Lovelace
GeForce RTX 4090 24GB
A 24 GiB discrete GPU with substantial capacity for larger quantized models, without implying a particular speed or context guarantee.
- Vendor
- NVIDIA
- Type
- Discrete
- Dedicated VRAM
- 24 GiB
- Architecture
- Ada Lovelace
Radeon RX 7600 8GB
An 8 GiB discrete GPU; actual llama.cpp backend support and practical results depend on the selected runtime and driver stack.
- Vendor
- AMD
- Type
- Discrete
- Dedicated VRAM
- 8 GiB
- Architecture
- RDNA 3
Radeon RX 7800 XT 16GB
A 16 GiB discrete GPU with capacity for a wider range of quantized models; compatibility still depends on runtime support and system memory.
- Vendor
- AMD
- Type
- Discrete
- Dedicated VRAM
- 16 GiB
- Architecture
- RDNA 3
Radeon RX 7900 XTX 24GB
A 24 GiB discrete GPU with high memory capacity; no backend performance or universal compatibility claim is made.
- Vendor
- AMD
- Type
- Discrete
- Dedicated VRAM
- 24 GiB
- Architecture
- RDNA 3
Intel Arc A750 8GB
An 8 GiB discrete GPU; backend and driver support should be checked for the chosen llama.cpp build.
- Vendor
- Intel
- Type
- Discrete
- Dedicated VRAM
- 8 GiB
- Architecture
- Alchemist
Intel Arc A770 16GB
A 16 GiB discrete GPU; local LLM results depend on the llama.cpp backend, driver, and offload configuration.
- Vendor
- Intel
- Type
- Discrete
- Dedicated VRAM
- 16 GiB
- Architecture
- Alchemist
Intel UHD Graphics 770
Integrated graphics with no dedicated VRAM entry; LLMGauge conservatively treats shared system memory as CPU-only capacity.
- Vendor
- Intel
- Type
- Integrated
- Dedicated VRAM
- 0 GiB (integrated)
- Architecture
- Xe integrated graphics
GeForce RTX 2060 6GB
An older 6 GiB discrete card; useful for evaluating smaller quantized models, with limited dedicated-memory headroom.
- Vendor
- NVIDIA
- Type
- Discrete
- Dedicated VRAM
- 6 GiB
- Memory type
- GDDR6
- Architecture
- Turing
GeForce RTX 4060 Ti 16GB
The 16 GiB RTX 4060 Ti variant offers more model-memory capacity than the 8 GiB version; the catalog treats capacity separately from speed.
- Vendor
- NVIDIA
- Type
- Discrete
- Dedicated VRAM
- 16 GiB
- Memory type
- GDDR6
- Architecture
- Ada Lovelace
GeForce RTX 4070 Ti SUPER 16GB
A 16 GiB discrete GPU that broadens the catalog's mid/high capacity range without implying model speed or guaranteed fit.
- Vendor
- NVIDIA
- Type
- Discrete
- Dedicated VRAM
- 16 GiB
- Memory type
- GDDR6X
- Architecture
- Ada Lovelace
GeForce RTX 5080 16GB
A current 16 GiB discrete option; compatibility uses memory capacity only and does not infer backend availability or performance.
- Vendor
- NVIDIA
- Type
- Discrete
- Dedicated VRAM
- 16 GiB
- Memory type
- GDDR7
- Architecture
- Blackwell
Radeon RX 6600 8GB
An established 8 GiB discrete GPU; actual llama.cpp backend and driver support depend on the user's software stack.
- Vendor
- AMD
- Type
- Discrete
- Dedicated VRAM
- 8 GiB
- Memory type
- GDDR6
- Architecture
- RDNA 2
Radeon RX 7700 XT 12GB
A 12 GiB RDNA 3 card that fills the catalog's AMD mid-capacity range; memory fit does not confirm backend support.
- Vendor
- AMD
- Type
- Discrete
- Dedicated VRAM
- 12 GiB
- Memory type
- GDDR6
- Architecture
- RDNA 3
Radeon RX 9070 16GB
A newer 16 GiB discrete option; LLMGauge reports its memory class while leaving runtime and driver compatibility to the user's setup.
- Vendor
- AMD
- Type
- Discrete
- Dedicated VRAM
- 16 GiB
- Memory type
- GDDR6
- Architecture
- RDNA 4
Radeon RX 9070 XT 16GB
The RX 9070 XT provides a current 16 GiB AMD option; no speed or universal llama.cpp support is inferred.
- Vendor
- AMD
- Type
- Discrete
- Dedicated VRAM
- 16 GiB
- Memory type
- GDDR6
- Architecture
- RDNA 4
Intel Arc B570 10GB
A 10 GiB Intel discrete GPU; check the selected llama.cpp build and driver for backend support before relying on offload.
- Vendor
- Intel
- Type
- Discrete
- Dedicated VRAM
- 10 GiB
- Memory type
- GDDR6
- Architecture
- Battlemage
Intel Arc B580 12GB
A 12 GiB Intel discrete GPU that expands the memory-capacity range; available runtime paths remain software-dependent.
- Vendor
- Intel
- Type
- Discrete
- Dedicated VRAM
- 12 GiB
- Memory type
- GDDR6
- Architecture
- Battlemage
AMD Radeon 780M integrated graphics
Integrated graphics use system memory rather than a fixed dedicated VRAM capacity; LLMGauge conservatively classifies this as CPU-only capacity.
- Vendor
- AMD
- Type
- Integrated
- Dedicated VRAM
- 0 GiB (integrated)
- Architecture
- RDNA 3