GeForce GTX 670
Everyday Tasks
Evaluation - Excellent
Compared with GeForce GT 1030, a baseline GPU for everyday tasks such as Office work and web browsing, it delivers +124% performance.
Light Gaming
Evaluation - Difficult
Compared with GeForce GTX 1650, a baseline GPU for light games such as Valorant, Minecraft, Roblox, UNDERTALE, and League of Legends at FHD / 120 FPS, it delivers -32% performance.
Mid-range Gaming
Evaluation - Insufficient VRAM
Regardless of its performance score, this GPU falls short of the 4 GB of VRAM this task calls for, so it will struggle.
Heavy Gaming
Evaluation - Insufficient VRAM
Regardless of its performance score, this GPU falls short of the 8 GB of VRAM this task calls for, so it will struggle.
1080p Video Editing
Evaluation - Difficult
Compared with GeForce GTX 1650, a baseline GPU for 1080p video editing in tools like Premiere Pro, DaVinci Resolve, and Filmora, it delivers -32% performance.
4K Video Editing
Evaluation - Insufficient VRAM
Regardless of its performance score, this GPU falls short of the 8 GB of VRAM this task calls for, so it will struggle.
FHD Image Editing
Evaluation - Borderline
Compared with GeForce GTX 1050 Ti, a baseline GPU for FHD image editing in Photoshop, Lightroom, Figma, and Clip Studio Paint, it delivers -16% performance.
3D CG Creation
Evaluation - Insufficient VRAM
Regardless of its performance score, this GPU falls short of the 4 GB of VRAM this task calls for, so it will struggle.
Game Development
Evaluation - Insufficient VRAM
Regardless of its performance score, this GPU falls short of the 4 GB of VRAM this task calls for, so it will struggle.
CAD
Evaluation - Difficult
Compared with GeForce GTX 1650, a baseline GPU for CAD work in tools like AutoCAD and Fusion 360, it delivers -32% performance.
Local LLM
Evaluation - Not applicable
Launched in 2012, its design predates the generations current AI tooling assumes (2018 or later), so it is outside the scope of this quick check.
As a rule of thumb, (quantized) 8B models such as Gemma, Qwen, Llama, and DeepSeek need 8 GB of VRAM, 22B models need 16 GB, and 32B models need 24 GB. At the same VRAM capacity, newer generations also run faster. The field moves fast, so this is a quick check of where GeForce GTX 670's VRAM stands.
AI Image Generation
Evaluation - Not applicable
Launched in 2012, its design predates the generations current AI tooling assumes (2018 or later), so it is outside the scope of this quick check.
As a rule of thumb, (quantized) 2–6B models such as Krea, Flux, SDXL, Pony, Anima, and Illustrious need 8 GB of VRAM, 12B models need 16 GB, and 20B models need 24 GB. At the same VRAM capacity, newer generations also run faster. The field moves fast, so this is a quick check of where GeForce GTX 670's VRAM stands.
AI Video Generation
Evaluation - Not applicable
Launched in 2012, its design predates the generations current AI tooling assumes (2018 or later), so it is outside the scope of this quick check.
As a rule of thumb, (quantized) models such as Wan, LTX, and Hunyuan need at least 16 GB of VRAM for practical use. At the same VRAM capacity, newer generations also run faster. The field moves fast, so this is a quick check of where GeForce GTX 670's VRAM stands.
GPUMETA / price / minimum
GeForce GTX 670
This site uses affiliate links.
GPUMETA / ranking / score