Desktop GPU Performance Ladder (2005–2026)
NVIDIA GeForce vs AMD Radeon — 21 years of desktop GPUs tier-ranked by real-world performance.
How to use this ladder
This tier list groups Desktop GPU Ladder into seven performance tiers from S (Supreme / flagship) to D (Legacy / entry-level). Each tier is ordered by release year, newest first. A newer chip one tier down often outperforms an older chip one tier up. Use this as a reference for relative positioning, not as a benchmark — actual FPS varies by game, resolution and driver. Tiers are updated quarterly to reflect new releases and re-benchmarks.
- NVIDIA GeForce RTX 5090 SuperNVIDIA2026
Blackwell GB202 refresh, 32GB GDDR7, 24000 CUDA, +8% over 5090, consumer flagship king
- AMD Radeon RX 9080 XTAMD2026
RDNA 4 Navi 48 refresh, 32GB GDDR6, 12500 SP, AMD raster flagship
- NVIDIA GeForce RTX 5090NVIDIA2025
Blackwell GB202, 32GB GDDR7, 21760 CUDA, consumer flagship
- NVIDIA GeForce RTX 4090NVIDIA2022
Ada Lovelace AD102, 24GB GDDR6X, 16384 CUDA, flagship gaming
- AMD Radeon RX 7900 XTXAMD2022
RDNA 3 Navi 31, 24GB GDDR6, 6144 SP, first chiplet GPU
- NVIDIA GeForce RTX 3090NVIDIA2020
Ampere GA102, 24GB GDDR6X, 10496 CUDA, last-gen flagship
- AMD Radeon RX 6900 XTXAMD2020
RDNA 2 Navi 21, 16GB GDDR6, 5120 SP, flagship raster
- NVIDIA GeForce RTX 2080 TiNVIDIA2018
Turing TU102, 11GB GDDR6, 4352 CUDA, first RT flagship
- NVIDIA GeForce RTX 5080 SuperNVIDIA2026
Blackwell GB203 refresh, 16GB GDDR7, 11904 CUDA, +7% over 5080
- AMD Radeon RX 9070 XTAMD2026
RDNA 4 Navi 48, 16GB GDDR6, 8192 SP, high-end value king
- NVIDIA GeForce RTX 5080NVIDIA2025
Blackwell GB203, 16GB GDDR7, 10752 CUDA, high-end gaming
- AMD Radeon RX 9080AMD2025
RDNA 4 Navi 48, 24GB GDDR6, 11200 SP, AMD high-end
- NVIDIA GeForce RTX 4080 SUPERNVIDIA2024
Ada Lovelace AD103, 16GB GDDR6X, 10240 CUDA, high-end
- NVIDIA GeForce RTX 4080NVIDIA2022
Ada Lovelace AD103, 16GB GDDR6X, 9728 CUDA
- AMD Radeon RX 7900 XTAMD2022
RDNA 3 Navi 31, 20GB GDDR6, 5376 SP
- AMD Radeon RX 6800 XTAMD2020
RDNA 2 Navi 21, 16GB GDDR6, 4608 SP
- NVIDIA GeForce RTX 3080NVIDIA2020
Ampere GA102, 10GB GDDR6X, 8704 CUDA, value flagship
- NVIDIA GeForce RTX 2080 SUPERNVIDIA2019
Turing TU104, 8GB GDDR6, 3072 CUDA
- NVIDIA GeForce GTX 1080 TiNVIDIA2017
Pascal GP102, 11GB GDDR5X, 3584 CUDA, classic flagship
- NVIDIA GeForce RTX 5070 SuperNVIDIA2026
Blackwell GB203 refresh, 12GB GDDR7, 7168 CUDA, refreshes the 5070 tier
- AMD Radeon RX 9070 GREAMD2026
RDNA 4 Navi 44, 12GB GDDR6, 4096 SP, China/value refresh
- NVIDIA GeForce RTX 5070 TiNVIDIA2025
Blackwell GB203, 12GB GDDR7, 8960 CUDA
- NVIDIA GeForce RTX 5070NVIDIA2025
Blackwell GB203, 12GB GDDR7, 6144 CUDA
- NVIDIA GeForce RTX 4070 Ti SUPERNVIDIA2024
Ada Lovelace AD104, 16GB GDDR6X, 8448 CUDA
- NVIDIA GeForce RTX 4070 SUPERNVIDIA2024
Ada Lovelace AD104, 12GB GDDR6X, 7168 CUDA
- NVIDIA GeForce RTX 4070NVIDIA2023
Ada Lovelace AD104, 12GB GDDR6X, 5888 CUDA
- AMD Radeon RX 7800 XTAMD2023
RDNA 3 Navi 32, 16GB GDDR6, 3840 SP
- AMD Radeon RX 7700 XTAMD2023
RDNA 3 Navi 32, 12GB GDDR6, 3072 SP
- AMD Radeon RX 6700 XTAMD2021
RDNA 2 Navi 22, 12GB GDDR6, 2560 SP
- NVIDIA GeForce RTX 3070NVIDIA2020
Ampere GA104, 8GB GDDR6, 5888 CUDA, value high-end
- NVIDIA GeForce RTX 2070 SUPERNVIDIA2019
Turing TU106, 8GB GDDR6, 2560 CUDA
- NVIDIA GeForce GTX 1070 TiNVIDIA2017
Pascal GP104, 8GB GDDR5, 2432 CUDA
- AMD Radeon RX Vega 64AMD2017
Vega 10, 8GB HBM2, 4096 SP, first HBM2 GPU
- NVIDIA GeForce RTX 5060 TiNVIDIA2025
Blackwell GB206, 8GB GDDR7, 3840 CUDA
- NVIDIA GeForce RTX 4060 TiNVIDIA2023
Ada Lovelace AD106, 8GB GDDR6X, 4352 CUDA
- NVIDIA GeForce RTX 4060NVIDIA2023
Ada Lovelace AD106, 8GB GDDR6, 3072 CUDA
- AMD Radeon RX 7600AMD2023
RDNA 3 Navi 33, 8GB GDDR6, 2048 SP
- AMD Radeon RX 6650 XTAMD2022
RDNA 2 Navi 23, 8GB GDDR6, 2048 SP
- AMD Radeon RX 6600 XTAMD2021
RDNA 2 Navi 23, 8GB GDDR6, 2048 SP
- AMD Radeon RX 6600AMD2021
RDNA 2 Navi 23, 8GB GDDR6, 1792 SP
- NVIDIA GeForce RTX 3060 TiNVIDIA2020
Ampere GA104, 8GB GDDR6, 4864 CUDA
- NVIDIA GeForce RTX 2060 SUPERNVIDIA2019
Turing TU106, 8GB GDDR6, 1920 CUDA
- NVIDIA GeForce RTX 2060NVIDIA2018
Turing TU106, 6GB GDDR6, 1920 CUDA
- NVIDIA GeForce GTX 1060 6GBNVIDIA2016
Pascal GP106, 6GB GDDR5, 1280 CUDA, legendary mid-range
- NVIDIA GeForce RTX 5060NVIDIA2026
Blackwell GB207, 8GB GDDR7, 2880 CUDA, mainstream Blackwell entry
- AMD Radeon RX 9060 XTAMD2026
RDNA 4 Navi 44, 16GB GDDR6, 3072 SP, mainstream 1080P/2K
- NVIDIA GeForce RTX 5050NVIDIA2025
Blackwell GB207, 8GB GDDR6, 2560 CUDA
- NVIDIA GeForce RTX 4050NVIDIA2024
Ada Lovelace AD107, 6GB GDDR6, 3072 CUDA
- AMD Radeon RX 6500 XTAMD2022
RDNA 2 Navi 24, 4GB GDDR6, 1024 SP
- NVIDIA GeForce RTX 3060NVIDIA2021
Ampere GA106, 12GB GDDR6, 3584 CUDA
- AMD Radeon RX 5700 XTAMD2019
RDNA 1 Navi 10, 8GB GDDR6, 2560 SP
- NVIDIA GeForce GTX 1660 TiNVIDIA2018
Turing TU116, 6GB GDDR6, 1536 CUDA
- NVIDIA GeForce GTX 1660NVIDIA2018
Turing TU116, 6GB GDDR5, 1408 CUDA
- AMD Radeon RX 580AMD2017
Polaris 20, 8GB GDDR5, 2304 SP
- NVIDIA GeForce GTX 1060 3GBNVIDIA2016
Pascal GP106, 3GB GDDR5, 1152 CUDA
- NVIDIA GeForce RTX 3050NVIDIA2022
Ampere GA107, 4GB GDDR6, 2048 CUDA
- AMD Radeon RX 5600 XTAMD2020
RDNA 1 Navi 10, 6GB GDDR6, 2304 SP
- NVIDIA GeForce GTX 1650 SUPERNVIDIA2019
Turing TU117, 4GB GDDR6, 1280 CUDA
- NVIDIA GeForce GTX 1650NVIDIA2019
Turing TU117, 4GB GDDR5, 1024 CUDA
- AMD Radeon RX 5500 XTAMD2019
RDNA 1 Navi 14, 8GB GDDR6, 1408 SP
- AMD Radeon RX 5700AMD2019
RDNA 1 Navi 10, 8GB GDDR6, 1792 SP
- AMD Radeon RX 570AMD2017
Polaris 20, 4GB GDDR5, 2048 SP
- NVIDIA GeForce GTX 1050 TiNVIDIA2016
Pascal GP107, 4GB GDDR5, 768 CUDA
- NVIDIA GeForce GTX 1050NVIDIA2016
Pascal GP107, 2GB GDDR5, 640 CUDA
- AMD Radeon RX 470AMD2016
Polaris 10, 4GB GDDR5, 2048 SP
- NVIDIA GeForce GT 1030NVIDIA2017
Pascal GP108, 2GB GDDR5, 384 CUDA, passively cooled
- NVIDIA GeForce GTX 750 TiNVIDIA2015
Maxwell GM107, 2GB GDDR5, 640 CUDA, first Maxwell
- AMD Radeon R7 360AMD2014
Tonga PRO, 2GB GDDR5, 768 SP
- NVIDIA GeForce GTX 660NVIDIA2012
Kepler GK106, 2GB GDDR5, 960 CUDA, classic budget
- AMD Radeon HD 7750AMD2012
Cape Verde, 1GB GDDR5, 512 SP
- NVIDIA GeForce 9600 GTNVIDIA2008
G94, 512MB GDDR3, 64 CUDA, classic budget
- AMD Radeon HD 3850AMD2007
RV670, 512MB GDDR3, 320 SP, first unified shader
- NVIDIA GeForce 7600 GTNVIDIA2006
G73, 256MB GDDR3, 12 pixel pipelines
Methodology & update cadence
Tiers are assigned based on real-world benchmark consensus across multiple publications (AnandTech, TechPowerUp, Tom's Hardware, NotebookCheck). Gaming performance at 1080P/1440P/4K is weighted for consumer chips; multi-core and productivity benchmarks are weighted for workstation parts. Tiers are relative, not absolute — a chip's position reflects its standing at the time of its launch, not its performance today. New entries are added quarterly (March, June, September, December) as new chips launch and comprehensive benchmarks become available.
Last updated: 2026-08-02 · Next update: quarterly cadence