readygpu

NVIDIA GeForce RTX 5090 vs AMD Radeon RX 9070

For running models locally, priced in the US. At 128k context the first runs 50 models fully in memory, the second 21.

In short

The NVIDIA GeForce RTX 5090 has more memory for models: 32 GB against 16 GB. At 128k context the NVIDIA GeForce RTX 5090 runs 50 of the models we track fully in memory and the AMD Radeon RX 9070 21; only the NVIDIA GeForce RTX 5090 holds Qwen 3.8 27B. On the models both run fully in memory, the NVIDIA GeForce RTX 5090 is about 3.2× as fast as the AMD Radeon RX 9070 (estimated, median across models).

Updated . Speeds are estimates; how they are worked out.

NVIDIA GeForce RTX 5090

$4,921

eBay

Add to build
AMD Radeon RX 9070

Not priced here

Add to build

Specifications

SpecNVIDIA GeForce RTX 5090AMD Radeon RX 9070
GPUGeForce RTX 5090Radeon RX 9070
Memory32 GB GDDR716 GB GDDR6
Memory bus512-bit256-bit
Memory bandwidth1,792 GB/s640 GB/s
FP16 compute209 TFLOPS36.1 TFLOPS
SoftwareCUDAROCm
Board power575 W220 W
PCIe generationGen 5Gen 5
PCIe lanesx16x16

What each runs

Estimated decode speed with the most precise quant that fits each one est.

ModelNVIDIA GeForce RTX 5090AMD Radeon RX 9070
Qwen 3.8 27B36tok/sUD-Q6_Koffload
Qwen 3.6 27B36tok/sQ6_Koffload
Gemma 4 31B37tok/sQ4_K_Moffload
Qwen 3.6 35B-A3B128tok/sUD-Q6_Koffload
Qwen 3.5 27B36tok/sQ6_Koffload
Gemma 4 26B-A4B95tok/sQ8_0offload
Qwen 3.5 35B-A3B129tok/sQ6_Koffload
GLM 4.7 Flash60tok/sQ6_Koffload
Nemotron 3.5 Lightning 30B-A3B179tok/sUD-Q5_K_Moffload
Kimi Linear 48B-A3B227tok/sQ4_K_Moffload
gpt-oss 20B118tok/sMXFP4offload
Qwen3 Coder 30B-A3B45tok/sQ4_K_Moffload
Gemma 4 12B73tok/sQ8_029tok/sQ6_K
Qwen 3.5 9B80tok/sQ8_025tok/sQ8_0
Gemma 3 27B37tok/sQ5_K_Moffload
Ministral 3 14B36tok/sQ5_K_Moffload
Qwen3 14B35tok/sQ5_K_Moffload
DeepSeek R1 0528 Qwen3 8B40tok/sQ8_0offload
Qwen3 8B40tok/sQ8_0offload
Qwen 3.5 4B126tok/sQ8_040tok/sQ8_0
Ministral 3 8B41tok/sQ8_0offload
Granite 4.2 8B36tok/sQ8_0offload
Gemma 4 E4B93tok/sQ8_029tok/sQ8_0
LFM2.5 8B-A1B89tok/sQ8_029tok/sUD-Q6_K
Llama 3.1 8B47tok/sQ6_Koffload
Qwen2.5 0.5B519tok/sQ8_0164tok/sQ8_0
Qwen3.5 0.8B455tok/sQ8_0144tok/sQ8_0
Qwen3.5 2B363tok/sQ5_K_M115tok/sQ5_K_M
DeepSeek R1 Distill Qwen 1.5B197tok/sQ8_062tok/sQ8_0
Mellum2.1 12B A2.5B Thinking146tok/sQ8_049tok/sQ6_K
Spark X2.5 4B140tok/smlx-4bit44tok/smlx-4bit
MiniCPM5 2B134tok/sQ8_042tok/sQ8_0
FrogNano 4B 2609124tok/sQ8_039tok/sQ8_0
LFM2.5 2.6B102tok/sQ8_032tok/sQ8_0
LFM2.5 VL 3B102tok/sQ8_032tok/sQ8_0
Kimi VL A3B Thinking 250693tok/sQ8_035tok/sQ4_K_M
Kimi VL A3B91tok/smlx-8bitoffload
GLM 4 9B 041482tok/smlx-5bit26tok/smlx-5bit
granite vision 4.1 4b77tok/sQ8_025tok/sQ8_0
granite 4.2 3b76tok/sQ8_024tok/sQ8_0
GLM 4.6V Flash72tok/sQ8_031tok/sQ4_K_M
DeepSeek R1 Distill Qwen 7B71tok/sQ8_026tok/sQ6_K
GLM 4.1V 9B Thinking65tok/smlx-8bitoffload
Ministral 3 3B Instruct 251263tok/sQ8_0offload
LFM2.5 24B A2B53tok/sQ6_Koffload
Muse Glimmer 30B44tok/sQ6_Koffload
DeepSeek R1 Distill Llama 8B43tok/sQ8_0offload
Ministral 3 8B Reasoning 251241tok/sQ8_0offload
Qwen3.8 27B OBLITERATED38tok/smlx-4bitoffload
Ministral 3 14B Reasoning 251236tok/sQ5_K_Moffload
Qwen 3.5 122B-A10Boffloadwon't run
Qwen3 Coder Nextoffloadwon't run
Nemotron 3 Super 120B-A12Boffloadwon't run
Mistral Small 4 119Boffloadwon't run
Qwen3 Next 80B-A3Boffloadwon't run
Devstral Small 2 24Boffloadoffload
Qwen3 32Boffloadwon't run
Mistral Small 3.2 24Boffloadoffload
DeepSeek R1 Distill Qwen 32Boffloadwon't run
Phi-4 14Boffloadoffload
DeepSeek R1 Distill Qwen 14Boffloadoffload
Devstral Small 2507offloadoffload
Magistral Small 2506offloadoffload
Magistral Small 2507offloadoffload
Magistral Small 2509offloadoffload
granite 4.2 30boffloadwon't run
DeepSeek V4 Flashwon't runwon't run
GLM 5.3 Flashwon't runwon't run
Qwen 3.8 Flash Nextwon't runwon't run
MiniMax M2.7won't runwon't run
gpt-oss 120Bwon't runwon't run
GLM 4.5 Airwon't runwon't run
Llama 3.3 70Bwon't runwon't run
DeepSeek R1won't runwon't run
DeepSeek R1 0528won't runwon't run
DeepSeek R1 Distill Llama 70Bwon't runwon't run
DeepSeek R1 Zerowon't runwon't run
DeepSeek V3won't runwon't run
DeepSeek V3 0324won't runwon't run
DeepSeek V3.1won't runwon't run
DeepSeek V3.1 Terminuswon't runwon't run
DeepSeek V3.2won't runwon't run
DeepSeek V4 Flash 0731won't runwon't run
DeepSeek V4 Flash Vision Expwon't runwon't run
Devstral 2 123B Instruct 2512won't runwon't run
GLM 4.5won't runwon't run
GLM 4.5Vwon't runwon't run
GLM 4.6won't runwon't run
GLM 4.6Vwon't runwon't run
GLM 4.7won't runwon't run
GLM 5won't runwon't run
GLM 5.1won't runwon't run
GLM 5.2won't runwon't run
GLM 5.3won't runwon't run
Kimi Dev 72Bwon't runwon't run
Kimi K2won't runwon't run
Kimi K2 Instruct 0905won't runwon't run
Kimi K2 Thinkingwon't runwon't run
Kimi K2.5won't runwon't run
MiniMax M2won't runwon't run
MiniMax M2.1won't runwon't run
MiniMax M2.5won't runwon't run
MiniMax M3won't runwon't run
Mistral Medium 3.5 128Bwon't runwon't run

Questions

Which is better for local AI, the NVIDIA GeForce RTX 5090 or the AMD Radeon RX 9070?
Choose the NVIDIA GeForce RTX 5090 to run bigger models. Choose the NVIDIA GeForce RTX 5090 for speed.
Which can run bigger models, the NVIDIA GeForce RTX 5090 or the AMD Radeon RX 9070?
At 128k context the NVIDIA GeForce RTX 5090 runs 50 of the models we track fully in memory and the AMD Radeon RX 9070 21; only the NVIDIA GeForce RTX 5090 holds Qwen 3.8 27B.
Which is faster for local AI, the NVIDIA GeForce RTX 5090 or the AMD Radeon RX 9070?
On the models both run fully in memory, the NVIDIA GeForce RTX 5090 is about 3.2× as fast as the AMD Radeon RX 9070 (estimated, median across models).

More comparisons

More hardware for local AI