readygpu

NVIDIA GeForce RTX 3090 Ti vs NVIDIA RTX 6000 Ada Generation

For running models locally, priced in Germany. At 8k context the first runs 60 models fully in memory, the second 68.

In short

The NVIDIA RTX 6000 Ada Generation has more memory for models: 48 GB against 24 GB. At 8k context the NVIDIA RTX 6000 Ada Generation runs 68 of the models we track fully in memory and the NVIDIA GeForce RTX 3090 Ti 60; only the NVIDIA RTX 6000 Ada Generation holds Qwen 3.5 122B-A10B. On the models both run fully in memory, the NVIDIA GeForce RTX 3090 Ti is about 1.1× as fast as the NVIDIA RTX 6000 Ada Generation (estimated, median across models).

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

NVIDIA GeForce RTX 3090 Ti

2.500 €

Amazon

Add to build
NVIDIA RTX 6000 Ada Generation

Not priced here

Add to build

Specifications

SpecNVIDIA GeForce RTX 3090 TiNVIDIA RTX 6000 Ada Generation
GPUGeForce RTX 3090 TiRTX 6000 Ada Generation
Memory24 GB GDDR6X48 GB GDDR6
Memory bus384-bit384-bit
Memory bandwidth1,008 GB/s960 GB/s
FP16 compute80 TFLOPS91.1 TFLOPS
SoftwareCUDACUDA
Board power450 W300 W
PCIe generationGen 4Gen 4
PCIe lanesx16x16

What each runs

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

ModelNVIDIA GeForce RTX 3090 TiNVIDIA RTX 6000 Ada Generation
Qwen 3.8 27B28tok/sUD-Q6_K20tok/sQ8_0
Qwen 3.5 122B-A10Boffload97tok/sUD-Q2_K_XL
Qwen 3.6 27B27tok/sQ6_K20tok/sQ8_0
Gemma 4 31B27tok/sQ5_K_M17tok/sQ8_0
Qwen3 Coder Nextoffload135tok/sQ4_K_M
Mistral Small 4 119Boffload139tok/sUD-Q2_K_XL
Qwen 3.6 35B-A3B179tok/sQ4_K_M108tok/sQ8_0
Qwen 3.5 27B27tok/sQ6_K20tok/sQ8_0
Gemma 4 26B-A4B95tok/sUD-Q6_K79tok/sQ8_0
GLM 4.5 Airoffload49tok/sQ2_K
Qwen 3.5 35B-A3B181tok/sQ4_K_M108tok/sQ8_0
GLM 4.7 Flash94tok/sQ5_K_M62tok/sQ8_0
Devstral Small 2 24B30tok/sQ6_K23tok/sQ8_0
Qwen3 Next 80B-A3Boffload173tok/sQ4_K_M
Nemotron 3.5 Lightning 30B-A3B163tok/sMXFP4104tok/sUD-Q6_K_XL
gpt-oss 20B142tok/sMXFP4136tok/sMXFP4
Qwen3 Coder 30B-A3B128tok/sQ5_K_M89tok/sQ8_0
Kimi Linear 48B-A3Boffload139tok/sQ6_K
Qwen3 32B29tok/sQ4_K_M16tok/sQ8_0
Gemma 4 12B48tok/sQ8_045tok/sQ8_0
Qwen 3.5 9B64tok/sQ8_060tok/sQ8_0
Mistral Small 3.2 24B30tok/sQ6_K23tok/sQ8_0
DeepSeek R1 Distill Qwen 32B28tok/sQ4_K_M16tok/sQ8_0
Gemma 3 27B27tok/sQ6_K20tok/sQ8_0
Llama 3.3 70Boffload13tok/sQ4_K_M
Ministral 3 14B40tok/sQ8_038tok/sQ8_0
Qwen3 14B37tok/sQ8_035tok/sQ8_0
DeepSeek R1 0528 Qwen3 8B63tok/sQ8_060tok/sQ8_0
Qwen3 8B63tok/sQ8_060tok/sQ8_0
Phi-4 14B36tok/sQ8_035tok/sQ8_0
Qwen 3.5 4B130tok/sQ8_0124tok/sQ8_0
Ministral 3 8B62tok/sQ8_059tok/sQ8_0
Granite 4.2 8B59tok/sQ8_056tok/sQ8_0
Gemma 4 E4B74tok/sQ8_070tok/sQ8_0
LFM2.5 8B-A1B256tok/sQ8_0244tok/sQ8_0
Llama 3.1 8B82tok/sQ6_K78tok/sQ6_K
Qwen2.5 0.5B989tok/sQ8_0942tok/sQ8_0
Qwen3.5 0.8B671tok/sQ8_0639tok/sQ8_0
Qwen3.5 2B402tok/sQ5_K_M383tok/sQ5_K_M
DeepSeek R1 Distill Qwen 1.5B294tok/sQ8_0280tok/sQ8_0
MiniCPM5 2B206tok/sQ8_0196tok/sQ8_0
LFM2.5 2.6B185tok/sQ8_0176tok/sQ8_0
LFM2.5 VL 3B185tok/sQ8_0176tok/sQ8_0
Spark X2.5 4B184tok/smlx-4bit175tok/smlx-4bit
LFM2.5 24B A2B154tok/sQ6_K128tok/sQ8_0
granite vision 4.1 4b146tok/sQ8_0139tok/sQ8_0
Ministral 3 3B Instruct 2512138tok/sQ8_0132tok/sQ8_0
granite 4.2 3b137tok/sQ8_0130tok/sQ8_0
Mellum2.1 12B A2.5B Thinking133tok/sQ8_0127tok/sQ8_0
FrogNano 4B 2609127tok/sQ8_0121tok/sQ8_0
Kimi VL A3B Thinking 2506112tok/sQ8_0106tok/sQ8_0
Kimi VL A3B107tok/smlx-8bit102tok/smlx-8bit
DeepSeek R1 Distill Qwen 7B73tok/sQ8_070tok/sQ8_0
GLM 4 9B 041473tok/smlx-5bit69tok/smlx-5bit
DeepSeek R1 Distill Llama 8B65tok/sQ8_062tok/sQ8_0
Ministral 3 8B Reasoning 251262tok/sQ8_059tok/sQ8_0
GLM 4.6V Flash61tok/sQ8_058tok/sQ8_0
GLM 4.1V 9B Thinking52tok/smlx-8bit49tok/smlx-8bit
Ministral 3 14B Reasoning 251240tok/sQ8_038tok/sQ8_0
DeepSeek R1 Distill Qwen 14B36tok/sQ8_034tok/sQ8_0
Devstral Small 250735tok/sQ5_K_M23tok/sQ8_0
Magistral Small 250735tok/sQ5_K_M23tok/sQ8_0
Magistral Small 250935tok/sQ5_K_M23tok/sQ8_0
Magistral Small 250630tok/sQ6_K23tok/sQ8_0
Qwen3.8 27B OBLITERATED29tok/smlx-4bit28tok/smlx-4bit
granite 4.2 30b27tok/sQ5_K_M18tok/sQ8_0
Muse Glimmer 30B27tok/sQ6_K18tok/sQ8_0
DeepSeek R1 Distill Llama 70Boffload13tok/sQ4_K_M
MiniMax M2.7won't runoffload
Nemotron 3 Super 120B-A12Bwon't runoffload
gpt-oss 120Bwon't runoffload
Devstral 2 123B Instruct 2512won't runoffload
GLM 4.5Vwon't runoffload
GLM 4.6Vwon't runoffload
Kimi Dev 72Bwon't runoffload
Mistral Medium 3.5 128Bwon't runoffload
DeepSeek V4 Flashwon't runwon't run
GLM 5.3 Flashwon't runwon't run
Qwen 3.8 Flash Nextwon't runwon't run
DeepSeek R1won't runwon't run
DeepSeek R1 0528won'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
GLM 4.5won't runwon't run
GLM 4.6won'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 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

Questions

Which is better for local AI, the NVIDIA GeForce RTX 3090 Ti or the NVIDIA RTX 6000 Ada Generation?
Choose the NVIDIA RTX 6000 Ada Generation to run bigger models. Choose the NVIDIA GeForce RTX 3090 Ti for speed.
Which can run bigger models, the NVIDIA GeForce RTX 3090 Ti or the NVIDIA RTX 6000 Ada Generation?
At 8k context the NVIDIA RTX 6000 Ada Generation runs 68 of the models we track fully in memory and the NVIDIA GeForce RTX 3090 Ti 60; only the NVIDIA RTX 6000 Ada Generation holds Qwen 3.5 122B-A10B.
Which is faster for local AI, the NVIDIA GeForce RTX 3090 Ti or the NVIDIA RTX 6000 Ada Generation?
On the models both run fully in memory, the NVIDIA GeForce RTX 3090 Ti is about 1.1× as fast as the NVIDIA RTX 6000 Ada Generation (estimated, median across models).

More comparisons

More hardware for local AI