NVIDIA · consumer

GeForce RTX 4070 GDDR6

GeForce RTX 4070 GDDR6 has 12 GB of VRAM at 480 GB/s — about 11.16 GiB usable after driver and compositor overhead. 1791 of 2118 indexed models fit at 8K context with q4_0 KV.

Spec sheet· bandwidth, theoreticalFrom the file· fit from summed bytesPredicted· speed
Memory
12 GB
GDDR6
Bandwidth
480 GB/s
192-bit bus
Tensor FP16
117 TF
dense
TDP
200 W
$599 MSRP
Context4K8K16K32K64K128K
KV cachef16q8_0q4_0quantizing the KV cache is a ~2× lever on the dominant term at long context
text 1540vision language 149video 14audio asr 39audio tts 21image 2embedding 26

What fits at 8K context

largest quantization that fits, per model · 1791 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
Rocinante-XL-16B-v1Q4_K_L16.1B9.84 GiB0.47 GiB11.16 GiB0.00 GiB33±12.9%
gemma-4-26B-A4B-itMoEQ2_K26.5B10.20 GiB0.17 GiB11.16 GiB0.00 GiB33±12.9%
Wan2.1-FLF2V-14B-720PQ4_116.4B10.32 GiB0.00 GiB11.16 GiB0.00 GiB33±12.9%
v6-Finch-14B-HFQ4_014.1B8.17 GiB2.14 GiB11.16 GiB0.00 GiB33±12.9%
Wan2.1-I2V-14B-480PQ4_116.4B10.32 GiB0.00 GiB11.15 GiB0.01 GiB33±12.9%
Wan2.1-I2V-14B-720PQ4_116.4B10.32 GiB0.00 GiB11.15 GiB0.01 GiB33±12.9%
gemma-4-A4B-98e-v6-coder-itMoEQ3_K_L20.5B10.19 GiB0.17 GiB11.15 GiB0.01 GiB33±12.9%
gemma-4-A4B-98e-v7-coder-itMoEQ3_K_L20.5B10.19 GiB0.17 GiB11.15 GiB0.01 GiB33±12.9%
gemma-4-A4B-98e-v7-coderx-itMoEQ3_K_L20.5B10.19 GiB0.17 GiB11.15 GiB0.01 GiB33±12.9%
Skyfall-31B-v4.2IQ2_XS31.4B9.75 GiB0.47 GiB11.14 GiB0.02 GiB34±12.9%
NVIDIA-Nemotron-Nano-12B-v2Q6_K_L12.3B9.72 GiB0.54 GiB11.14 GiB0.02 GiB34±12.9%
NSFW_13B_sftQ5_013.3B8.54 GiB1.76 GiB11.14 GiB0.02 GiB33±12.9%
MythoMax-L2-Kimiko-v2-13bQ5_K_S13.0B8.54 GiB1.76 GiB11.14 GiB0.02 GiB33±12.9%
MythoMax-L2-13bI1-Q5_K_S13.0B8.54 GiB1.76 GiB11.14 GiB0.02 GiB33±12.9%
CallerIQ2_S32.8B9.67 GiB0.56 GiB11.14 GiB0.02 GiB34±12.9%
Dumpling-Qwen2.5-32BIQ2_S32.8B9.67 GiB0.56 GiB11.14 GiB0.02 GiB34±12.9%
OREAL-32BIQ2_S32.8B9.67 GiB0.56 GiB11.14 GiB0.02 GiB34±12.9%
QwQ-32B-Preview-abliterated-linear25I1-IQ2_S32.8B9.67 GiB0.56 GiB11.14 GiB0.02 GiB34±12.9%
openhands-lm-32b-v0.1I1-IQ2_S32.8B9.67 GiB0.56 GiB11.14 GiB0.02 GiB34±12.9%
Qwen2.5-Coder-32B-abliteratedI1-IQ2_S32.8B9.67 GiB0.56 GiB11.14 GiB0.02 GiB34±12.9%
m1-32bI1-IQ2_S32.8B9.67 GiB0.56 GiB11.14 GiB0.02 GiB34±12.9%
XMainframe-v2-Instruct-32bI1-IQ2_S32.8B9.67 GiB0.56 GiB11.14 GiB0.02 GiB34±12.9%
Qwen2.5-Coder-32B-Python-SpecialistI1-IQ2_S32.8B9.67 GiB0.56 GiB11.14 GiB0.02 GiB34±12.9%
Qwen2.5-32b-RP-InkI1-IQ2_S32.8B9.67 GiB0.56 GiB11.14 GiB0.02 GiB34±12.9%
LongWriter-Zero-32BIQ2_S32.8B9.67 GiB0.56 GiB11.14 GiB0.02 GiB34±12.9%
OpenCodeReasoning-Nemotron-32B-IOIIQ2_S32.8B9.67 GiB0.56 GiB11.14 GiB0.02 GiB34±12.9%
Qwen2.5-Coder-32B-Instruct-abliteratedIQ2_S32.8B9.67 GiB0.56 GiB11.14 GiB0.02 GiB34±12.9%
OlympicCoder-32BIQ2_S32.8B9.67 GiB0.56 GiB11.14 GiB0.02 GiB34±12.9%
OpenCodeReasoning-Nemotron-32BIQ2_S32.8B9.67 GiB0.56 GiB11.14 GiB0.02 GiB34±12.9%
OpenThinker-32BIQ2_S32.8B9.67 GiB0.56 GiB11.14 GiB0.02 GiB34±12.9%
QwQ-32B-ArliAI-RpR-v4IQ2_S32.8B9.67 GiB0.56 GiB11.14 GiB0.02 GiB34±12.9%
Qwen2.5-Coder-32B-InstructIQ2_S32.8B9.67 GiB0.56 GiB11.14 GiB0.02 GiB34±12.9%
Qwen2.5-Coder-32BIQ2_S32.8B9.67 GiB0.56 GiB11.14 GiB0.02 GiB34±12.9%
QwQ-32B-abliteratedIQ2_S32.8B9.67 GiB0.56 GiB11.14 GiB0.02 GiB34±12.9%
DeepSeek-R1-Distill-Qwen-32B-hereticI1-IQ2_S32.8B9.67 GiB0.56 GiB11.14 GiB0.02 GiB34±12.9%
InnoSpark-HPC-RM-32BI1-IQ2_S32.8B9.67 GiB0.56 GiB11.14 GiB0.02 GiB34±12.9%
OpenThinker2-32BIQ2_S32.8B9.67 GiB0.56 GiB11.14 GiB0.02 GiB34±12.9%
INTELLECT-2IQ2_S32.8B9.67 GiB0.56 GiB11.14 GiB0.02 GiB34±12.9%
Qwen2.5-32B-InstructIQ2_S32.8B9.67 GiB0.56 GiB11.14 GiB0.02 GiB34±12.9%
Qwen2.5-Coder-32B-Instruct-UncensoredI1-IQ2_S32.8B9.67 GiB0.56 GiB11.14 GiB0.02 GiB34±12.9%
QwQ-32B-PreviewIQ2_S32.8B9.67 GiB0.56 GiB11.14 GiB0.02 GiB34±12.9%
TinyR1-32B-PreviewIQ2_S32.8B9.67 GiB0.56 GiB11.14 GiB0.02 GiB34±12.9%
deepseek-r1-qwen-2.5-32B-ablatedIQ2_S32.8B9.67 GiB0.56 GiB11.14 GiB0.02 GiB34±12.9%
Rombos-LLM-V2.5-Qwen-32bIQ2_S32.8B9.67 GiB0.56 GiB11.14 GiB0.02 GiB34±12.9%
DeepSeek-R1-Distill-Qwen-32B-abliteratedIQ2_S32.8B9.67 GiB0.56 GiB11.14 GiB0.02 GiB34±12.9%
Qwen2.5-32B-ArliAI-RPMax-v1.3IQ2_S32.8B9.67 GiB0.56 GiB11.14 GiB0.02 GiB34±12.9%
DeepSeek-R1-Distill-Qwen-32BIQ2_S32.8B9.67 GiB0.56 GiB11.14 GiB0.02 GiB34±12.9%
Qwen2.5-VL-32B-InstructIQ2_S33.5B9.67 GiB0.56 GiB11.14 GiB0.02 GiB34±12.9%
EVA-Qwen2.5-32B-v0.2IQ2_S32.8B9.67 GiB0.56 GiB11.14 GiB0.02 GiB34±12.9%
EVA-Qwen2.5-32B-v0.1IQ2_S32.8B9.67 GiB0.56 GiB11.14 GiB0.02 GiB34±12.9%
Ornith-1.0-35BMoEUD-IQ1_M34.7B10.29 GiB0.04 GiB11.13 GiB0.03 GiB181±37%
cogito-v1-preview-qwen-32BI1-IQ2_S32.8B9.67 GiB0.56 GiB11.13 GiB0.03 GiB34±12.9%
QwQ-32B-Snowdrop-v0I1-IQ2_S32.8B9.67 GiB0.56 GiB11.13 GiB0.03 GiB34±12.9%
DeepSeek-R1-Distill-Qwen-32B-UncensoredI1-IQ2_S32.8B9.67 GiB0.56 GiB11.13 GiB0.03 GiB34±12.9%
RoguePlanet-DeepSeek-R1-Qwen-32B-RPI1-IQ2_S32.8B9.67 GiB0.56 GiB11.13 GiB0.03 GiB34±12.9%
ThinkingCap-Qwen3.6-27BIQ2_M27.4B10.13 GiB0.14 GiB11.13 GiB0.03 GiB34±12.9%
Tess-4-27BIQ2_M27.8B10.13 GiB0.14 GiB11.13 GiB0.03 GiB34±12.9%
Qwen3.5-40B-RoughHouse-Claude-4.6-Opus-Polar-Deckard-Uncensored-Heretic-ThinkingI1-IQ1_M39.5B10.05 GiB0.21 GiB11.13 GiB0.03 GiB34±12.9%
Qwen3-30B-A3BMoEUD-IQ2_M30.5B10.12 GiB0.21 GiB11.12 GiB0.04 GiB129±37%
Pantheon-Reasoning-27BI1-Q2_K27.8B10.12 GiB0.14 GiB11.12 GiB0.04 GiB34±12.9%
●From the file◐Predictedwhat these mean

Speed is modeled, not measured: decode is memory-bandwidth bound, so tokens per second is bytes read per token against achievable bandwidth. Mixture-of-experts models carry a wider band because only the routed experts are read each step, and few have been measured publicly.

Questions people ask

What AI models can a GeForce RTX 4070 GDDR6 run?
1791 of 2118 indexed open-weight models fit a GeForce RTX 4070 GDDR6 at 8,192 context with q4_0 KV cache, the largest being Rocinante-XL-16B-v1 at Q4_K_L. That covers text, vision-language, image, video and speech models.
How much usable memory does a GeForce RTX 4070 GDDR6 actually have?
Its nameplate is 12 GB, but about 11.16 GiB is available to a model once driver and compositor overhead is accounted for.
Is a GeForce RTX 4070 GDDR6 fast for local AI?
Its memory bandwidth is 480 GB/s, and that figure — not teraflops — is what governs token generation speed. Capacity decides what you can run; bandwidth decides how fast it runs.