How much VRAM does the NVIDIA RTX 4070 have?

The NVIDIA RTX 4070 has 12 GB of GDDR6X with 504 GB/s memory bandwidth.

What is the NVIDIA RTX 4070 best for?

With 12 GB of VRAM, the NVIDIA RTX 4070 comfortably handles 7B–8B models and can stretch to some 13B–14B models at aggressive quantization, a solid entry point into local LLM inference.

What LLMs can the NVIDIA RTX 4070 run locally?

The NVIDIA RTX 4070 can run 30 of the 97 open-weight models tracked by CanItRun natively in VRAM at 8k context. Top options include: Ornith 1.5 9B at Q8_0, Gemma 4 26B (MoE) at Q2_K, Qwen 3.5 9B at Q8_0.

Can the NVIDIA RTX 4070 run Gemma 4 31B?

The NVIDIA RTX 4070 can run Gemma 4 31B with CPU offload at Q6_K quantization, but inference will be slower than native VRAM execution.

Can the NVIDIA RTX 4070 run Qwen 3.6 27B?

The NVIDIA RTX 4070 can run Qwen 3.6 27B with CPU offload at Q8_0 quantization, but inference will be slower than native VRAM execution.

Can the NVIDIA RTX 4070 run Qwen3 8B?

Yes. The NVIDIA RTX 4070 runs Qwen3 8B natively in VRAM at Q8_0 quantization, achieving approximately 33.7 tokens per second.

NVIDIA RTX 4070

The NVIDIA RTX 4070 has 12 GB VRAM and 504 GB/s memory bandwidth. It can run 30 of our 97 tracked models natively in VRAM at 8k context.

With 12 GB GDDR6X, the NVIDIA RTX 4070 is a consumer-tier GPU that can run 30 models natively. This site's calculator computes decode speed from VRAM capacity and memory bandwidth, not CUDA core count, so this card returns the exact same fit and tok/s results as the pricier RTX 4070 Ti and RTX 4070 SUPER for every model this site tracks: Qwen3 8B fits natively at Q8_0 (10.88 GB, 33.7 tok/s), Qwen 2.5 7B fits at Q8_0 (9.57 GB, 38.3 tok/s), and Qwen3 14B fits at its recommended Q3_K_M (9.48 GB, 38.7 tok/s). The 12GB ceiling, shared by all three cards, is what actually limits this GPU: GPT-OSS 20B needs CPU offload (25.23 GB, 3.6 tok/s) and Qwen 3.6 27B fares worse (32.75 GB, 1.3 tok/s), the identical results the $200-more RTX 4070 Ti produces for both models. Buying the Ti or the SUPER over this card changes nothing about what fits or how fast it decodes; it only changes gaming frame rates and prompt-processing speed, neither of which this site models.

The NVIDIA RTX 4070 launched April 13, 2023 at a $599 MSRP, sharing the RTX 4070 Ti's AD104 die but in a more cut-down configuration: AD104-250 (still 35.8 billion transistors) with 5,888 CUDA cores, 23.3% fewer than the Ti's 7,680, at a 200W TDP against the Ti's 285W. Memory is untouched: the identical 12GB GDDR6X on a 192-bit bus at 504 GB/s the RTX 4070 Ti and RTX 4070 SUPER also carry. Since this site's decode-speed model runs on VRAM capacity and memory bandwidth rather than core count, it returns the exact same tok/s and VRAM fit results as those two pricier cards for every tracked model: the fewer cores here mean lower gaming frame rates and slower prompt processing, not slower token generation.

NVIDIA RTX 4070: Launched April 13, 2023 at a $599 MSRP, the RTX 4070 shares the RTX 4070 Ti's AD104 die but in a more cut-down configuration, AD104-250 (still 35.8 billion transistors on the same 295mm² die) with 5,888 CUDA cores and 46 SMs, 23.3% fewer cores than the Ti's 7,680 (TechPowerUp's spec database; WCCFTech's contemporaneous leak coverage matches every figure here). That smaller cut also drops power draw further than it drops cores: a 200W TDP versus the Ti's 285W, a 30% cut TechPowerUp's own review credited with RTX 3080-class gaming performance at nearly 40% less power than the 3080's 320W. Memory is untouched: 12GB GDDR6X on a 192-bit bus at 504 GB/s, identical to the RTX 4070 Ti and RTX 4070 SUPER on either side of it in the lineup.

This site's calculator computes decode speed from VRAM capacity and memory bandwidth, not CUDA core count, so this card returns the exact same fit and tok/s results as the pricier RTX 4070 Ti and RTX 4070 SUPER for every model this site tracks: Qwen3 8B fits natively at Q8_0 (10.88 GB, 33.7 tok/s), Qwen 2.5 7B fits at Q8_0 (9.57 GB, 38.3 tok/s), and Qwen3 14B fits at its recommended Q3_K_M (9.48 GB, 38.7 tok/s). The 12GB ceiling, shared by all three cards, is what actually limits this GPU: GPT-OSS 20B needs CPU offload (25.23 GB, 3.6 tok/s) and Qwen 3.6 27B fares worse (32.75 GB, 1.3 tok/s), the identical results the $200-more RTX 4070 Ti produces for both models. Buying the Ti or the SUPER over this card changes nothing about what fits or how fast it decodes; it only changes gaming frame rates and prompt-processing speed, neither of which this site models.

Full CUDA support on Ada Lovelace, the same mature llama.cpp, Ollama, vLLM, and TensorRT-LLM coverage every Ada card in this family gets. Sharing its 12GB/504 GB/s memory subsystem with the RTX 4070 Ti and RTX 4070 SUPER means this card can't run NVFP4 either (Blackwell-only in this site's calculator); Q2_K remains the floor on the standard quant ladder for all three. This is the cheapest way into that shared 12GB ceiling in the Ada 4070 family, and for local LLM inference specifically, that's the whole story: the RTX 4070 Ti's extra 1,792 CUDA cores and the RTX 4070 SUPER's extra 1,280 buy faster gaming and prompt processing, not a single extra token per second of decode speed on this site's calculator. Anyone whose models actually need more than 12GB should look at the RTX 4070 Ti SUPER instead, the only card in the family whose memory subsystem changed at all.

Vendor	NVIDIA
Architecture	Ada Lovelace
VRAM	12 GB
Memory type	GDDR6X
Memory bandwidth	504 GB/s
Compute backend	CUDA
Tier	Consumer
Released	2023
Models (native)	30 / 97
Models (offload)	27 / 97

Software: Full llama.cpp and Ollama support out of the box. CUDA 12.x recommended; driver ≥ 525 required.

Popular models for this GPU

Bonsai 27B Gemma 4 26B (MoE)Qwen3 14B Qwen 2.5 14B Instruct Phi-4 14B Instruct

Models this GPU runs natively in VRAM (30)

Show 25 more

Models that fit with CPU offload (27)

These use system RAM for layers that don't fit in VRAM, so expect much slower inference.

Too large for this GPU (40)

Compare NVIDIA RTX 4070 with other GPUs

Continue reading

vram-guides8 min

Best LLMs for 12 GB VRAM (2026)

hardware10 min

NVIDIA RTX 40 Series for LLMs: Complete VRAM & Performance Guide

Frequently asked questions

How much VRAM does the NVIDIA RTX 4070 have?: The NVIDIA RTX 4070 has 12 GB of GDDR6X with 504 GB/s memory bandwidth.
What is the NVIDIA RTX 4070 best for?: With 12 GB of VRAM, the NVIDIA RTX 4070 comfortably handles 7B–8B models and can stretch to some 13B–14B models at aggressive quantization, a solid entry point into local LLM inference.
What LLMs can the NVIDIA RTX 4070 run locally?: The NVIDIA RTX 4070 can run 30 of the 97 open-weight models tracked by CanItRun natively in VRAM at 8k context. Top options include: Ornith 1.5 9B at Q8_0, Gemma 4 26B (MoE) at Q2_K, Qwen 3.5 9B at Q8_0.
Can the NVIDIA RTX 4070 run Gemma 4 31B?: The NVIDIA RTX 4070 can run Gemma 4 31B with CPU offload at Q6_K quantization, but inference will be slower than native VRAM execution.
Can the NVIDIA RTX 4070 run Qwen 3.6 27B?: The NVIDIA RTX 4070 can run Qwen 3.6 27B with CPU offload at Q8_0 quantization, but inference will be slower than native VRAM execution.
Can the NVIDIA RTX 4070 run Qwen3 8B?: Yes. The NVIDIA RTX 4070 runs Qwen3 8B natively in VRAM at Q8_0 quantization, achieving approximately 33.7 tokens per second.