NVIDIA · Ada Lovelace · 2024
NVIDIA GeForce RTX 4080 Super for local AI
A capable 16GB card, but a new RTX 5070 Ti is faster for similar money.
Updated · Speeds marked ~ are estimates; seemethodology
- VRAM
- 16GB
- Bandwidth
- 736 GB/s
- Power
- 320W
- Runs well
- 6 models
RTX 4080 Super
Typically about $1,150 used
Can the RTX 4080 Super run LLMs?
The RTX 4080 Super has 16GB of GDDR6X and 736 GB/s of memory bandwidth, which decides how big a model fits and how fast it generates text. It comfortably runs models up to about 15B parameters: the largest in our catalog is Phi-4 at Q5_K_M (13GB with 8K context, ~52 tokens/sec). Our top all-round pick for it is Qwen3.8 27B (Q3_K_M, ~43 tokens/sec). Llama 3.3 70B Instruct only runs with partial CPU offload, at about 1.4 tokens/sec with CPU offload.
Recommendations
Best models for the RTX 4080 Super
Best overall · also coding, reasoning, vision, largest that fits
Qwen3.8 27BQ3_K_M · 15GB · ~43 tokens/sec
Fits, tightRunner-up
Qwen3.5 9BQ8_0 · 12GB · ~52 tokens/sec
Runs wellFastest good model
Gemma 4 26B A4BQ3_K_M · 14GB · ~130 tokens/sec
Fits, tight
Compatibility
What runs on the RTX 4080 Super
Every current model in our catalog, at the best quantization that fits in 16GB with 8K tokens of context. "Partial offload" means part of the model runs from system RAM (we assume 64GB), which is much slower.
| Model | Parameters | Best quant | Memory | Speed | Verdict |
|---|---|---|---|---|---|
| Gemma 4 E4B | 8B | Q8_0 | 10 GB | ~59 tok/s | Runs well |
| Qwen3.5 9B | 9B | Q8_0 | 12 GB | ~52 tok/s | Runs well |
| Phi-4 | 14.7B | Q5_K_M | 13 GB | ~52 tok/s | Runs well |
| gpt-oss-20b | 20.9B (3.6B active) | MXFP4 | 15 GB | ~110 tok/s | Fits, tight |
| Gemma 4 26B A4B | 26B (3.8B active) | Q3_K_M | 14 GB | ~130 tok/s | Fits, tight |
| Qwen3.8 27B | 27B | Q3_K_M | 15 GB | ~43 tok/s | Fits, tight |
| Gemma 4 31B | 31B | Q4_K_M | 22 GB | ~5.6 tok/s | Partial offload |
| Qwen3.6 35B-A3B | 35B (3B active) | Q4_K_M | 24 GB | ~46 tok/s | Partial offload |
| GLM-4.5-Air | 106B (12B active) | — | — | — | Won't fitWhat it needs |
| Llama 4 Scout | 109B (17B active) | — | — | — | Won't fitWhat it needs |
| gpt-oss-120b | 117B (5.1B active) | — | — | — | Won't fitWhat it needs |
| Qwen3.5 122B-A10B | 122B (10B active) | — | — | — | Won't fitWhat it needs |
| DeepSeek V4 Flash | 284B (13B active) | — | — | — | Won't fitWhat it needs |
| GLM-5.3-Flash | 320B (18B active) | — | — | — | Won't fitWhat it needs |
| GLM-5.3 | 744B (40B active) | — | — | — | Won't fitWhat it needs |
Performance
RTX 4080 Super vs. alternatives on Qwen3.8 27B
Generation speed in tokens per second for a single user. Token generation is limited mostly by memory bandwidth, which is why bandwidth predicts these results so well.
RTX 5080
Q3_K_M · 16GB
~55tok/s
RTX 4080 Super
Q3_K_M · 16GB
~43tok/s
RTX 4070 Ti Super
Q3_K_M · 16GB
~39tok/s
RX 7900 XTX
Q4_K_M · 24GB
~36tok/s
Measured
RTX 4080 Super benchmark results
| Model | Quant | Runtime | Generation | Prompt | Source |
|---|---|---|---|---|---|
| Llama 2 7B Chat | Q4_0 | llama.cpp (CUDA) | 150 tok/s | 8125 tok/s | llama.cpp discussion #15013: Performance of llama.cpp on NVIDIA CUDA · October 2026 |
| Llama 2 7B Chat | Q4_0 | llama.cpp (Vulkan) | 145 tok/s | 7100 tok/s | llama.cpp discussion #10879: Performance of llama.cpp with Vulkan · October 2026 |
Our take
Should you buy the RTX 4080 Super for AI?
The RTX 4080 Super has 16GB of GDDR6X on a 256-bit bus for 736 GB/s. It runs 14B models at 8 bit and 24B models at 4 bit, the same range as other 16GB cards. Used cards sold for about $1,150 in October 2026, close to the price of a new RTX 5070 Ti that has 22 percent more bandwidth and a full warranty. Buy one only if it is clearly cheaper than a 5070 Ti.
What we like
- 16GB with solid 736 GB/s bandwidth
- Mature drivers and CUDA support
- Widely available on the used market
Watch out for
- Used prices close to a new RTX 5070 Ti
- 16GB limits model size
- 320W board power
Best for: Used-market buyers who find a good deal · Gaming PCs that also run 8B to 24B models
Specifications
RTX 4080 Super specs that matter for AI
- Memory
- 16GB GDDR6X
- Memory bandwidth
- 736 GB/s
- Memory bus
- 256-bit
- Architecture
- Ada Lovelace
- Power
- 320W
- Interface
- PCIe 4.0 x16
- Released
- 2024
- Launch price
- $999
- Street price (used)
- ~$1,150 (2026-10)
- Software
- CUDA, Vulkan
Cross-shop
Alternatives to the RTX 4080 Super
Frequently asked questions
How much VRAM does the RTX 4080 Super have?
The NVIDIA GeForce RTX 4080 Super has 16GB of GDDR6X with 736 GB/s of memory bandwidth on a 256-bit bus.
What is the largest LLM the RTX 4080 Super can run?
Fully in GPU memory, the largest model in our catalog that runs comfortably is Phi-4 (14.7B) at Q5_K_M, using about 13GB with 8K context at ~52 tokens/sec.
Can the RTX 4080 Super run Llama 3.3 70B?
Only with partial CPU offload. At Q4_K_M the model needs about 47GB with 8K context, so part of it spills into system RAM and speed drops to about 1.4 tokens/sec with CPU offload. For usable 70B speeds you want 48GB of VRAM or more (for example two 24GB cards).
What is the best LLM to run on the RTX 4080 Super?
Our top pick is Qwen3.8 27B at Q3_K_M, which fits in 16GB with room for 8K of context and runs at ~43 tokens/sec. See our full list of the best models for the RTX 4080 Super for coding, reasoning, and vision picks.
Is the RTX 4080 Super good for local AI in 2026?
A capable 16GB card, but a new RTX 5070 Ti is faster for similar money.