Skip to content
Llama Builds

NVIDIA · Blackwell · 2025

NVIDIA GeForce RTX 5080 for local AI

Fast, but 16GB is the limit, and the RTX 5070 Ti runs the same models nearly as quickly for less.

Updated · Speeds marked ~ are estimates; seemethodology

VRAM
16GB
Bandwidth
960 GB/s
Power
360W
Runs well
6 models

RTX 5080

Typically about $1,600

Check price at AmazonShop used on eBay

Can the RTX 5080 run LLMs?

The RTX 5080 has 16GB of GDDR7 and 960 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, ~67 tokens/sec). Our top all-round pick for it is Qwen3.8 27B (Q3_K_M, ~55 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 5080

  1. Best overall · also coding, reasoning, vision, largest that fits

    Qwen3.8 27B

    Q3_K_M · 15GB · ~55 tokens/sec

    Fits, tight

  2. Runner-up

    Qwen3.5 9B

    Q8_0 · 12GB · ~66 tokens/sec

    Runs well

  3. Fastest good model

    Gemma 4 26B A4B

    Q3_K_M · 14GB · ~155 tokens/sec

    Fits, tight
Read: the best LLMs to run on the RTX 5080

Compatibility

What runs on the RTX 5080

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.

Which open-weight models run on the NVIDIA GeForce RTX 5080, with the best quantization and estimated speed
ModelParametersBest quantMemorySpeedVerdict
Gemma 4 E4B8BQ8_010 GB~75 tok/sRuns well
Qwen3.5 9B9BQ8_012 GB~66 tok/sRuns well
Phi-414.7BQ5_K_M13 GB~67 tok/sRuns well
gpt-oss-20b20.9B (3.6B active)MXFP415 GB~130 tok/sFits, tight
Gemma 4 26B A4B26B (3.8B active)Q3_K_M14 GB~155 tok/sFits, tight
Qwen3.8 27B27BQ3_K_M15 GB~55 tok/sFits, tight
Gemma 4 31B31BQ4_K_M22 GB~5.8 tok/sPartial offload
Qwen3.6 35B-A3B35B (3B active)Q4_K_M24 GB~48 tok/sPartial offload
GLM-4.5-Air106B (12B active)———Won't fitWhat it needs
Llama 4 Scout109B (17B active)———Won't fitWhat it needs
gpt-oss-120b117B (5.1B active)———Won't fitWhat it needs
Qwen3.5 122B-A10B122B (10B active)———Won't fitWhat it needs
DeepSeek V4 Flash284B (13B active)———Won't fitWhat it needs
GLM-5.3-Flash320B (18B active)———Won't fitWhat it needs
GLM-5.3744B (40B active)———Won't fitWhat it needs

Performance

RTX 5080 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.

  1. RTX 5090

    Q6_K · 32GB

    ~59tok/s

  2. RTX 5080

    Q3_K_M · 16GB

    ~55tok/s

  3. RTX 5070 Ti

    Q3_K_M · 16GB

    ~51tok/s

  4. RTX 4080 Super

    Q3_K_M · 16GB

    ~43tok/s

Estimated (method)

Measured

RTX 5080 benchmark results

ModelQuantRuntimeGenerationPromptSource
Llama 2 7B ChatQ4_0llama.cpp (Vulkan)185 tok/s7445 tok/sllama.cpp discussion #10879: Performance of llama.cpp with Vulkan · October 2026
Llama 2 7B ChatQ4_0llama.cpp (CUDA)180 tok/s8295 tok/sllama.cpp discussion #15013: Performance of llama.cpp on NVIDIA CUDA · October 2026

Our take

Should you buy the RTX 5080 for AI?

The RTX 5080 uses the GB203 chip with 16GB of GDDR7 on a 256-bit bus for 960 GB/s. For local LLMs that is only about 7 percent more bandwidth than the RTX 5070 Ti, and both cards top out at the same 16GB, so they run the same set of models. It handles 14B models at 8 bit and 24B models at 4 bit, while 30B-class MoE models need some CPU offload. At about $1,600 new in October 2026 it is hard to justify over a 5070 Ti or a used 24GB card.

What we like

  • 960 GB/s GDDR7 is quick for 8B to 24B models
  • Mature CUDA software support
  • PCIe 5.0 x16 and FP4 tensor support

Watch out for

  • 16GB is the same capacity as much cheaper cards
  • About $1,600 new, well above its $999 MSRP
  • Little LLM speed gain over the RTX 5070 Ti

Best for: Gaming PCs that also run 8B to 24B models · Single-card setups where a 5090 is out of budget

Specifications

RTX 5080 specs that matter for AI

Memory
16GB GDDR7
Memory bandwidth
960 GB/s
Memory bus
256-bit
Architecture
Blackwell
Power
360W
Interface
PCIe 5.0 x16
Released
2025
Launch price
$999
Street price (new)
~$1,600 (2026-10)
Street price (used)
~$1,400 (2026-10)
Software
CUDA, Vulkan

Sources: NVIDIA GeForce RTX 5080 product page

Cross-shop

Alternatives to the RTX 5080

NVIDIA2025
16GB

GDDR7 · 896 GB/s

RTX 5070 Ti

~$1,200

NVIDIA2025
32GB

GDDR7 · 1,792 GB/s

RTX 5090

~$4,300

NVIDIA2024
16GB

GDDR6X · 736 GB/s

RTX 4080 Super

~$1,150 used

Frequently asked questions

How much VRAM does the RTX 5080 have?

The NVIDIA GeForce RTX 5080 has 16GB of GDDR7 with 960 GB/s of memory bandwidth on a 256-bit bus.

What is the largest LLM the RTX 5080 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 ~67 tokens/sec.

Can the RTX 5080 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 5080?

Our top pick is Qwen3.8 27B at Q3_K_M, which fits in 16GB with room for 8K of context and runs at ~55 tokens/sec. See our full list of the best models for the RTX 5080 for coding, reasoning, and vision picks.

Is the RTX 5080 good for local AI in 2026?

Fast, but 16GB is the limit, and the RTX 5070 Ti runs the same models nearly as quickly for less.