BGE-Large EN v1.5 from BAAI is the reference English embedding model for RAG pipelines — produces 1024-dim vectors, scores near the top of MTEB retrieval benchmarks. The tiny 335 M footprint runs on CPU comfortably or 100k+ embeddings/sec on a single RTX 4090. The 512-token context cap is the main limitation versus Nomic Embed v1.5.
BGE-Large EN v1.5 needs at minimum 0.2 GB of VRAM at Q4_K_M quantization (0.7 GB at FP16). The cheapest GPU that comfortably fits with KV-cache headroom is the Intel Arc A380 6GB (6 GB VRAM, $139 MSRP). Measured throughput hits 23500 emb/s on NVIDIA B200 192GB. This model fits a single consumer GPU under 48 GB, so a one-card build works.
Weights only. Add ~20-30% for KV cache at typical context lengths.
| Quant | VRAM | Quality | Notes |
|---|---|---|---|
| FP16 | 0.7 GB | Reference | Training-precision weights |
| Q8_0 | 0.4 GB | Near-lossless | 8-bit, ~0.1% perplexity hit |
| Q6_K | 0.3 GB | Very high | 6-bit, near Q8 quality |
| Q5_K_M | 0.3 GB | High | Strong middle ground |
| Q4_K_M | 0.2 GB | Balanced (recommended) | Default for local deployments |
| Q4_0 | 0.2 GB | Legacy 4-bit | Older GGUF, kept for compatibility |
| Q3_K_M | 0.2 GB | Lossy | When VRAM is very tight |
| Q2_K | 0.1 GB | Extreme | Rescue option, quality degrades visibly |
Filtered from MyAIHardware's GPU database. FP16 needs 0.7 GB; Q4_K_M needs 0.2 GB.
31 measurement(s) for this model from MyAIHardware's benchmark database.
| Device | Speed | Quant | Context | Power |
|---|---|---|---|---|
| NVIDIA B200 192GB | 23500 emb/s | FP16 | 512 | 1000W |
| NVIDIA H200 141GB | 15200 emb/s | FP16 | 512 | 700W |
| AMD Instinct MI300X 192GB | 14800 emb/s | FP16 | 512 | 750W |
| NVIDIA H100 SXM5 80GB | 12400 emb/s | FP16 | 512 | 700W |
| NVIDIA GeForce RTX 5090 32GB | 12200 emb/s | FP16 | 512 | 575W |
| NVIDIA L40S 48GB | 10500 emb/s | FP16 | 512 | 350W |
| AMD Instinct MI300X 192GB | 9800 emb/s | FP16 | 512 | 750W |
| AMD Instinct MI300X 192GB | 9600 emb/s | FP16 | 512 | 750W |
| NVIDIA A100 SXM4 80GB | 9500 emb/s | FP16 | 512 | 400W |
| NVIDIA A100 80GB SXM | 9100 emb/s | FP16 | 512 | 400W |
| NVIDIA GeForce RTX 5090 32GB | 9100 emb/s | FP16 | 512 | 575W |
| NVIDIA GeForce RTX 4090 24GB | 8500 emb/s | FP16 | 512 | 450W |
| NVIDIA GeForce RTX 5090 32GB | 8200 emb/s | FP16 | 512 | 575W |
| NVIDIA GeForce RTX 5080 16GB | 8200 emb/s | FP16 | 512 | 360W |
| NVIDIA RTX 6000 Ada 48GB | 8200 emb/s | FP16 | 512 | 300W |
| NVIDIA L40S 48GB | 7400 emb/s | FP16 | 512 | 350W |
| NVIDIA GeForce RTX 5080 16GB | 7200 emb/s | FP16 | 512 | 360W |
| NVIDIA GeForce RTX 5070 Ti 16GB | 7100 emb/s | FP16 | 512 | 300W |
| NVIDIA GeForce RTX 4080 Super 16GB | 6900 emb/s | FP16 | 512 | 320W |
| NVIDIA GeForce RTX 5080 16GB | 6800 emb/s | FP16 | 512 | 360W |
Official weights + popular runtime tags.
ollama pull bge-largeStep-by-step guides that use this model.
Models in a similar size or capability class.
Use these in papers, blog posts, or internal docs.
@misc{bge-large-en-2023,
title={BGE-Large EN v1.5},
author={BAAI},
year={2023},
url={https://huggingface.co/BAAI/bge-large-en-v1.5}
}BAAI (2023). BGE-Large EN v1.5 [Model card]. Hugging Face. https://huggingface.co/BAAI/bge-large-en-v1.5
BGE-Large EN v1.5 (BGE, BAAI, 2023). Available at https://huggingface.co/BAAI/bge-large-en-v1.5.