NVIDIA
Operating mode
Use this to bias workload recommendations toward responsiveness, background autonomy, lighter serving, or multi-GPU scale-out.
Current mode
Balanced
Balanced for general local use. Keeps the ranking neutral across personal and serving workflows.
The NVIDIA L40 is the first Ada Lovelace 48 GB datacenter GPU, released ahead of the L40S, combining 48 GB of GDDR6 with Ada Tensor Cores in a PCIe form factor. It was positioned for visualization, rendering, and inference workloads. The L40S later superseded it for pure inference with higher INT8 throughput and better compute balance. The L40 remains a capable option for teams running mixed visualization and inference workloads, with 90 TFLOPS FP16 and 720 INT8 TOPS in a 300W envelope. It handles 30B models at Q4 and 13B models at FP16.
Beyond LLMs
What AI tasks this GPU can handle — from text generation to image and video creation.
| Capability | Status | Representative Model |
|---|---|---|
| LLM Chat (7B) | Runs natively | Llama 3.1 8B Q4 |
| LLM Coding (30B) | Runs natively | Qwen 3 30B Q4 |
| LLM Large (70B) | Needs offload | Llama 3.1 70B Q4 |
| Image Gen (SDXL) | Runs natively | SDXL 1.0 FP16 |
| Image Gen (Flux) | Runs natively | Flux.1 Dev FP16 |
| Image Gen (SD 3.5) | Runs natively | SD 3.5 Large FP16 |
| Video Short (25f) | Runs natively | LTX Video 2B |
| Video Long (100f) | Won't fit | Wan Video 14B |
Architecture
Ada Lovelace is NVIDIA's fourth-generation RTX architecture, manufactured on TSMC's custom 4N process. It introduces 4th-generation Tensor Cores with FP8 support, 3rd-generation ray tracing cores, and the Shader Execution Reordering (SER) engine for improved workload scheduling.
AI Relevance
FP8 Tensor Core operations provide a significant uplift for quantized LLM inference compared to Ampere's FP16-only Tensor Cores. DLSS 3 Frame Generation demonstrates the architecture's AI processing capabilities.
Consejo de compra
Excelente opción para IA local
Ejecuta 28 de 50 modelos principales bien — un todoterreno sólido para inferencia local.
48.0 GB
VRAM
$5,500
PVP
$115/GB
Coste por GB de VRAM
Mejores modelos para esta GPU
What will limit you first
Este setup está bastante equilibrado para este modelo.
No hay grandes señales de alerta
Esta recomendación tiene margen de memoria suficiente y una velocidad estimada razonable para la carga de trabajo seleccionada.
Best upgrade itinerary
Desbloquea 5 modelos adicionales que hoy no caben en tu setup.
¿Quieres más margen? AMD Instinct MI210 64GB (64.0 GB VRAM) es el siguiente paso.
Chat
SQwen 3.5 27B matches Chat and keeps a practical fit profile. It is a recent-generation family, which helps on current local SOTA workloads. It fits natively with comfortable headroom. Context coverage stays within the requested workload envelope. Known distribution channels: huggingface, ollama, lm-studio.
Coding
SQwen 3.6 27B is a specialized fit for Coding. It is a recent-generation family, which helps on current local SOTA workloads. It fits natively with comfortable headroom. Context coverage stays within the requested workload envelope. Known distribution channels: huggingface, lm-studio.
Agentic Coding
SQwen 3.6 27B is a specialized fit for Agentic Coding. It is a recent-generation family, which helps on current local SOTA workloads. It fits natively with comfortable headroom. Context coverage stays within the requested workload envelope. Known distribution channels: huggingface, lm-studio.
Reasoning
SDevstral Small 2 24B Instruct matches Reasoning and keeps a practical fit profile. It is a recent-generation family, which helps on current local SOTA workloads. It fits natively with comfortable headroom. Context coverage stays within the requested workload envelope. Known distribution channels: huggingface, ollama, lm-studio.
RAG
SQwen 3.5 27B matches RAG and keeps a practical fit profile. It is a recent-generation family, which helps on current local SOTA workloads. It fits natively with comfortable headroom. Context coverage stays within the requested workload envelope. Known distribution channels: huggingface, ollama, lm-studio.
Qwen 3.6 35B A3BQwen 3.5 35B A3BQwen3-Coder 30B A3B InstructQwen3-VL 30B A3B InstructQwen 3 30B A3BQwen 3.5 27BNemotron Cascade 2 30B A3BQwen 3 32B
Magistral Small 2507Devstral Small 2 24B InstructQwen 3 14B
GPT-OSS 20BQwen 3.6 27BQwen 3.5 9BNemotron 3 Nano 30B
Phi-4-reasoning-plus 14BDevstral Small 1.1
Gemma 4 31BQwen 3 8B
EXAONE 4.0 32BQwen 3.5 4BMinistral 3 14BNemotron Nano 8BPhi-4 Mini Reasoning 4BQwen3-Coder-NextQwen 2.5 VL 72B
Jina Embeddings v3
BGE M3Qwen 3.5 397B A17BDevstral 2 123B Instruct
Kimi K2.5Kimi K2.6
DeepSeek V4 ProQwen 3.5 122B A10BDeepSeek V4 FlashMistral Small 4 119B
Command A 111BGPT-OSS 120B
GLM-5.2GLM-5.1Pixtral Large 124BGLM-5DeepSeek V3.2Qwen 3 235B A22BQwen3-Coder 480B A35B InstructLeanstral 119B A6BDeepSeek Coder V2 236BDeepSeek R1 671BDeepSeek V3.1 671BCasi al alcance
Modelos de alta calidad que necesitan un poco más de memoria
Image & Video Generation
54 of 56 models can generate images or video on your NVIDIA L40 48GB
| Model | Max Resolution | Gen Time | Grade |
|---|---|---|---|
| SD TurboImage | 512×512 | 400ms | S |
| Stable Diffusion 1.5Image | 512×768 | 900ms | S |
| Realistic Vision v5.1Image | 512×768 | 900ms | S |
| DreamShaper 8Image | 512×768 | 900ms | S |
| LCM DreamShaper v7Image | 512×768 | 300ms | S |
| PixArt-SigmaImage | 1024×1024 | ~3.6s | S |
| FramePack I2VVideo | 640×480 | ~11.3s/frame | S |
| SDXL TurboImage | 512×512 | 400ms | S |
| SDXL LightningImage | 1024×1024 | ~1.3s | S |
| Stable Diffusion XL 1.0Image | 1024×1024 | ~3.6s | S |
| Playground v2.5Image | 1024×1024 | ~5.3s | S |
| RealVisXL v5.0Image | 1024×1024 | ~4s | S |
| DreamShaper XLImage | 1024×1024 | ~4s | S |
| Juggernaut XL v9Image | 1024×1024 | ~4s | S |
| Animagine XL 3.1Image | 1024×1024 | ~4s | S |
| Pony Diffusion V6 XLImage | 1024×1024 | ~4s | S |
| Animagine XL 4.0Image | 1024×1024 | ~4s | S |
| Illustrious XLImage | 1024×1024 | ~4s | S |
| Wan Video 2.1 1.3BVideo | 480×832 | ~2.6s/frame | S |
| Stable Diffusion 3.5 MediumImage | 1024×1024 | ~6.2s | S |
| Flux.2 Klein 4BImage | 1024×1024 | ~1.1s | S |
| LTX Video 2BVideo | 1280×720 | ~3.1s/frame | S |
| KolorsImage | 1024×1024 | ~7.1s | S |
| Stable CascadeImage | 1024×1024 | ~8.9s | S |
| AuraFlow v0.3Image | 1536×1536 | ~16s | S |
| Stable Diffusion 3.5 LargeImage | 1024×1024 | ~19.5s | S |
| Stable Diffusion 3.5 Large TurboImage | 1024×1024 | ~3.6s | S |
| CogVideoX 2BVideo | 720×480 | ~3.1s/frame | S |
| HunyuanVideoVideo | 256×256 | ~11.3s/frame | S |
| ChromaImage | 1024×1024 | ~3.6s | S |
| Z-Image TurboImage | 1536×1536 | ~3.7s | S |
| Flux.1 DevImage | 1024×1024 | ~16s | S |
| Flux.1 SchnellImage | 1024×1024 | ~3.1s | S |
| LTX Video 13BVideo | 768×512 | ~6.5s/frame | S |
| Flux.1 Kontext DevImage | 1024×1024 | ~17.8s | S |
| AnimateDiff v1.5.3Video | 512×768 | ~1.6s/frame | S |
| Cosmos Diffusion 7BVideo | 1024×576 | ~5.1s/frame | S |
| CogVideoX 5BVideo | 720×480 | ~4.4s/frame | S |
| Wan2.2 TI2V 5BVideo | 832×480 | ~4.4s/frame | S |
| Flux.2 Klein 9BImage | 1024×1024 | ~1.8s | S |
| Flux.1 Fill DevImage | 1024×1024 | ~15.1s | S |
| Krea 2Image | 1024×1024 | ~4.8s | S |
| Sulphur 2Video | 1280×720 | ~5.6s/frame | S |
| Ideogram 4Image | 2048×2048 | ~4.4s | S |
| Mochi 1 PreviewVideo | 848×480 | ~5.9s/frame | S |
| HunyuanVideo 1.5Video | 720×1280 | ~5.4s/frame | A |
| Helios 14BVideo | 832×480 | ~6.7s/frame | B |
| SkyReels V2 14BVideo | 256×256 | ~6.7s/frame | B |
| Wan Video 2.1 14BVideo | 256×256 | ~11.5s/frame | D |
| Wan Video 2.2 14BVideo | 256×256 | ~11.5s/frame | D |
| Qwen ImageImage | 256×256 | ~9.8s | D |
| Qwen Image EditImage | 256×256 | ~9.8s | D |
| LTX-2 22BVideo | 256×256 | ~8s/frame | D |
| Flux.2 DevImage | 256×256 | ~2m 48s | D |
| MAGI-1Video | 256×256 | ~8.3s/frame | F |
| HunyuanImage 3.0Image | 256×256 | ~10.5s | F |
Image models estimated at 1024×1024 (28 steps, FP16). Video models estimated at 768×512 (25 frames, 30 steps, FP16). Actual performance varies with runtime and system load.
Multi-GPU scaling
Scale out with multiple GPUs for larger models. PCIe interconnect with 25% scaling overhead.
| Config | Effective memory | Models that fit | Est. bandwidth |
|---|---|---|---|
| 1× NVIDIA | 48 GB | 343/380 | 864 GB/s |
| 2× NVIDIA | 96 GB | 356/380 | 1,296 GB/s |
Model counts use default quantization at coding workload settings. Multi-GPU scaling factor: 0.75× per additional GPU.
Upgrade paths
See what you unlock with more powerful hardware
Opciones de mejora
2× NVIDIA L40 48GBMulti-GPUDesbloquea 13 modelos adicionales que hoy no caben en tu setup.
Eleva la velocidad media de decodificación en torno a un 13% en los modelos que sí caben.
El scale-out solo compensa si la plataforma host tiene suficientes PCIe lanes, separación entre slots, potencia y refrigeración.
Cuanto más grande es el setup, más importa el runtime. En multi-GPU y serving para varios usuarios es donde vLLM, SGLang, TGI, TensorRT-LLM o un llama.cpp afinado empiezan a justificar su complejidad.
~$5,500 MSRP
Desbloquea 5 modelos adicionales que hoy no caben en tu setup.
Eleva la velocidad media de decodificación en torno a un 15% en los modelos que sí caben.
~$10,000 MSRP
NVIDIA A100 80GBMejora NVIDIADesbloquea 12 modelos adicionales que hoy no caben en tu setup.
Eleva la velocidad media de decodificación en torno a un 37% en los modelos que sí caben.
~$15,000 MSRP
Desbloquea 13 modelos adicionales que hoy no caben en tu setup.
~$2,499 MSRP
Desbloquea 26 modelos adicionales que hoy no caben en tu setup.
Eleva la velocidad media de decodificación en torno a un 127% en los modelos que sí caben.
~$8,000 MSRP
NVIDIA L40 48GB (48 GB VRAM) can run these top models: Qwen 3.6 35B A3B (score: 98/100), Qwen 3.5 35B A3B (score: 96/100), Qwen3-Coder 30B A3B Instruct (score: 96/100). See the full compatibility list above.
NVIDIA L40 48GB has 48 GB of VRAM available for AI model inference. This determines which models and quantization levels you can run locally.
Yes, NVIDIA L40 48GB is excellent for running LLMs locally with top compatibility scores above 80/100.
For coding on NVIDIA L40 48GB, we recommend Qwen 3.6 27B. It achieves 24.7 tokens per second with 262K context window. Qwen 3.6 27B is a specialized fit for Coding. It is a recent-generation family, which helps on current local SOTA workloads. It fits natively with comfortable headroom. Context coverage stays within the requested workload envelope. Known distribution channels: huggingface, lm-studio.
There are 5 upgrade path(s) from NVIDIA L40 48GB: NVIDIA L40 48GB, AMD Instinct MI210 64GB. Upgrading would unlock larger models and faster inference speeds.
Yes, NVIDIA L40 48GB with 48 GB of usable memory can run Flux.1 Dev at FP16 natively. Flux is a 12B parameter diffusion transformer that produces high-quality images. You can also run the Schnell variant for faster generation.
NVIDIA L40 48GB (48 GB VRAM) can handle various AI generation tasks beyond LLMs. For image generation, SDXL and Stable Diffusion 3.5 run well. Flux.1 Dev also runs natively for state-of-the-art image quality. For video, LTX Video 2.3 can generate short clips. Check the AI Capability Matrix above for detailed compatibility.
NVIDIA L40 48GB is excellent for AI image generation. With 48 GB of usable memory, it runs all major diffusion models including Flux.1, SDXL, and Stable Diffusion 3.5 at full precision. You can generate high-resolution images quickly and even handle video generation models.
Yes, NVIDIA L40 48GB with 48 GB of usable memory can run Qwen 3.5 27B at Q8 (near-lossless, ~28.9 GB) or even FP16 (~55.4 GB) depending on your context needs. This setup provides an excellent experience with this model. Use Ollama or vLLM for best results.
With 48 GB VRAM on NVIDIA L40 48GB, use Q8_0 for most models — it is near-lossless and you have the memory for it. For 70B+ models, Q6_K offers excellent quality. Reserve Q4_K_M for 100B+ models or when you need maximum context length.
NVIDIA L40 48GB has enough memory for many local LLMs, but bandwidth still matters a lot for real speed. Once a model fits, a faster-memory GPU can feel significantly better than a slower setup with similar capacity.
NVIDIA L40 48GB supports up to 2× GPU scaling via PCIe. With 2× GPUs, you get 96 GB effective memory with a 0.75× scaling factor per GPU. This enables running models like Qwen 3.5 397B A17B and Devstral 2 123B Instruct that don't fit on a single card.
NVIDIA L40 48GB uses PCIe for multi-GPU communication, which has approximately 25% scaling overhead. For best multi-GPU performance, consider NVLink-equipped variants.
Usually yes. If you want to run 2-4× NVIDIA L40 48GB for local AI, the bottleneck often becomes the platform, not the card. Workstation and server boards give you more CPU PCIe lanes, better x16 slot wiring, more spacing between cards, stronger power delivery, and usually more RAM capacity. Consumer x8/x8 layouts can work, but they are a common weak point in multi-GPU builds.
Compare with similar
