NVIDIA
RTX 5090 Laptop 24GB
RTX 50 LaptopLaptopBlackwellMOBILECUDA
Operating mode
Choose the operating mode for this hardware
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.
About this GPU for AI
The RTX 5090 Laptop is NVIDIA's Blackwell mobile flagship, featuring 24 GB of GDDR7 at 896 GB/s bandwidth in a 95–150W TGP package. Based on the GB203 die (not the desktop RTX 5090's GB202), it delivers 52 TFLOPS FP16 and 1,824 AI TOPS — making it the first laptop GPU with enough VRAM to run 70B models at Q3/Q4 without CPU offloading. Available from March 2025, it represents a major step forward for portable AI inference compared to the 16 GB Ada laptop generation.
Beyond LLMs
AI Capability Matrix
What AI tasks this GPU can handle — from text generation to image and video creation.
| Capability | Status | Representative Model | Detail |
|---|
| LLM Chat (7B) | Runs natively | Llama 3.1 8B Q4 | — |
| LLM Coding (30B) | Runs natively | Qwen 3 30B Q4 | — |
| LLM Large (70B) |
portablethermally-limitedlaptopblackwelllarge-vrammobile-flagship
Specifications
Compute
FP1652 TFLOPS
INT8832 TOPS
ArchitectureBlackwell
Memory
VRAM24 GB
Bandwidth896 GB/s
General
FamilyRTX 50 Laptop
SegmentLaptop
InterconnectMOBILE
Compute PlatformCUDA
Key Features
24 GB GDDR7 VRAM on a 256-bit busBlackwell GB203 die with 5th-gen Tensor Cores, FP4 and FP8 support52 TFLOPS FP16 / 832 INT8 TOPS / 1,824 AI TOPS896 GB/s memory bandwidth95–150W configurable TGPDLSS 4 with Multi-Frame Generation
For AI Workloads
Strengths
- 24 GB GDDR7 is the largest VRAM ever shipped in a laptop GPU — fits 70B Q3/Q4 models without CPU offloading
- 896 GB/s bandwidth delivers fast decode for large quantized models in a portable chassis
- 5th-gen Tensor Cores with FP4 support enable next-generation quantization formats for maximum throughput
- First laptop GPU capable of single-card 70B inference — a meaningful capability leap
Considerations
- Based on GB203 die, not desktop GB202 — delivers approximately 35–40% of desktop RTX 5090 sustained compute
- 95–150W TGP means performance varies significantly between laptop models — verify TGP before purchasing
- Laptops equipped with this GPU carry a significant premium ($2,899+ laptop price)
- Thermal throttling under sustained long inference sessions limits effective throughput in compact chassis
Blackwell is NVIDIA's fifth-generation RTX architecture, built on TSMC's 4NP process. It introduces 5th-generation Tensor Cores with native FP4 precision support, enabling double the inference throughput per watt compared to Ada Lovelace's FP8 operations. Key innovations include the Neural Rendering Pipeline for AI-driven shading and the debut of GDDR7 memory in consumer GPUs.
AI Relevance
FP4 Tensor Cores deliver the highest tokens-per-watt efficiency in any consumer architecture. Native FP4 quantization means models can run at lower precision with minimal quality loss, effectively doubling the effective VRAM for model weights.
Process: TSMC 4NPPlatform: CUDATensor Cores: Gen 5Precisions: FP32, FP16, BF16, FP8, FP4, INT8, INT4
Recommendations by Workload
Qwen 3 14B 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.
Decode 74.3 tok/s · 60K ctx · llama.cppEST.
Codestral 2 25.08 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.
Decode 53.8 tok/s · 48K ctx · llama.cppEST.
Just out of reach
Models you could run with an upgrade
High-quality models that need a bit more memory
397BTier 100Needs ~246.9 GB
123BTier 100Needs ~81.0 GB
1000BTier 100Needs ~617.0 GB
1000BTier 100Needs ~617.0 GB
1600BTier 100Needs ~866.2 GB
Image & Video Generation
Diffusion Model Compatibility
41 of 52 models can generate images or video on your RTX 5090 Laptop 24GB
Upgrade paths
Upgrade from RTX 5090 Laptop 24GB
See what you unlock with more powerful hardware
Upgrade options
Upgrade options
MacBook Pro M4 Max 36GBNext step up
36 GB Unified (+12)
AUnlocks 1 additional models that do not fit on the current setup.Unlocks Gemma 4 31B
Unlocks 1 additional models that do not fit on the current setup.
~$2,499 MSRP
32 GB VRAM (+8)
AUnlocks 6 additional models that do not fit on the current setup.Unlocks Gemma 4 31B, Kimi Linear 48B A3B, Falcon 40B Instruct+3 more
Unlocks 6 additional models that do not fit on the current setup.
~$4,000 MSRP
Mac mini M4 64GBBest value
64 GB Unified (+40)
BUnlocks 17 additional models that do not fit on the current setup.Unlocks Qwen 2.5 VL 72B, Gemma 4 31B, Llama 3.3 70B+14 more
Unlocks 17 additional models that do not fit on the current setup.
~$1,099 MSRP
AMD Instinct MI350X 288GBBiggest leap
288 GB VRAM (+264)8000 GB/s (+7104)
BUnlocks 45 additional models that do not fit on the current setup.Unlocks Qwen 3.5 397B A17B, Devstral 2 123B Instruct, Qwen 3.5 122B A10B+42 more · +109% faster avg
Unlocks 45 additional models that do not fit on the current setup.
Lifts average decode speed across fitting models by about 109%.
~$8,000 MSRP
Frequently Asked Questions
24
GB
RTX 5090 Laptop 24GBCategory AvgMacBook Pro M4 Max 36GB
| Image Gen (SDXL) | Runs natively | SDXL 1.0 FP16 | ~~5.7s per image |
| Image Gen (Flux) | Runs with offload | Flux.1 Dev FP16 | ~~25.7s per image |
| Image Gen (SD 3.5) | Runs natively | SD 3.5 Large FP16 | ~~31.4s per image |
| Video Short (25f) | Runs natively | LTX Video 2B | ~~5s/frame |
| Video Long (100f) | Won't fit | Wan Video 14B | ~~14.6s/frame |
Qwen 3.6 27B is a specialized fit for Agentic Coding. It is a recent-generation family, which helps on current local SOTA workloads. It should run, but memory headroom will be limited. Context coverage stays within the requested workload envelope. Known distribution channels: huggingface, lm-studio.
Decode 34.3 tok/s · 69K ctx · llama.cppEST.
Qwen 3 14B 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.
Decode 95.2 tok/s · 80K ctx · llama.cppEST.
Granite 4.1 8B matches RAG and keeps a practical fit profile. It sits in the middle of the current generation mix. It fits natively with comfortable headroom. Context coverage stays within the requested workload envelope. Known distribution channels: huggingface, ollama.
Decode 112.0 tok/s · 93K ctx · llama.cppEST.
21B18.6 GB145 tok/s52K ctx
Image
| MAGI-1Video | 256×256 | ~13.4s/frame | 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.
Buying advice
Should you buy RTX 5090 Laptop 24GB for local AI?
Excellent choice for local AI
Runs 26 of 50 top models well — a strong all-rounder for local inference.
What will limit you first
This setup is broadly balanced for this model.
Very little memory headroom
You can run the model, but there is not much room left for longer context, bigger batches, extra apps, or future model updates.
Best upgrade itinerary
Buy headroom, not only minimum fit
A slightly larger memory tier gives you safer context growth and makes the recommendation more future-proof.
Unlocks 1 additional models that do not fit on the current setup.
Want more headroom? MacBook Pro M4 Max 36GB (36.0 GB unified memory) is the next step up.
