Nvidia publie Nemotron 3.5 Bolt 74B-A9B, MoE 24 Go pour RTX Spark
TL;DR
- Sparse MoE and 2-bit quantization let Bolt 74B-A9B consume compute equivalent to a 9B model at inference, fitting into 24 GB VRAM.
- RTX Spark ships October 16 across eight OEM laptops from $2,599; Microsoft's exclusive Dev Box at $5,999 claims one petaflop of AI compute.
- Nemotron 3.5 spans three tiers: Super VL 120B (cloud), Bolt 74B-A9B (24 GB local agents), and Bolt E4B 4B (on-device gaming characters).
NVIDIA a publié le 15 octobre Nemotron 3.5 Bolt 74B-A9B, un MoE hybride Mamba-Transformer destiné à exécuter « everyday agentic tasks locally on NVIDIA RTX/DGX Spark » dans une empreinte de 24 Go, avec un objectif d'au moins 80 tokens par seconde en décodage.
Le modèle descend du Super 3.5 VL 120B-A12B via Star Elastic, que le rapport présente comme « a push-button, router-guided architecture search and end-to-end compression framework invented at NVIDIA ». Le pipeline ramène les paramètres actifs de 12 à 9 milliards et le top-K des experts de 22 à 9. La quantification est agressive : les couches FC1 et FC2 « constitute the majority of the model weights are quantized to IQ2_XXS at 2.0625 bits per element », tandis que les couches Mamba passent en NVFP4 et que l'attention, les embeddings, le lm_head et les poids MoE latents restent en Q8_0.
Le pari revendiqué par le laboratoire : « the resulting model is better than similarly sized models that are trained from scratch and quantized to 4 bit ». Côté vitesse, le drafter de speculative decoding LiLiCorr, 1,13 Md de paramètres, affiche des longueurs d'acceptation moyennes de 4,68 en greedy et 4,25 en échantillonnage probabiliste. Nvidia mesure 83,99 TPS sur AIPerf SPEED-Bench-Coding et un temps de rejeu de 600,80 secondes sur AA AgentPerf Local, pour un classement #1 PinchBench, #2 WebArena et #2 BrowseComp parmi les « models with a similar memory footprint ».
C'est la seconde pièce de la série 3.5 en deux jours après Nemotron 3.5 Super VL 120B : le gros frère sert ici de donneur, Bolt vise la station locale. Les checkpoints BF16, GGUF et les variantes LiLiCorr sont publiés sur HuggingFace.
Ce qu'en disent les autres médias
-
NVIDIA Blog Lire →
First-party account cataloging the full software partner lineup at launch, from agent frameworks (GitHub Copilot, Hermes, Perplexity) to creative suites (Adobe, ComfyUI, Blender) and developer tools.
RTX Spark delivers the full stack of NVIDIA RTX and CUDA technologies for creators, developers and gamers, while unlocking a new paradigm via powerful local AI agents.
-
NVIDIA Newsroom Lire →
Details the Adobe deal terms: 2x AI processing gains in Photoshop and Premiere, 5x faster Frame.io embeddings, and Brand Concierge optimized on Nemotron open models.
AI agents are helping creators unlock deeper and richer workflows using the Adobe applications they know and love.
-
VentureBeat Lire →
Provides benchmark comparisons against Claude Opus 5 and GPT-6.1 Sol on domain tasks (medical claims, security resilience), plus throughput numbers (625 tok/s) for enterprise evaluation.
The potential value for enterprises is the ability to adapt a model to their documents and procedures, then choose where it runs.
-
TechTimes Lire →
Explains sparse MoE and 2-bit quantization letting the 74B model behave like a 9B model at inference, framing local AI as a full alternative to cloud subscriptions.
A 74-billion-parameter model may behave, in terms of compute and memory bandwidth, like a 9-billion-parameter model on any given inference.
-
Techzine Lire →
Emphasizes Super VL efficiency versus Nvidia's own largest reasoning model and quantifies Dynamo backend gains (8x user capacity, 85% lower latency) for cloud-scale alongside local Bolt.
Super VL achieves accuracy comparable to the company's largest reasoning model...with about one-fifth the parameters.
-
Videocardz Lire →
Anchors the launch with retail pricing across eight OEMs ($2,599 to ~$7,000) and highlights Microsoft's exclusive $5,999 Dev Box claiming one petaflop of AI compute.
Microsoft's Surface RTX Spark Dev Box, exclusive to Microsoft.com in the US, lands at $5,999.99 and claims one petaflop of AI compute.
Article original publié par research.nvidia.com
Lire l'article original →Titre original : Nvidia publie Nemotron 3.5 Bolt 74B-A9B compressé en 24 Go pour accompagner le lancement des RTX Spark