TechnologyTracking
NVIDIA open-sources three models: how will enterprises choose by tier and post-train?
After NVIDIA open-sources Nemotron 3.5 Super VL and two Bolt edge models, how will enterprises select models by deployment tier and conduct post-training with their own data?
Watching: 英伟达未公布各模型的具体性能指标、硬件要求及后训练工具链细节，企业实际选择标准尚不明确。
Current read
NVIDIA's Nemotron 3.5 family spans data-center to desktop tiers; enterprises can select models by deployment needs and post-train with NeMo on proprietary data.
Evidence shows Nemotron 3.5 Lightning (30B MoE) supports deployment across PC, workstation, data center, and cloud, and can be post-trained via NVIDIA NeMo on enterprise domain data, tools, and workflows. Bolt 74B-A9B is a 74B-total/9B-active MoE with 2-bit quantization fitting 24GB, suited for local deployment. The reports do not specify Super VL's positioning or full performance comparisons, nor detailed post-training toolchain specifics.
How we read it
Nemotron 3.5 Lightning supports deployment from PC to cloud and can be post-trained with enterprise data via NeMo.
Bolt 74B-A9B is a 74B-total/9B-active MoE, 2-bit quantized to run locally in 24GB.
Evidence does not specify Super VL's positioning or model performance comparisons, nor detailed post-training steps.
Earlier updates
- — Public tracking started; no independent probability estimate yet.; Starting public context: NVIDIA said the three models span deployment tiers from data center to desktop, allowing enterprises to select by deployment level and conduct their own post-training.
- — NVIDIA's Nemotron 3.5 family spans data-center to desktop tiers; enterprises can select models by deployment needs and post-train with NeMo on proprietary data.; Evidence shows Nemotron 3.5 Lightning (30B MoE) supports deployment across PC, workstation, data center, and cloud, and can be post-trained via NVIDIA NeMo on enterprise domain data, tools, and workflows. Bolt 74B-A9B is a 74B-total/9B-active MoE with 2-bit quantization fitting 24GB, suited for local deployment. The reports do not specify Super VL's positioning or full performance comparisons, nor detailed post-training toolchain specifics.