NVIDIA repeats its Blackwell Ultra throughput claim on DeepSeek-V3 671B
NVIDIAAI · x · 2026-07-21
NVIDIA repeats its claim that Blackwell Ultra set a pre-training record on DeepSeek-V3 671B, reaching 1,648 TFLOPs per GPU.
- The company says that is about 3× the prior generation’s delivered performance.
- The gain is attributed to co-design and software optimization across Megatron-Core, TorchTitan, and JAX.
Related event: NVIDIA Blackwell Ultra Sets DeepSeek-V3 Training Throughput Record(2 posts)→
More from Infra
- DeepSpeed shows AutoEP/AutoTP/AutoSP parallelism gains, hosts office hours — PyTorch · 2026-10-27
- Cloudflare Durable Objects validated as top architecture for AI agent WebSocket servers — tristanbob · 2026-10-27
- Gary Marcus: With Q3 data added, Anthropic's compute-efficiency growth shows deceleration — GaryMarcus · 2026-10-27
- Inference is just Kubernetes with a multi-trillion-dollar market, and the communities overlap — matt_slotnick · 2026-10-27
- vLLM eyes Rust migration as inference engines race to rewrite the stack — AccBalanced · 2026-10-27
- Resin lets an agent build standalone LLM inference engines; vLLM and SGLang optimize DeepSeek kernels differently — AccBalanced · 2026-10-27