AITOPAI热报
IndustrySelected

DeepSeek V4.1 Would Need Thousands More Blackwell GPUs

Teortaxes@teortaxesTex
Post
View the post
Why we picked it

The back-of-envelope math suggests DeepSeek's RL training pipeline is compute-bound at the GPU level, not CPU-bound, which changes how analysts should read its published infrastructure claims. Investors and policy watchers tracking US export-control impact on Chinese labs can use the 3,000-GPU figure as a yardstick for scaling bottlenecks. Compare it with NVIDIA Blackwell allocation tiers announced for US hyperscalers.

A widely shared estimate on X calculates that DeepSeek could generate roughly 165 trillion tokens per month using 3,000 Blackwell GPUs if maximising throughput. Under a pessimistic reading of V4.1's architecture with 1M-token contexts, that would yield about 90 million trajectories, or 3 million per day. The author suggests this volume could saturate one DeepSeek reinforcement-learning compute cluster, possibly four. The analysis concludes DeepSeek would need far more GPUs than CPUs.

Original · Teortaxes

Maximizing for throughput, DeepSeek could generate ≈165T tokens a month off 3K Blackwell GPUs. With 1M contexts, *pessimistically* given V4.1's shape – 90M trajectories, 3M/day. That might saturate one DSec RL cluster. Maybe 4. idk? Seems they need WAY more GPUs than CPUs. https://t.co/Q2sUnZaVCU

#DeepSeek#DeepSeek V4.1#NVIDIA Blackwell#reinforcement learning#reasoning models

Keep reading

Recent stories with the same tag or section
ProductsUS vs ChinaSelected12:24
DeepSeek Search ranks 17th on Artificial Analysis Search Index

The headline price is low, but the benchmark shows the real cost lives in tokens: ~7k-token search results make DeepSeek's model cost the 2nd highest on the board at ~US$0.023 per task. Developers building search-augmented agents should weigh that against the weak BrowseComp score of 63%, which trails Perplexity variants and Octen at 85% to 87% on multi-hop tasks. Compared with alternatives like Exa and Firecrawl, DeepSeek wins on price, not on factual accuracy.

#DeepSeek#Search API#Artificial Analysis#agents
官方账号Artificial Analysis@ArtificialAnlysSource
ModelsUS vs ChinaMulti-sourceSelected13:20
DeepInfer Release Draws Criticism Over Heterogeneous Hardware Support

The debate centres on whether DeepInfer's reported throughput depends on a tightly controlled, homogeneous GPU fleet rather than general infrastructure design. Engineers planning large-scale inference deployments on GB200 clusters can weigh whether NVIDIA Dynamo plus 3FS achieves comparable results outside DeepSeek's own setup.

#DeepSeek#DeepInfer#NVIDIA Dynamo#GB200
Event
Teortaxes@teortaxesTex11 sourcesSource
IndustryUS vs ChinaMulti-sourceSelected79°13:09
DeepSeek V4.1F shows 5x-7.4x decode edge on Ascend 950DT vs GB200

The cost crossover at 6.8K tokens/s/chip reframes the Ascend-versus-Blackwell debate from raw benchmark losses to workload-dependent economics, suggesting DeepSeek could run its API on a sub-US$1B Blackwell cluster while pivoting post-training to Ascend. Practitioners tracking inference infrastructure or Huawei's CANN ecosystem get a rare concrete per-chip price and throughput comparison. The proposed 128-chip Ascend supernode topology has no Nvidia equivalent, which matters for anyone planning large-scale decode deployments.

#DeepSeek#Ascend 950DT#GB200#DeepEP-Ascend
Event
Teortaxes@teortaxesTex11 sourcesSource
ModelsUS vs ChinaMulti-sourceSelected77°09:43
DeepSeek Open-Sources DeepInfer Inference Framework for V4.1 Flash

Open-sourcing the framework that runs DeepSeek's own production fleet gives operators a rare look at inference internals optimised for both NVIDIA and Huawei Ascend hardware, a dual-platform angle few comparable stacks like vLLM or SGLang emphasise. Teams running DeepSeek models on Ascend 950 in Southeast Asia gain a first-party option instead of community ports. Deterministic sampling is currently NVIDIA-only, so check hardware fit before adopting.

#DeepSeek#DeepInfer#open-source models#inference
Event
SuperTechFans11 sourcesSource
ProductsBuildersMulti-sourceSelected08:37
vLLM Matches DeepInfer Decode Performance on DeepSeek-V4.1-Flash

Open-source inference stacks closing the gap with a lab's proprietary serving setup changes the economics of self-hosting large MoE models. Teams weighing DeepSeek's API against running their own GPUs can use these Wide-EP and EPLB results to estimate revenue per GPU at off-peak prices. The GB200 and GB300 benchmarks also give infrastructure planners concrete throughput targets for the latest NVIDIA hardware.

#vLLM#DeepSeek#DeepInfer#inference
Event
官方账号vLLM@vllm_project7 sourcesSource
IndustryUS vs ChinaMulti-sourceSelected20:40
DeepSeek Reaches US$1B Annualized Revenue With 33 GB200 Racks

The implication is that inference efficiency, not GPU count, is becoming the competitive lever for API revenue among Chinese labs. Builders pricing their own token workloads can use the 155T tokens-per-day figure as a benchmark for what a small rack footprint can sustain. Compared with US labs running far larger fleets, the capital-per-revenue ratio here is unusually lean.

#DeepSeek#GB200#inference#China AI
Event
Teortaxes@teortaxesTex3 sourcesSource