Unverified50% confidenceBenchmarkExact time
On 8 NVIDIA B300 GPUs, a 47-billion-parameter MoE model reaches about 52,000 tokens per second per GPU on the new Olmo-core 3 framework, whereas the old implementation reached only about 19,400, a throughput improvement of about 2.7 times; Ai2 labels this as a "preliminary result."
1
Sources
50%
Confidence
Short-term (~14 days)
Relevance
10/14/2026
First Seen
Valid until: 10/28/2026
Sources
Related Entities
Related Claims
UnverifiedOfficial measurement: a 24-frame 720p MP4 at 64 samples consumes about 58 GPU seconds, equivalent to about 6 yen.74% similarUnverifiedJANCTION Render gives each new key 500 yen of GPU time (5000 GPU seconds), valid for 14 days, with no card binding required; afterwards preview tasks get 2 GPU minutes free per day, and other GPU time is billed at 0.1 yen per GPU second.70% similarUnverifiedRunning a 16-bit precision 70B model on 2×80GB GPUs leaves only 2.9 GB of VRAM after the weights are loaded, accommodating only about 9,000 in-flight tokens, which in practice cannot serve requests effectively.70% similarUnverifiedCompressing the context from 128K to 32K may quadruple the number of concurrent users a single GPU can serve, representing a trade-off between throughput and context length.69% similarVerifiedTaking a 70B parameter model with FP16 precision and 32 attention head layers as an example, the KV Cache for a single 200K-length request requires approximately 80–120 GB of VRAM, far exceeding the 80 GB VRAM limit of a single A100 GPU.69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1025298API
curl https://kongchang.com/api/v1/knowledge/claims/1025298MCP
get_claim(id=1025298)