Unverified50% confidenceBenchmarkExact time
For Spark X 2.5, at 32,000 tokens the KV cache is approximately 1.3 GB with a total footprint of approximately 9 GB; at 128,000 tokens the KV cache is approximately 4.9 GB; and at 1 million tokens the KV cache is approximately 39 GB plus a model file of approximately 43 GB
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
9/27/2026
First Seen
Valid until: 12/26/2026
Sources
Related Entities
Related Claims
UnverifiedThe 8-bit version of the Spark X 2.5 model file is approximately 4 GB, and the full-precision version is approximately 8 GB79% similarUnverifiedThe combined unified memory of two DGX Sparks totals approximately 256 GB.75% similarVerified以LLaMA-3 70B（80层、64头、每头128维）为例，单个token的KV Cache约占2.6MB（FP16精度），32K上下文的KV Cache总量约达83GB71% similarUnverifiedWhen the context of LLaMA-2 7B is extended to 32K tokens, KV Cache GPU memory usage expands to approximately 16 GB, approaching the total memory of a single consumer-grade GPU.71% similarUnverifiedThe PagedAttention paper uses OPT-13B in FP16 as an example, where each token requires approximately 800 KB, and 2048 tokens require approximately 1.6 GB of KV Cache.69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/954251API
curl https://kongchang.com/api/v1/knowledge/claims/954251MCP
get_claim(id=954251)