Unverified50% confidenceBenchmarkExact time
TIDE's peak memory is up to 4.7 GiB lower than FSDP2+SymmMem, and on Wan2.2 with Slot-Reuse enabled it is even lower than Stock FSDP2.
1
Sources
50%
Confidence
Short-term (~14 days)
Relevance
10/30/2026
First Seen
Valid until: 11/13/2026
Sources
Related Entities
Related Claims
UnverifiedCompared to FP16, FP8 reduces memory footprint and bandwidth consumption by half, and the AIC's compute throughput for FP8 can reach twice that of FP1669% similarUnverifiedFP4-Local is not smaller than 4-bit GGUF: UD-IQ4_XS GGUF is 157 GB, while FP4-Local is about 166 GiB.68% similarUnverifiedAt 64-GPU scale, TIDE reduces step time by up to 27.1% on Llama-3.1 70B, Wan2.2 and Qwen3-30B-A3B (compared with Stock FSDP2).67% similarUnverifiedFP4-Local uses about 166 GiB of VRAM, and on two RTX PRO 6000 GPUs (192GB total) it can fit a 1.8M token FP8 KV cache; a dual DGX Spark setup can reach 2M tokens.66% similarUnverifiedDeepSeek-V4.1-Flash has 763B parameters, BF16 weights are about 1.5 TB, the official FP4/FP8 checkpoint is 475.2 GiB, and coolbho3k's 3-bit EXL3 package compresses it to 309.3 GiB.63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1114851API
curl https://kongchang.com/api/v1/knowledge/claims/1114851MCP
get_claim(id=1114851)