Buy Crypto
Pay with
USD
USD
Buy & Sell
Hot
Supports Visa, Mastercard, SEPA & more
P2P
0 Fees
Flexible trading, zero fees
Gate Card
Use your crypto for payments worldwide
Markets
Trade
Basic
Advanced
Futures
Futures
Access hundreds of perpetual contracts
CFD
Gold
One platform for global traditional assets
Options
Hot
Trade European-style vanilla options
Unified Account
Maximize your capital efficiency
Demo Trading
Introduction to Futures Trading
Learn the basics of futures trading
Futures Events
Join events to earn rewards
Demo Trading
Use virtual funds to practice risk-free trading
Earn
Launch
CandyDrop
tag
Collect candies to earn airdrops
Launchpool
Quick staking, earn potential new tokens
HODLer Airdrop
Hold GT and get massive airdrops for free
Pre-IPOs
Unlock full access to global stock IPOs
Alpha Points
Trade on-chain assets and earn airdrops
Futures Points
Earn futures points and claim airdrop rewards
Investment
Square
Square
Post, share, and explore crypto trends
Live
moments live
Live crypto market analysis
Chat
Chat with crypto traders
News
What is happening in crypto
flower_head
More
Promotions
AI
Others
TradFi
Gate Card
Rewards
Square
Latest
Hot
News
Profile

Zhipu Releases GLM-5.1 High-Speed API, Setting Global Speed Record at 400 Tokens/s

AirdropBlackHole
According to monitoring by Dongcha Beating, Zhipu has launched the GLM-5.1 high-speed API for select enterprise clients, achieving a model output speed of 400 tokens/s, setting a new global record for the end-to-end speed limit of large model official interfaces. This high-speed version retains the capabilities of the original flagship model while being powered by a high-performance inference engine developed jointly by Zhipu and the TileRT team. This engine completely restructured the GPU's operational scheduling mechanism, statically arranging the model into a persistent Engine Kernel that resides on the GPU during the compilation phase. During single-card inference, computation, asynchronous I/O, and communication are all decomposed into tile-level micro-tasks, initiating the kernel only once. Intermediate results between operators are directly transmitted via registers and shared caches, eliminating the latency caused by frequent kernel startups and memory read/write in traditional inference. When scaled to a multi-card setup, TileRT further extends the specialization parallelism approach across an 8-card NVL topology, transforming originally homogeneous GPU nodes into heterogeneous Workers responsible for different tasks. In handling the attention layer computations of GLM-5.1, the system assigns GPU 0 to run a sparse index Worker dedicated to sparse index construction and routing decisions, while GPUs 1 to 7 run MLA Workers responsible for computation-intensive phases, fully integrating communication into the tile-level task pipeline, achieving deep overlap between computation and inter-card communication. This high-speed service is currently available to select enterprise clients on the Zhipu MaaS platform. In the future, this technology will further optimize FP8 inference and ultra-long context production environments, providing more deterministic performance support for low-latency sensitive scenarios such as AI programming, real-time interaction, and real-time voice.
ZHIPU26.55%
This page may contain third-party content, which is provided for information purposes only (not representations/warranties) and should not be considered as an endorsement of its views by Gate, nor as financial or professional advice. See Disclaimer for details.
  • Reward
  • Comment
  • Repost
  • Share
Comment
Add a comment
Add a comment
No comments

  • Pinned

Sitemap