从业者公开求租约4.5万小时H100等效算力，愿以灵活条件换取低价

全部来源

• 从业者公开求租约4.5万小时H100等效算力，愿以灵活条件换取低价

  LLMJunky · 主条目

  作者（Beni）称，因收到大量私信询问，公开发帖求租约4.5万小时的H100等效算力，时间为未来一个月内，理想方案为32个节点7-8天、16个节点14-16天或8个节点28-30天。作者表示愿以极高灵活性换取低价，包括预付、接受30天窗口、可拆分多个时段、不挑地区与服务器厂商，并可考虑更低SLA或不可退款预订等安排。作者还称可接受H200、B200、B300、A100 SXM及MI300X/MI325X/MI355X。硬性要求是GPU须在同一集群并具备NVLink加InfiniBand或RDMA级互联；作者强调只接受真正激进的折扣。

  查看原始帖子 ↗
  查看原文

  RT Beni I fear I'll regret doing this but given the amount of dms I am getting about it I might as well: Looking to rent ~45k hours of H100-equivalent sometime over the next month Ideally: 32 nodes for ~7-8 days 16 nodes for ~14-16 days 8 nodes for ~28-30 days I'm willing to be EXTREMELY flexible if it gets me a better price: -Can pay upfront / commit to a minimum spend -Can give you a 30-day window to fit me in whenever your clusters would otherwise be idle -Can start on short notice -Can commit to an exact runtime -Don't care about region, start date, time of day, weekends, holidays etc. -Open to splitting into multiple blocks -Can move between clusters/regions between blocks if needed -Don't care about server manufacturer, bare metal vs virtualization, orchestration stack etc -Can handle deployment, monitoring, debugging, data preprocessing, checkpointing etc -Open to reduced SLA requirements, non-refundable reservations or other commercial arrangements depending on the discount ***Provided the economics justify it*** 1. Flexible on cluster size = more GPUs for less time or fewer GPUs for longer 2. Open to H200, B200, B300, A100 SXM, MI300X/MI325X/MI355X 3.Potentially open to scheduled interruptions/restarts if we can checkpoint safely and the economics justify it Non-negotiable requirements: GPUs are on the same cluster with NVLink + InfiniBand/RDMA or equivalent If you've got stranded capacity and can offer genuinely aggressive pricing, please reach out PS: Competitive does not mean 5% below on-demand, pls don't waste my time or I will make you regret it, thank you for your understanding

前后事件

相关事件