ScaleTensor Docs ← scaletensor.fun

Documentation

How we build and run your GPU stack

These pages describe the reference designs we deploy and how a project runs commercially, from the first brief to handover. Every project gets its own statement of work. Where it differs from these pages, the statement of work applies.

PlatformNVIDIA GPU Operator

Drivers, toolkit, device plugin, MIG and monitoring managed as one Kubernetes component.

 PlatformImages & model artifacts

Private registry in Singapore, proxy caches, signing, scanning and offline model storage.

 ModelsModel serving

vLLM and Triton behind an OpenAI-compatible gateway, sized and benchmarked on your workload.

 CommercialExport control & KYC

The end-user and end-use checks we run before supplying any GPU hardware.

 CommercialPayment

Currencies, accepted methods, typical payment schedules and invoice verification.

Engagement process

  1. Project brief. You send the brief from scaletensor.fun or by email. We reply within 1–2 business days.
  2. Scoping call. 30–45 minutes on workload, models, site power and cooling, and timeline.
  3. Spec and quote. Bill of materials, architecture, delivery plan and price. Hardware quotes are valid for 7 days because GPU pricing moves quickly.
  4. KYC. For any hardware supply we complete the export-control checks before we accept a purchase order.
  5. Purchase order and first payment. See payment for typical schedules.
  6. Build and burn-in. Assembly, firmware alignment and a 48-hour burn-in per node (GPU stress, NCCL all-reduce, memory and storage tests).
  7. Deploy. Kubernetes or Slurm, the GPU Operator, registry, monitoring and your models.
  8. Acceptance. We run the acceptance tests agreed in the statement of work and share the results.
  9. Handover or managed operations. Runbooks, credentials and diagrams handed over, or we keep running it under a monthly retainer.

Contact

Email hello@scaletensor.fun or use the project brief on the main site.

SCALETENSOR PTE. LTD. · UEN 202650485R · Singapore

文档中心

我们如何搭建并运维你的 GPU 平台

这里介绍我们交付的参考架构,以及一个项目从需求到交接的商务流程。每个项目都会签订单独的工作说明书(SOW),如与本页内容不同,以 SOW 为准。

平台NVIDIA GPU Operator

驱动、容器工具包、设备插件、MIG 与监控,作为一个 Kubernetes 组件统一纳管。

 平台镜像与模型管理

新加坡本地私有镜像仓库、代理缓存、签名与漏洞扫描,以及离线模型存储。

 模型模型推理服务

vLLM 与 Triton,统一 OpenAI 兼容网关,按你的实际负载做容量规划与压测。

 商务出口管制与 KYC

供应任何 GPU 硬件之前,我们执行的最终用户与最终用途审查。

 商务付款方式

币种、付款渠道、常见付款节点,以及如何核实发票与收款信息。

合作流程

  1. 提交项目需求。在 scaletensor.fun 填写需求表或直接发邮件,我们 1–2 个工作日内回复。
  2. 需求沟通。30–45 分钟,了解负载、模型、机房电力与散热、时间要求。
  3. 方案与报价。提供物料清单、架构设计、交付计划与价格。GPU 价格波动大,硬件报价有效期 7 天。
  4. KYC 审查。凡涉及硬件供应,接受采购订单前必须完成出口管制审查。
  5. 下单与首付款。常见付款节点见付款方式。
  6. 组装与老化测试。组装、固件统一,每节点 48 小时老化测试(GPU 压力、NCCL all-reduce、内存与存储测试)。
  7. 部署。Kubernetes 或 Slurm、GPU Operator、镜像仓库、监控以及你的模型。
  8. 验收。按 SOW 约定的验收用例执行测试,并提交测试结果。
  9. 交接或托管运维。交付运维手册、账号与架构图;或由我们按月托管继续运维。

联系我们

邮件 hello@scaletensor.fun,或在官网提交项目需求。

SCALETENSOR PTE. LTD. · UEN 202650485R · 新加坡