# 产业日报 2026-10-06 来源范围:`/workspace/intel/archive/2026-10-06/`(industry/ 主;vmware/redhat/vates/scalecomputing 等 vendor 仅旁证)。覆盖窗口:10-05(周一)早报之后 → 10-06(周二)早。抓取统计:主跑 Saved 267 / skipped 28 / errors 40(+ Scale Computing 源修复后定向补抓 Saved 14 / errors 0);blocked_hosts:blocksandfiles.com、docs.redhat.com、e.huawei.com、support.huawei.com、www.nextplatform.com、www.reddit.com、www.theregister.com。与 `industry/daily/2026-10-05.md` 对照:industry 目录 **仅 1 个新增 URL**(STH Kioxia CM9-R E3.S SSD 评测,轻–中);**本日无公开战略级产业新稿**。**勘误(重要):** STH 已修订 Xeon 6+ 原文——"12 channels of DDR5 **(not MRDIMM)**"、带宽测算改为 "DDR5-8000 **(not MRDIMMs)**";10-05 日报/专题中「含 MRDIMM」表述**作废**,以本日为准。中文频道(至顶网/奇客)新增标题为媒体快讯(Google 冻结 OSS VRP、Project Suncatcher 卫星 TPU、WSL Containers GA、CNNIC 生成式 AI 报告),**轻条/非 HCI 主线**。Blocks&Files / Register / NextPlatform 本日仍 host_blocked,**无标题可列,不发明**。数字标「媒体评测」「媒体快讯」「厂商自测」「委托研究」「续档」。 ## 摘要 今日产业信号收成六句话: 1. **本日无公开战略级产业新稿;续档主线 + 勘误:** 10-05 主线「**Xeon 6+ Clearwater Forest 广泛供货 × 内存涨价传导**」继续有效——4 SKU、144–288 E-core、低 TDP 1.14–1.15 W/core、list 约 $4,600→接近 $15,000;DGX Spark 128GB $3,999→约 $6,950、STH 主编"$2000 for 64GB ECC RDIMMs"。**但 STH 修订原文:Xeon 6+ 为 12 通道 DDR5、不支持 MRDIMM**,每核带宽 2.67–5.34 GB/s 的测算前提改为普通 DDR5-8000——**10-05「含 MRDIMM」须更正**;对 HCI 节点选型的含义:288 核密度下每核内存带宽偏低的判断不变,且不能指望 MRDIMM 补带宽。 2. **industry 轻–中 NEW · Kioxia CM9-R 15.36TB E3.S 评测(STH):** PCIe 5.0 / NVMe 2.0,单 x4 或双 x2(**双端口**,面向 HA 阵列)、BiCS Gen8 TLC、1 DWPD / 5 年;额定顺序读写 14,800 / 11,000 MB/s、4K 随机 3.4M / 540K IOPS;STH 称"在 245/256TB 级 SSD 时代,15.36TB 已像中低容量"——**媒体评测(本地仅取到首页摘要段与规格表,结论页未取得);与 10-05 BiCS NAND 展台互证 NAND 需求叙事。** 3. **Vendor 旁证 · 内存/CPU 配比话术跟进:** VMware「Optimizing AI Deployments Part 3」称 GPU 推理 VM 的 vCPU(16→128)与内存(32→512GB)增减对 Llama3.1 8B 推理"little or no impact",主张把余量给 Agentic AI / 数据库(Llama3 70B 与 HammerDB 1.8M TPM 同机"no measurable impact")——**Broadcom 性能团队自测**;同日 Forrester **受 Broadcom 委托** Spotlight(90%/70%/33%/50%)。在内存涨价背景下,厂商开始把「少配内存、提高共置密度」做成 TCO 话术——**与 Sangfor aSV 内存分层、VMware memtier 同一语境(续档对照,勿双抬)**。 4. **Vendor 旁证 · 平台补丁与生态:** XCP-ng 8.3 LTS 10 月更新 #1(存储在线迁移 2–2.5×、CBT/NFS/iSCSI 修复、Windows tools 修数据损坏、OpenSSL 3.5.5——**GA 补丁**);Red Hat DH2i 三用例(Windows VM rehost 到 OpenShift Virtualization → RHEL → 容器化 SQL AG)、MLOps edge("KServe now in MicroShift"——**无版本,勿 GA**);Scale Computing 源修复后补录 c-store 势头 PR(Casey's 2,600+ 店等——**厂商自称**)——**详情归 vendor / openshift / migration / edge 专题。** 5. **媒体快讯(轻,非 HCI):** 奇客转述 **Google 因大量无效 AI bug 报告冻结 OSS VRP(10-01 生效,2027 Q1 更新)**——AI 生成漏洞报告挤占维护者,安全运营侧信号;**Google Project Suncatcher** 卫星搭载 4 颗 TPU 测试太空 AI 数据中心(热约束每次运行 15 分钟);**微软 WSL Containers GA**;CNNIC 称 2026 上半年国内生成式 AI 用户破 7 亿——**均为媒体转述,轻条**。 6. **对规划:** 今日无新主线,早报**续抬 10-05 主线并发布勘误**:Clearwater Forest 无 MRDIMM → 高核密度 HCI 节点的每核内存带宽与内存配比需按普通 DDR5 重算;内存涨价下「GPU/AI 节点少配内存 + 共置 CPU 业务」成为厂商 TCO 新话术(VMware 自测),我方需准备可复核的共置/超分基准;E3.S 双端口 Gen5 SSD 进入评测期,可作全闪 HCI 节点介质选型参考。 ## 原文摘录 ### 1. 勘误 · Intel Xeon 6+ SKU List and Value Analysis(STH,原文修订) - 归档:`industry/intel-xeon-6-sku-list-and-value-analysis-clearwater-forest-runs-wide-0d5dd057.*`(对照 `archive/2026-10-05/industry/` 同名文件) - URL:https://www.servethehome.com/intel-xeon-6-sku-list-and-value-analysis-clearwater-forest-runs-wide/ - 口径:**媒体分析;页面修订**(diff 两处,均为 MRDIMM 措辞)。 > (10-05 版)… 2 socket support, 12 channels of DDR5 (and MRDIMM) memory, 96 PCIe lanes… > (10-06 版)… 2 socket support, 12 channels of DDR5 (not MRDIMM) memory, 96 PCIe lanes, and Intel's new Application Energy Telemetry (AET)… > (10-06 版)… assuming 12 channels of DDR5-8000 (not MRDIMMs), this spans from 2.67GB/second of memory bandwidth per core for the 288-core SKU to 5.34GB/second for the 144-core SKU. ### 2. industry 轻–中 NEW · Kioxia CM9-R 15.36TB E3.S NVMe SSD Review(STH) - 归档:`industry/kioxia-cm9-r-15-36tb-e3-s-nvme-ssd-review-98361b41.*` - URL:https://www.servethehome.com/kioxia-cm9-r-15-36tb-e3-s-nvme-ssd-review/ - 来源:ServeTheHome;`fetched_at` 2026-10-06T00:34:35Z(=10-06 08:34 CST);本地 txt 仅 1,770 B(首页摘要 + 规格表,后续评测页未取得) - 口径:**媒体评测**;规格为 Kioxia 额定值。 > Today we are looking at the Kioxia CM9-R 15.36TB E3.S PCIe Gen5 SSD. > It is strange to think of that as a midrange or even lower capacity drive, but now that we are in the era of 245/256TB-class SSDs, it is where we are. > Kioxia specifies PCIe Gen5 operation in either single x4 or dual x2 modes. With its CMx series of drives, Kioxia usually targets enterprise deployments where dual-port operation is important for high-availability arrays. > | Flash | BiCS FLASH generation 8 TLC | Endurance / warranty | 1 DWPD / 5 years | Rated sequential read / write | Up to 14,800 / 11,000 MB/s | Rated 4K random read / write | Up to 3,400,000 / 540,000 IOPS | ### 3. Vendor 旁证 · VMware AI Part 3 / Forrester Spotlight - 归档:`vmware/optimizing-ai-deployments-with-vmware-cloud-foundation-part-3-f4be665b.*`、`vmware/forrester-spotlight-finds-that-vmware-cloud-foundation-streamlines-pri-4533d242.*` - 口径:**厂商自测**;**委托研究(commissioned by Broadcom)**。 > Even when we allocated only 25%-50% of available CPU cores to the VM, the inference performance is on par with bare metal. > These tests show that we don't need to allocate more vCPUs and memory than a workload uses. > Source: Results are based on interviewed customers. … a commissioned study conducted by Forrester Consulting on behalf of Broadcom, October 2026. ### 4. Vendor 旁证 · XCP-ng / DH2i×OpenShift / RH MLOps edge / Scale c-store - 归档:`vates/october-2026-updates-1-for-xcp-ng-8-3-lts-b006f9d7.*`、`redhat/from-ai-pilots-to-production-how-dh2i-and-red-hat-help-customers-run-r-49d27c81.*`、`redhat/mlops-for-edge-ai-preparing-ai-models-for-deployment-at-the-edge-5f307e38.*`、`scalecomputing/scale-computing-inc-fuels-momentum-in-the-convenience-store-f263711b.*` > Storage migration is much faster (2 to 2.5 times in some of our tests)…(XCP-ng) > Crawl: Rehost existing Windows VMs on Red Hat OpenShift Virtualization.(DH2i 用例) > KServe, for example, which is available in OpenShift AI, OpenShift, and now in MicroShift too…(RH 博客,无版本) > Casey's General Stores … deployed SC//Reliant™ Manager Edge Computing as a Service across more than 2,600 stores in just ten months.(Scale PR 09-24,补录) ### 5. 媒体快讯(轻)· 奇客 / 至顶网 - 归档:`industry/奇客资讯网-ed448ea7.txt`、`industry/至顶网-af3b259d.txt` - 口径:**中文媒体转述/频道摘要**;非 HCI。 > 因涌入大量无效的 AI bug 报告,Google 宣布冻结其 Bug 悬赏计划 Open Source Software Vulnerability Reward Program (OSS VRP),该决定于 10 月 1 日生效。Google 承诺将于 2027 年第一季度提供相关更新… > Google 的太空 AI 数据中心项目称之为 Project Suncatcher,最新发射的卫星…安装了 4 个 Tensor Processing Units 芯片…因热管理约束每次只运行 15 分钟。 > 微软宣布 WSL Containers GA,该工具为 Windows 11 开发者提供了一种通过 Windows Subsystem for Linux 构建、运行和部署 Linux 容器的内置方案。 > OpenAI ChatGPT 与 Codex 负责人 Tibo Sottiaux 在 DevDay 当天接受专访,谈新产品 Dots 背后的判断:手工搭建的智能体循环终将被淘汰…(至顶网频道摘要) ### 6. 续档 / 空壳 / blocked | 条目 | 状态 | |---|---| | DGX Spark 64GB / 128GB 涨价(STH) | **续档**(10-05 中强 NEW;正文无变化) | | BiCS NAND 展台 / Thales Quantum & AI / F5 BNK 赞助实验室 / Venice·EPYC 9006×Agentic / Joey / Digital Realty 40%/50%/37% / Server Virt MQ(VR 09-23,本地无 PDF)/ Gartner 工业 AI agents / Blueprint Alliance / OpenAI Plugins / TRX50 | **续档**(本日再抓正文无变化) | | Blocks&Files / The Register / NextPlatform | host_blocked,0-byte → **不摘、不发明标题** | | 科技行者 / 至顶网服务器频道 | 频道壳,无变化或非 HCI → **跳过** | | Reddit | host_blocked / 403 | ### 7. 对规划一句 **本日无公开战略级产业新稿:续抬「Xeon 6+ Clearwater Forest 广泛供货 × 内存涨价传导」并更正——Xeon 6+ 不支持 MRDIMM(12ch DDR5),高核密度 HCI 节点每核带宽按普通 DDR5 估算;旁证 VMware「GPU 推理 VM 少配 CPU/内存、共置业务」自测话术 + Forrester 委托 TCO、XCP-ng 存储迁移提速 GA 补丁、Kioxia CM9-R 双端口 E3.S 评测(轻–中)。**