---
格式版本: 2
标题: "Dell's new PowerEdge XE8812 crams 144 NVIDIA Vera Rubin GPUs into a single rack | iTWire"
原文链接: "https://itwire.com/business-it-news/enterprise-solutions/dells-new-poweredge-xe8812-crams-144-nvidia-vera-rubin-gpus-into-a-single-rack"
发布日期: "未标注"
发现时间: "2026-06-23T08:31:12+08:00"
入库时间: "2026-07-07T09:39:35.705Z"
来源平台: "Exa"
搜索渠道: "exa_search"
搜索词: "AI Rack"
匹配关键词:
  - "GPU"
  - "Liquid Cooling"
  - "NVL72"
  - "Nvlink"
相关厂家:
  - "NVIDIA"
相关专家:
  []
内容类型: "网页"
抓取工具: "XCrawl Scrape"
清洗工具: "XCrawl Markdown + LLM 正文裁剪"
原始附件:
  []
图片摘要:
  - "✗ ./assets/img-02994806.webp | photo | 装饰性配图，内容为产品实拍展示，与正文技术细节无关"
  - "✓ ./assets/img-07a74e1f.webp | photo | Dell PowerEdge XE8812 机架正面实拍图，展示内部144块Vera Rubin GPU布局"
  - "✓ ./assets/img-5c20c992.webp | photo | Dell XE8812 整排机架实拍图，展示机架集群部署形态"
  - "✗ ./assets/img-555c1ca0.webp | photo | 装饰性配图，内容为产品局部实拍，无技术价值"
  - "✓ ./assets/img-5254656c.webp | photo | Dell XE8812 单台设备实拍照，展示机箱外观与正面接口"
  - "✗ ./assets/img-f0432946.webp | infographic | 推广性信息图，展示Dell AI Factory品牌标志和合作方Logo，无正文技术数据"
  - "✗ ./assets/img-4deedabe.webp | infographic | 品牌宣传信息图，展示Dell和NVIDIA合作标识及标语，无实质技术信息"
AI优质: "是"
AI打分: 98
AI分档: "高置信优质"
AI质检状态: "通过"
AI打分理由: "Dell PowerEdge XE8812单机架144组Vera Rubin GPU，全液冷无风扇，OCP ORv3标准，有客户案例（NERSC、InstaDeep等）及部署时间，技术细节丰富，商业信号强，来源专业媒体，时效性高。"
AI质检模型: "deepseek-v4-flash"
AI质检时间: "2026-07-30T01:19:01+08:00"
AI主题相关性: 20
AI来源权威性: 13
AI新颖性: 20
AI技术细节: 20
AI商业部署信号: 15
AI完整性: 10
采集批次: "2026年7月7日16点36分44秒"
采集批次ID: "20260707-163644-932"
去重键: "https://itwire.com/business-it-news/enterprise-solutions/dells-new-poweredge-xe8812-crams-144-nvidia-vera-rubin-gpus-into-a-single-rack"
---

![Dell AI Factory with NVIDIA](https://media.itwire.com/2026/06/23/dells-new-poweredge-xe8812-crams-144-nvidia-vera-rubin-gpus-into-a-single-rack-726e3da894b4.webp)

Dell has pushed its AI Factory deeper into supercomputer territory, and the server at the centre of it is a dense, fanless, fully liquid-cooled machine built for the hardest science on the planet. There's an Australian thread running through it too.

[Dell Technologies](https://dell.com.au/) has unveiled the PowerEdge XE8812, a new server built for HPC and AI work that makes ordinary data centre gear sweat.

It's the latest addition to the [Dell AI Factory with NVIDIA](https://www.dell.com/en-us/lp/dt/nvidia-ai), and it rides on NVIDIA's new Vera Rubin NVL4 architecture to pack up to 144 GPUs into a single rack.

Dell announced it out of Sydney on 23 June, a small reminder that this race isn't only a US-and-China story.

The timing isn't accidental. Dell cites projections that AI investment will grow 44% year-on-year in 2026, and says 87% of organisations now call innovation key to their strategy, with 75% saying the same about AI. When the demand curve looks like that, incremental upgrades stop cutting it.

![](https://media.itwire.com/2026/06/23/dells-new-poweredge-xe8812-crams-144-nvidia-vera-rubin-gpus-into-a-single-rack-inline-1-e61770f6f485.webp)

This infographic independently created by iTWire using Gemini Nano Banana

**What's actually new here**

The headline number is density. Dell reckons the XE8812 will be one of the industry's densest platforms, with up to 144 GPUs, 300kW+ of power and 100% direct liquid cooling on both CPUs and GPUs. No fans anywhere.

The jump comes from swapping NVIDIA's GB200 NVL4 for the newer [Vera Rubin NVL4](https://nvidianews.nvidia.com/news/nvidia-vera-rubin-delivers-world-class-supercomputers-for-science). That shift brings more host memory, more GPU memory, more compute, and a core count that climbs from 144 to 176 per node.

Dell's pitch is 50% more memory per socket and per GPU than the prior generation. That reads like a spec-sheet brag until you think about what it buys you.

More memory means researchers can run bigger models and simulations entirely in-memory. No staging data from storage, no swapping it in and out.

Both of those tricks add latency measured in microseconds to milliseconds and gut your effective bandwidth. For molecular and multi-physics work, that's the gap between a result today and a result next week.

![260623 - DT - ISC Image XE8812_row.webp](https://media.itwire.com/2026/06/23/dells-new-poweredge-xe8812-crams-144-nvidia-vera-rubin-gpus-into-a-single-rack-inline-6-b5fd83e19a6e.webp)

A row of Dell XE8812 servers

**Open rack, less pain to run**

The XE8812 sits in the Dell PowerRack 9100, built on the OCP-backed ORv3 open standard. Open standards here mean modular deployment and better efficiency, rather than a proprietary box you can't service.

Once it's running, Dell's iDRAC controller lets IT teams deploy, update and monitor the servers remotely. Rack-level telemetry comes through the Dell Integrated Rack Controller and OpenManage Enterprise, including automated leak detection (which matters a lot when you've got coolant running through every CPU and GPU).

The turnkey angle is the PowerRack itself: factory-integrated, pre-validated racks that, with Dell ProDeploy white-glove service, can go from delivery to live workloads in just over 6 hours. That's Dell's figure, not mine, but if it holds up it's a serious cut to the usual integration slog.

![](https://media.itwire.com/2026/06/23/dells-new-poweredge-xe8812-crams-144-nvidia-vera-rubin-gpus-into-a-single-rack-inline-2-aa12dbb605ed.webp)

This infographic independently created by iTWire using Gemini Nano Banana

**Who's already buying in**

Dell says more than 5,000 customers are now deploying the AI Factory globally. A few marquee names show the range.

In the US, Dell, NVIDIA and NERSC are [building Doudna](https://www.nersc.gov/news-and-events/news/doe-announces-new-supercomputer-powered-by-dell-and-nvidia-to-speed-scientific-discovery), the next flagship US Department of Energy supercomputer at Lawrence Berkeley National Laboratory. It runs on XE8812 servers with Vera Rubin NVL4 and [NVIDIA Quantum-X800 InfiniBand](https://www.nvidia.com/en-us/networking/products/infiniband/quantum-x800/) networking. It's named after Jennifer Doudna, the Berkeley biochemist who shared the 2020 Nobel Prize in Chemistry for CRISPR, and NERSC expects it to deliver at least 10x the performance of its current Perlmutter system.

In France, Dell and NVIDIA are helping AI company [InstaDeep](https://www.delltechnologies.com/asset/en-us/solutions/business-solutions/customer-stories-case-studies/instadeep-case-study.pdf) scale its Kyber cluster to roughly 0.5 exaFLOPs of FP16 performance, used for jobs like automated design of printed circuit boards, the parts inside everything from phones to factory gear.

In the UK, the Wellcome Sanger Institute is running Dell PowerEdge XE-Series servers with NVIDIA GPUs to [sequence DNA](https://www.dell.com/en-us/blog/wellcome-sanger-institute-accelerates-genomic-discovery-with-dell-technologies/) at a pace that's hard to picture: a fully assembled genome every 7 hours, more than 100 petabytes of curated genetic data on-premises, and over 70% of the genomes in the global Earth BioGenome Project.

![260623 - DT - ISC Image XE8812_single.webp](https://media.itwire.com/2026/06/23/dells-new-poweredge-xe8812-crams-144-nvidia-vera-rubin-gpus-into-a-single-rack-inline-7-62ab53ba43e4.webp)

A single Dell XE8812 server

**The Australian angle**

Closer to home, [Monash University](https://www.monash.edu/news/articles/monash-university-to-build-australian-first-supercomputer-maveric-with-global-technology-partners) has built MAVERIC with Dell, NVIDIA and [CDC Data Centres](https://cdc.com/resources/news/monash-university-cdc-nvidia-dell-technologies-to-build-maveric-australia-s-first-higher-education-ai-supercomputer/). It's the country's first higher-education AI supercomputer, using liquid-cooled Dell PowerRack systems with PowerEdge XE9712 servers and NVIDIA GB200 NVL72.

The roughly AU$60 million machine sits in CDC's Brooklyn facility in Melbourne and targets cancer detection, climate work and genomics. CDC's cooling is the quiet headline: its data centres are designed to run with no water consumption for primary cooling, which in a dry country is no small thing.

MAVERIC runs the current GB200 generation, not the new Vera Rubin gear. Still, it proves the same Dell-plus-NVIDIA recipe is already landing on Australian soil, and it gives local researchers a front-row seat for whatever the XE8812 generation brings next.

![](https://media.itwire.com/2026/06/23/dells-new-poweredge-xe8812-crams-144-nvidia-vera-rubin-gpus-into-a-single-rack-inline-3-7125ce1f43c7.webp)

This infographic independently created by iTWire using Gemini Nano Banana

**Dell's got company**

The announcement leaves out the obvious context: Dell is far from alone in racing to wrap NVIDIA's Rubin silicon in something you can actually buy.

[Supermicro](https://www.hpcwire.com/off-the-wire/supermicro-delivers-nvidia-vera-rubin-nvl4-end-to-end-dcbbs-blueprint-with-native-fp64-performance-for-converged-hpc-and-ai-infrastructure/) got there a day earlier. On 22 June it announced its own Vera Rubin NVL4 blueprint, part of its Data Center Building Block Solutions line, also hitting 144 GPUs per rack and leaning hard on FP64 double-precision for science. Same chips, same density target, different packaging philosophy.

[HPE](https://www.hpe.com/us/en/newsroom/press-release/2026/03/hpe-unveils-next-generation-ai-factory-and-supercomputing-advancements-with-nvidia.html) is coming at it from the rack-scale end. Its Vera Rubin NVL72 by HPE system is due in December 2026, alongside new Cray GX5000 and GX240 gear, aimed more at trillion-parameter frontier models than in-memory simulation. Lenovo is bringing its Neptune liquid cooling to the Rubin platform too.

So what's Dell's edge? Three things stand out: the ORv3 open-standard rack, the management stack, and marquee wins like Doudna that prove the thing works at national-lab scale. Dell has also leaned on liquid cooling for more than a decade, its Infrastructure Solutions Group posted 80% year-on-year growth in a recent quarter, and the company landed at number 5 on Time's 10 Most Influential Hardware Companies of 2026.

The NVL4-versus-NVL72 split is the bit to watch. NVL4 ganged up to 144 GPUs a rack is tuned for HPC and simulation that wants huge coherent memory. NVL72's 72-GPU NVLink domain is built for giant AI training runs. Most big labs will end up wanting both.

**What Dell and NVIDIA are saying**

Dell Technologies senior vice president of Compute and Networking [Arun Narayanan](https://www.linkedin.com/in/arunkumar-narayanan-6a68ba/) framed it around the science.

"The institutions doing the world's most important research like decoding the human genome, modeling the energy systems of the future and building the sovereign AI infrastructure that nations depend on deserve infrastructure that matches the ambition of their work," he said. "The Dell PowerEdge XE8812 reflects Dell's commitment to pushing the boundaries of what's possible, giving these organisations the density, memory and open architecture they need to tackle workloads that once seemed impossible."

NVIDIA vice president of Enterprise Platforms [Chris Marriott](https://www.linkedin.com/in/chris-marriott-44036a5/) put the focus on convergence.

"The convergence of AI and HPC is redefining what organisations should expect from their infrastructure," he said. "Dell and NVIDIA are raising that bar together, combining NVIDIA Vera Rubin NVL4 architecture and CUDA-X libraries with Dell's engineering and at-scale deployment expertise to provide the performance, efficiency and openness required for the world's most demanding AI and scientific computing workloads."

![Dell detailed infographic.webp](https://media.itwire.com/2026/06/23/dells-new-poweredge-xe8812-crams-144-nvidia-vera-rubin-gpus-into-a-single-rack-inline-5-82eddbe2620a.webp)

This infographic independently created by iTWire using Gemini Nano Banana

**When can you get one**

The PowerEdge XE8812 will be globally available early next year, so early 2027 in practice. That lines it up against HPE's December 2026 NVL72 and Supermicro's freshly announced blueprint.

The bigger story is that "supercomputer" is quietly turning into a product you order rather than a one-off you build. Dell's selling pre-validated racks that boot live workloads in an afternoon, NVIDIA's shipping the same Rubin silicon to every server maker willing to cool it, and national labs and universities (Monash included) are the ones writing the cheques.

The density race won't slow down. It's 144 GPUs a rack today, and Rubin's successor is already sketched on someone's whiteboard. Dell can clearly build the things. The open question is whether the grid, and the water table, can keep up.

*Further reading: Dell's full announcement is on its* *[investor newsroom](https://investors.delltechnologies.com/news-releases/news-release-details/dell-ai-factory-nvidia-advances-supercomputing-class)**, and* *[TechPowerUp](https://www.techpowerup.com/350175/dell-introduces-poweredge-xe8812-server-with-nvidia-vera-rubin-nvl4-architecture)* *has a tidy spec rundown. Connect with Dell on* *[X](https://x.com/Dell)* *and* *[LinkedIn](https://www.linkedin.com/company/delltechnologies/)**.*