CoreWeave demonstrated a Dell PowerEdge XE9712 rack built around NVIDIA’s GB200 NVL72 platform in December 2024, and Dell formally confirmed the shipment on December 9. The system combines 72 interconnected Blackwell GPUs, Grace CPUs, liquid cooling, high-speed networking and CoreWeave’s cloud software stack. It was a rack-scale AI infrastructure milestone—not a conventional standalone server launch.
What CoreWeave and Dell announced
StorageReview reported the live demonstration on December 2, 2024, at a Switch data center. Dell’s formal announcement followed on December 9, saying it was shipping the first liquid-cooled PowerEdge XE9712 racks with NVIDIA GB200 NVL72 systems to CoreWeave.
The word “first” needs qualification. Dell’s claim concerned its first shipment of this specific XE9712/GB200 NVL72 rack solution to CoreWeave. It should not be interpreted as proof that Dell produced the first GB200 system worldwide.
The platform consisted of Dell PowerEdge XE9712 server systems integrated into Dell’s IR7000 liquid-cooled rack infrastructure, with NVIDIA Quantum-2 InfiniBand networking and CoreWeave’s Kubernetes, scheduling and observability software.
The Tool Desk
Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →#1 Best Overall
- Dell PowerEdge R730xd 24B SFF 2U Server
- 2x Intel Xeon E5-2690 v4 2.6Ghz 14-Core (28-cores Total)
- 128GB DDR4 RAM – 4x 1.2TB 10K SAS 2.5” 12Gb/s
- Dell H730P mini 2GB 12Gb/s RAID
- 2x 750W PSU - 2x 10Gb SFP+ 2x 1Gb (RJ45) NIC
CoreWeave’s announcement described Dell Professional Services, factory integration and testing, and ProSupport One as part of the deployment.
System specifications at a glance
| Component | Role |
|---|---|
| NVIDIA GB200 | Grace Blackwell superchip platform |
| NVL72 | Rack-scale configuration linking 72 Blackwell GPUs through a high-bandwidth NVLink fabric |
| Dell PowerEdge XE9712 | Server platform carrying the GB200 NVL72 architecture |
| Dell IR7000 | Integrated 21-inch liquid-cooled rack infrastructure |
| Grace CPUs | Up to 36 NVIDIA Grace CPUs in Dell’s stated configuration |
| Quantum-2 InfiniBand | High-speed cluster networking |
| CoreWeave software | CoreWeave Kubernetes Service, SUNK, Mission Control and observability |
Dell described the XE9712 as supporting up to 72 NVIDIA Blackwell GPUs and 36 Grace CPUs. In an NVL72 system, those accelerators are designed to operate as a tightly coupled domain rather than as 72 loosely connected PCIe cards.
What “GB200 NVL72” means
GB200 refers to NVIDIA’s Grace Blackwell superchip platform. NVL72 identifies a rack-scale design built around 72 Blackwell GPUs connected through NVLink.
The important feature is the interconnect. Large-model training and inference often require GPUs to exchange model weights, activations and other data rapidly. A high-bandwidth scale-up fabric can reduce communication bottlenecks compared with treating each accelerator as an isolated device.
Dell said the 72-GPU NVLink domain could function as a single GPU for real-time trillion-parameter large-language-model inference. That is a vendor positioning claim, not an independently verified result established by the cited sources.
Why liquid cooling matters
A rack containing dozens of high-performance accelerators produces far more heat than a conventional air-cooled server rack. Direct-to-chip liquid cooling transfers heat more efficiently than air alone, but it also changes the facility requirements.
The deployment needs cooling distribution units, compatible facility water loops, leak detection, water-quality controls, service procedures, high-density power delivery and trained technicians. Power, cooling and networking therefore become part of the computing platform rather than separate data-center details.
Rank #2
- Renewed server with the highest quality standards
- Ideal for a robust enterprise environment or data center
- All servers include power cords, and other parts detailed in full product description below
- Custom configurations available upon request
Dell said its IR7000 rack was designed for native liquid cooling and could support future deployments of up to 480 kW. Dell also claimed it could capture nearly 100% of generated heat. Those are Dell specifications; a 480-kW cooling capability must not be confused with the rack’s normal or demonstrated power consumption.
Recommended Free Tools
StorageReview said the demonstration took place in Rob Roy’s Evo Chamber, described as providing 1 MW of power and cooling per rack, including 250 kW of air-cooling capacity and 750 kW of direct-to-chip liquid-cooling capacity. That describes the demonstration environment, not necessarily the operating draw of every production CoreWeave rack.
What the live demonstration showed
According to StorageReview’s report, the demonstration included:
- NCCL All-Reduce: a test of communication performance across the 72-GPU NVLink domain.
- GPU Blaze: a matrix-multiplication workload used to show accelerator throughput.
- Megatron training: a live model-training run using CoreWeave’s SUNK, or Slurm-on-Kubernetes.
- Cooling response: the in-rack cooling distribution unit adjusted cooling output as GPU workloads changed.
- Power monitoring: a dashboard displayed the system’s energy requirements during operation.
These demonstrations showed system bring-up, interconnect behavior, workload execution and thermal-management coordination. They did not establish independent production benchmarks, customer economics, total cost of ownership or superiority over competing systems.
When customers could use the platform
The demonstration and shipment were not the same as cloud availability. CoreWeave’s release note states that GB200 NVL72-powered cloud instances became available on February 3, 2025, initially through CoreWeave Kubernetes Service in the US-WEST-01 region.
CoreWeave’s documentation listed the platform’s claimed figures as up to 1.4 exaFLOPS of AI compute and 13.4 TB of fifth-generation, NVLink-connected GPU memory per rack. It also reported training improvements of up to four times versus previous-generation GPU instances.
Those figures are CoreWeave’s published specifications and claims. The cited release note does not provide a universal public hourly price. Current regions, quotas and commercial terms should be confirmed directly with CoreWeave because the original release note is historical and the page was later modified.
Rank #3
- Dell 13th Generation Rack Mount 1U 8-Bay 2.5" SFF Server
- Enterprise Server For Home Use
- 2x Intel Xeon Processor E5-2690 v4 2.60GHz 14-Core CPUs
- 128GB PC4-2133 DDR4 Memory
- 2x 1TB 2.5" SATA SSDs - Solid State Drives -
What the numbers do—and do not—tell buyers
| Documented or observed | Qualification |
|---|---|
| 72 Blackwell GPUs and up to 36 Grace CPUs | Dell’s rack-level configuration |
| NCCL, GPU Blaze and Megatron demonstrations | Reported live demonstrations, not independent certification |
| Up to 1.4 exaFLOPS per rack | CoreWeave’s published claim; precision and measurement context are not specified in the cited release note |
| 13.4 TB of memory per rack | CoreWeave’s published rack-level figure |
| Up to four-times training improvement | CoreWeave claim against previous-generation GPU instances |
| 25-times efficiency and 30-times inference positioning | Dell product-positioning claims, not independently reproduced results in the cited sources |
Who benefits from an NVL72 system?
The likely workloads include foundation-model training, large-model fine-tuning, real-time inference, scientific computing and tightly coupled engineering simulations.
This is not aimed at individual developers, ordinary business applications, general-purpose virtualization or small jobs that need one GPU. The practical customer is a cloud provider such as CoreWeave—or a large enterprise with substantial capital, power, cooling, networking and operations expertise.
Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchPC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11Rack-scale computing can be valuable when a workload keeps many accelerators busy and benefits from fast GPU-to-GPU communication. It can be uneconomical when jobs are intermittent, poorly parallelized or too small to use the rack effectively.
Infrastructure and operational trade-offs
- Facility readiness: Buyers must validate utility capacity, redundancy, rack power density, cooling-water availability, CDUs, floor loading, cabling and maintenance access.
- Large minimum footprint: A rack-scale system commits substantial capital before utilization is proven.
- Scheduling complexity: Job placement and allocation become more important when workloads depend on a complete high-performance fabric.
- Liquid-cooling operations: Pumps, CDUs, connectors, water quality, leak detection and service procedures introduce requirements beyond ordinary server maintenance.
- Vendor dependence: Integrated factory testing and professional services can reduce deployment risk but may increase reliance on Dell, NVIDIA and CoreWeave.
- Pricing opacity: The cited sources do not provide a public hourly rate or a complete cost-of-ownership model.
For many organizations, managed CoreWeave access is more practical than constructing a private liquid-cooled rack. Even then, buyers should compare utilization, reservation commitments, regional availability, networking requirements and portability before committing.
What happened next
By 2026, the original GB200/XE9712 deployment was no longer the newest generation of the partnership. Dell later described GB300 NVL72 systems delivered to CoreWeave and subsequent Vera Rubin-based systems using XE9812 servers.
Those later systems should not be confused with the December 2024 announcement. The original story is specifically about the Dell XE9712 and NVIDIA GB200 NVL72 combination.
Quick wins for a faster PC:
Scan for outdated or missing drivers - takes under a minuteDriver Scan →Clear out junk files and repair common Windows errorsFree Scan →Sources: Dell AI Factory announcement, CoreWeave GB200 NVL72 release note, and Dell investor materials.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
