Lava reported finding roughly 2,100 internet-accessible NVIDIA DCGM Exporter hosts exposing metrics for more than 12,000 unique GPUs in scans conducted from March through May 2026. Separately, NVIDIA disclosed CVE-2026-47483, a high-severity flaw in DCGM Exporter’s Go profiling endpoints that can enable resource exhaustion and information disclosure. The scan describes a time-bounded exposure, not a live count; the vulnerability requires its own software and configuration response.
What was exposed—and what the findings mean
NVIDIA DCGM Exporter collects selected GPU telemetry and serves it over HTTP, commonly so a Prometheus monitoring system can scrape it. Lava found that the observed hosts returned metrics without authentication over plaintext HTTP. Those readings can reveal GPU models and UUIDs, utilization, memory use, power, temperature, errors and driver details. Labels may also offer clues about workloads or projects, while repeated readings can expose usage patterns.
As an Amazon Associate I earn from qualifying purchases.
That information can help an outsider map infrastructure and identify operational patterns. It does not, by itself, establish access to model weights, training data, or control of the server.
Lava’s four scans between March and May 2026 identified roughly 2,100 hosts associated with nearly 300 organizations and more than 12,000 unique GPU UUIDs. Lava estimated that the observed hardware represented about $100 million; that is its estimate, not an independently audited valuation. These results are observations from those scans, not a census of all exposed systems or proof that the hosts remained exposed after disclosure. Lava’s report provides its findings and methodology.
#1 Best Overall
- [ Maximum AI Compute Power ] Dominate complex workloads with the ASUS ESC8000A-E13. This 4U rack server is a powerhouse engineered for mass-scale AI, machine learning, and deep training. Featuring support for dual AMD EPYC 9005/9004 processors and up to eight dual-slot GPUs, it delivers the raw computational muscle required to train LLMs and run complex simulations effortlessly. Accelerate your data science pipeline and transform raw data into actionable intelligence faster than ever.
- [ Advanced Thermal Efficiency ] High performance demands elite cooling. The ESC8000A-E13 features a cutting-edge aerodynamic design with independent CPU and GPU airflow tunnels. Equipped with redundant hot-swap fans and optimized for liquid cooling integrations, this 4U server ensures maximum uptime under heavy, sustained workloads. Keep your data center running cool, quiet, and highly efficient while preventing thermal throttling during mission-critical enterprise operations.
- [ Scale with Flexible Storage ] Future-proof your infrastructure with unmatched storage and expansion flexibility. This offers comprehensive front-panel drive bays supporting Gen5 NVMe, SAS, or SATA drives alongside multiple PCIe 5.0 slots. Designed as a high-density 4U server capable of housing eight dual-slot GPUs: NVD H200, RTX PRO 6000 Blackwell, RTX PRO 4500 Blackwell or AMD Instinct MI350P PCIe Card, each supporting up to 600 watts.
- [ Enterprise-Grade Reliability ] Minimize downtime and secure your ecosystem with server-grade redundancy. The ESC8000A-E13 is built for 24/7 continuous operation, boasting 2+2 redundant (3200W total) 80 PLUS Titanium power supplies and integrated ASUS ASMB11-iKVM for comprehensive out-of-band management. Ideal for cloud service providers, rendering farms, and large enterprise infrastructure, it combines robust physical hardware with smart remote monitoring to safeguard your digital assets.
- [Reliability Guaranteed] Shop with total peace of mind knowing that every new computer component we sell is backed by our EPC 3-year warranty. Whether you are investing in high-speed DDR5 RAM or a powerhouse GPU, we protect your build against defects and performance failures. We stand firmly behind the quality of our hardware, ensuring that your setup remains fast, stable, and secure for years to come.
Where Lava observed the GPUs
The operator categories below are Lava’s classifications of the exposed GPU set it observed. They are not market-wide shares.
| Operator category | Share of observed exposed GPUs |
|---|---|
| Consumer GPU and mining farms | 35% |
| Neoclouds | 25% |
| General hosting and colocation | 19% |
| Telecoms and national ISPs | 10% |
| Hyperscalers | 6% |
| Universities and research institutes | 5% |
Lava’s scan observations counted 5,274 GPUs in the United States (44%), 2,054 in Romania (17%), and 1,967 in China (16%). These are scan results, not current national inventories. Lava also reported that roughly 60% of GPUs showed 0% utilization in each of its four scans; the snapshot does not establish why they were idle.
Rank #2
- NVIDIA Volta GV100 Architecture — 4,608 CUDA Cores, 640 1st-Gen Tensor Cores delivering 14 TFLOPS FP32 and 112 TFLOPS deep learning performance for AI training, inference, HPC, and scientific computing workloads
- 32GB HBM2 ECC Memory — 900 GB/s Bandwidth — High-bandwidth memory on a 4096-bit bus with ECC error correction provides the memory capacity and throughput required for the largest AI models, simulations, and datasets
- PCIe 3.0 x16 Interface — 250W TDP — Standard PCIe Gen3 connectivity with passive cooling designed for enterprise rack server deployment in HPE ProLiant, Dell PowerEdge, and Supermicro platforms with adequate chassis airflow
- NVLink — Scale to 96GB Unified Memory — Connect two V100 GPUs via NVLink at 300 GB/s bi-directional bandwidth to scale GPU memory from 32GB to 96GB for larger AI training and HPC workloads
- Multi-Precision Computing — Supports FP64 (7 TFLOPS), FP32 (14 TFLOPS), FP16 (112 TFLOPS) and INT8 precision modes for flexible deployment across training, inference, and scientific simulation workloads
How the exporter flaw differs from public exposure
Publicly reachable metrics and CVE-2026-47483 are related risks, but they are not the same problem. Metrics exposure is an access-control and network-configuration issue: an endpoint is reachable by people or systems that should not be able to query it. The CVE is a software vulnerability in DCGM Exporter’s Go /debug/pprof/ profiling endpoints.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
NVIDIA says concurrent unauthenticated profiling requests can cause uncontrolled resource consumption, with potential denial of service and information disclosure. Lava reproduced resource exhaustion in a controlled environment and warned that CPU and memory pressure could slow training or inference on the same host, especially without strict resource limits. It did not test the behavior against the public systems it scanned, and reported no observed workload disruption across those hosts. NVIDIA’s July 2026 security bulletin, updated September 3, 2026, assigns the issue CVSS v3.1 8.2, High.
Rank #3
- AI-Optimized: Designed to support up to 4 GPUs, it is perfect for handling intensive AI and machine learning tasks, ensuring high performance and scalability for advanced computational needs.
- Intelligent Storage: Equipped with 8 hot-swappable 3.5" SATA/SAS drives (12Gbps), featuring SGPIO and temperature control, it ensures efficient data management and reliable storage performance.
- Robust Cooling: The system includes 3x 12038 hot-swap PWM fans and 2x 8038 rear fans, providing advanced thermal management to maintain optimal temperatures and ensure stable operation under heavy workloads.
- Rack-Ready: Comes with a pre-installed rail kit, allowing for quick and easy installation in standard 19-inch server racks, making it ideal for data center environments and enterprise setups.
- Versatile Connectivity: Offers USB 3.0 and the latest USB 3.2 Type-C ports, ensuring high-speed data transfer and compatibility with a wide range of peripherals and devices for enhanced connectivity options.
Which versions NVIDIA lists as affected
NVIDIA’s bulletin gives separate version ranges for DCGM and DCGM Exporter. Check which component is actually deployed; updating one does not establish that the other is current.
| Component | Affected versions listed by NVIDIA | Updated version listed |
|---|---|---|
| DCGM | 0.0 through 4.5.2 | 4.5.3 |
| DCGM Exporter | 0.0 through 4.8.2 | 4.8.2 |
The version guidance above is from NVIDIA’s bulletin. NVIDIA says to clone or update the software from its DCGM Exporter repository. Confirm the exact deployed component, compatibility, and current vendor instructions before rollout.
Rank #4
- Professional AI & Creator Workstation: AMD Radeon AI PRO R9700 GPU with 32GB GDDR6 is engineered for AI development, professional content creation, and compute-intensive workloads.
- Massive 32GB Memory Capacity: 32GB of GDDR6 memory on a 256-bit bus provides ample bandwidth for large AI models, 8K video editing, and complex 3D rendering.
- Advanced RDNA 4 with AI Accelerators: 64 Compute Units with 3rd Gen Ray Tracing and dedicated 2nd Gen AI Accelerators for groundbreaking AI performance and visual computing.
- Professional Blower Cooling: Efficient single blower design exhausts heat directly out of the chassis, ideal for multi-GPU workstation and server configurations.
- Enterprise-Grade Thermal Solution: Vapor chamber heatsink with industrial Honeywell PTM7950 thermal interface material ensures reliable cooling under sustained professional loads.
What operators and customers should do
- Inventory the deployment. Identify DCGM and DCGM Exporter versions, where each runs, which interfaces and ports serve telemetry, and whether profiling endpoints are enabled.
- Update affected components. Apply NVIDIA’s listed updates—DCGM 4.5.3 and DCGM Exporter 4.8.2—after checking compatibility and current vendor guidance.
- Remove unnecessary public reachability. Where possible, bind exporters to loopback or private interfaces. Use firewall rules, security groups, or equivalent network controls so only authorized monitoring infrastructure can connect.
- Keep profiling off unless it is needed. Lava says profiling is opt-in in current versions and recommends not enabling
--enable-pprofunless there is an explicit operational need. Verify the setting and behavior for the deployed release. - Protect the monitoring layer. Restrict access to Prometheus query APIs and target pages as well as exporter endpoints; an internal monitoring service can expose useful infrastructure details if its own access is open.
- Set limits and alerts. Apply resource limits where appropriate and monitor for unusual CPU or memory pressure and service failures, so a monitoring or profiling incident is less likely to spill into GPU workloads.
How responsibility can be shared
Lava said exposed services appeared on infrastructure associated with providers including Voltage Park, Lambda, Northern Data, and DigitalOcean, among others. It also reported that providers told Lava services were primarily deployed and exposed by customers, and that provider security teams helped notify customers and drive remediation. An endpoint’s association with a provider does not establish that the provider configured it incorrectly.
In a hosted GPU environment, the practical split depends on who controls the operating system, exporter deployment, network rules, and monitoring stack. Providers should clarify which controls they manage; customers should verify the controls they own rather than assume that hosted infrastructure makes telemetry private. Both sides need an agreed path for reporting and closing exposure.
Quick Recap
A focused remediation check
- Are DCGM and DCGM Exporter versions known, and have the relevant NVIDIA updates been applied?
- Can a public network reach metrics or profiling endpoints? If so, is there a documented, controlled reason?
- Can only authorized monitoring systems connect, and are Prometheus query and target interfaces also restricted?
- Is profiling enabled only when operationally necessary?
- Are resource limits and alerts in place for the exporter and the host?
- Is it clear whether the customer or provider owns each configuration and remediation step?
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.
