CoreWeave says NVIDIA Vera Rubin NVL72 is in limited availability on CoreWeave Cloud, with Cognition the first customer running production workloads. Cognition reports up to 4.8× total token throughput on its SWE-2 inference workload versus a GB200 NVL72 baseline. That result is specific to Cognition’s workload and comparison; it is not a general performance guarantee.
What CoreWeave has put into production
CoreWeave’s September 30, 2026 announcement describes Vera Rubin NVL72 as in limited availability, not general availability. The company says hundreds of Rubin GPUs have been deployed across multiple regions and that customer workload onboarding is underway. Cognition is identified as the first customer running production agentic AI workloads on the platform. CoreWeave’s announcement does not provide public pricing or a general availability date.
As an Amazon Associate I earn from qualifying purchases.
One Vera Rubin NVL72 rack pairs 72 Rubin GPUs with 36 Vera CPUs, according to CoreWeave’s system description. For larger deployments, CoreWeave says hundreds of accelerators can be connected using Spectrum-X Ethernet. The company also says customers can use the same operating model and tooling as for its existing GB200 and GB300 NVL72 fleets.
What Cognition’s 4.8× result measures
Cognition reports up to 4.8× total token throughput for SWE-2 inference compared with GB200 NVL72. CoreWeave calls this an independent benchmark run by Cognition on CoreWeave Cloud. Its chart describes the inference comparison at matched interactivity and per GPU, but the published material does not provide enough methodological detail to extend the result to other models, configurations, or latency targets. See CoreWeave’s investor release.
#1 Best Overall
- 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
The same release reports a separate Cognition result: 3.8× output-token throughput for reinforcement-learning workloads versus GB200 NVL72. This is a different workload and metric from the 4.8× SWE-2 inference result, so the figures should not be combined or treated as interchangeable measures of overall system speed.
NVIDIA says Cognition sampled tasks from FrontierCode for its early software-engineering workload benchmark. That context helps explain the focus on agentic coding, but does not make the result representative of all coding agents or AI workloads. As Cognition’s Silas Alberti put it, “Agentic coding is an unforgiving workload that requires long contexts, high concurrency and rapid reasoning.”
Rank #2
- GPU-Modell: Gefoce RTX 3080
- Memory Type: GDDR6X Memory Capacity: 20GB Memory Bus Width: 320bit Output Interfaces: 3*DP + HDMI Core Clock: 1710MHz Memory Clock: 19Gbps Power Interface: 8+8pin Recommended Power Supply: 850W or higher
What Vera CPU adds
NVIDIA says CoreWeave will also offer Vera CPU, a CPU platform positioned for the orchestration and execution work surrounding AI agents. In its September 30 post, NVIDIA describes a CoreWeave configuration with 128 CPUs and 11,264 cores in one rack. NVIDIA reports that CoreWeave testing found more than 3× faster agent-sandbox startup and a 1.7× performance gain across all passing Terminal-Bench tasks. These are claims reported by NVIDIA about CoreWeave tests, not independently audited results in the cited material. Details appear in NVIDIA’s CoreWeave announcement.
Free tools Windows power users keep installed
One-click scans. No signup required.
That 128-CPU configuration is distinct from a separate Vera CPU rack description in NVIDIA’s Vera CPU announcement, which specifies 256 liquid-cooled CPUs and says the rack is intended to support more than 22,500 concurrent CPU environments. The two rack descriptions refer to different configurations and should not be merged.
Rank #3
- No Processor Installed; Supports 2x AMD EPYC 9004 Series Processors
- No Memory Installed; Supports 24x DDR5 4400/4800 Regsitered Memory Modules
- 8x 3.5" Trays; (Bring Your Own SATA/NVMe Drives)
- 4x H200 NVL Tensor Core 141GB HBM3e PCI Express 5.0 x16 GPU Accelerator Card
- In Original Packaging; Includes Rails and ASUS GPU Cables
What CoreWeave Forge is
CoreWeave Forge is an announced software environment for training, evaluating, and improving models and agents. NVIDIA describes it as bringing together Weights & Biases, OpenPipe post-training expertise, and the open-source marimo notebook project in a connected workflow for continuous model and agent improvement. The announcement describes its components and purpose, but does not establish public pricing or detailed availability terms.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What the announcement means for infrastructure buyers
The news is relevant to organizations evaluating cloud access to a new NVIDIA rack-scale system, but the public evidence supports a measured conclusion: capacity is being onboarded for customers, while availability remains limited. The reported throughput gains may be useful when assessing Cognition’s agentic-coding and reinforcement-learning workloads, but buyers should request workload-specific results and configuration details before projecting them onto their own systems.
Quick Recap
Rank #4
- 【Brilliant AI Performance for production】 on-device processing with up to 100 TOPS AI performance with low power and low latency, Due to the high thermal demands of Super mode, only the J30 Series supports upgrading to Super mode via the JetPack 6.2 update
- 【Hand-size edge AI device】 compact size at 130mm x120mm x 58.5mm, includes NVIDIA Jetson Orin NX 16GB production module, a cooling fan with a heatsink, enclosure, and a power adapter. Support desktop, wall mount, fit in anywhere
- 【Expandable with rich I/Os】4x USB 3.2, HDMI 2.1, 2xCSI, 1xRJ45 for GbE, M.2 Key E, M.2 Key M, CAN, and GPIO
- 【Accelerate solution to market】pre-installed Jetpack with NVIDIA JetPack 5.1 on the included 128GB NVMe SSD, Linux OS BSP, 128GB SSD, support Jetson software and leading AI frameworks and software platforms
- 【Comprehensive certificates】FCC, CE, RoHS, UKCA
- Availability: ask CoreWeave about current regional capacity and onboarding; the announcement does not promise general availability.
- Workload fit: distinguish SWE-2 inference from reinforcement learning, and compare results at relevant concurrency, interactivity, and latency targets.
- Deployment scale: confirm rack configuration, networking, storage, and how the proposed environment maps to an existing GB200 or GB300 fleet.
- Software operations: establish which Forge capabilities and integrations are available for the intended workflows.
- Commercial terms: no price comparison or public pricing is given in the cited announcements.
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.
