IBM’s Vela refresh focused on moving data between GPUs faster and fitting more GPU capacity into the system. IBM Research reported two to four times higher network throughput, six to 10 times lower network latency, roughly twice the previous GPU capacity, and a 50% reduction in the time its automation needed to find and understand hardware failures or degradation. Those are IBM-reported results, not an independent benchmark; the published material does not state the test conditions or an exact post-refresh GPU count.
What changed in the Vela refresh?
The main change was to the network path between GPUs. IBM added RoCE—RDMA over Converged Ethernet—and GPU-direct RDMA. RDMA allows systems to transfer data with less CPU and operating-system network-stack involvement; GPU-direct RDMA lets data move between GPUs and the network more directly rather than relying on the CPU to handle each transfer.
IBM Research’s December 2023 account reported these changes and results:
| Area | What IBM reported |
|---|---|
| Network throughput | Two to four times higher after the upgrade; IBM did not state the benchmark conditions in the cited account. |
| Network latency | Six to 10 times lower after the upgrade; IBM did not state the benchmark conditions in the cited account. |
| GPU capacity | Approximately twice the previous GPU capacity; no exact post-refresh GPU count was stated. |
| Hardware failure response | Automated failure detection cut the time to find and understand hardware failures and degradation in half. |
The upgrade also doubled server-rack density. IBM connected that denser design to fitting more capacity within existing power and cooling limits, but the cited material does not provide power or cooling measurements.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
#1 Best Overall
- Dell Precision 7920 Tower Workstation
- 2x Intel Xeon Gold 6130 16-Core 2.1GHz (3.7GHz Turbo)
- 192GB DDR4 Memory - upgradable to 1.5TB
- 2x 1TB SSD + 2x 4TB HDD (Removable Hot Swap Drive bays)
- Nvidia Quadro P1000 4GB - Windows 11 Professional 64-bit
Why does faster GPU networking matter?
Training a large model across many GPUs depends on frequent communication among them. If data transfers lag, GPUs can spend time waiting rather than computing, reducing the benefit of adding more accelerators. RoCE and GPU-direct RDMA are intended to reduce that communication bottleneck by making transfers more direct and reducing CPU and network-stack overhead.
IBM said the refreshed system scaled near-linearly to larger workloads and was used to train a 20-billion-parameter Granite model. IBM described that model as a key enabler for watsonx Code Assistant for Z. The reported network improvements support IBM’s case for the refresh, but the source material does not include independent comparisons or enough test detail to treat the figures as universal performance guarantees.
What hardware did Vela use?
IBM’s published description of Vela’s original node design specified eight 80GB NVIDIA A100 GPUs connected with NVLink and NVSwitch, two Intel Xeon Scalable processors, 1.5TB of DRAM, and four 3.2TB NVMe drives. Compute nodes connected through multiple 100G Ethernet interfaces arranged in a two-level Clos topology.
Rank #2
- [Local AI Inference & 70B Model Ready] Equipped with the AMD Ryzen 7 PRO 8845HS processor, NEXUS is engineered for heavy local AI workloads. With a full-size GPU bay, it runs 70B LLMs natively without an internet connection. Ideal for AI developers and tech enthusiasts who need private environment for coding and model testing.
- [132TB Mass Storage with ZFS Integrity] Features a hybrid storage architecture (3×NVMe + 4×3.5" HDD) supporting up to 132TB. Utilizing the enterprise-grade ZFS file system and ECC memory, it prevents data corruption and bit rot—a must-have for professional photographers and video editors safeguarding 4K/8K RAW footage.
- [OpenClaw-Driven Automation Workflow] The built-in OpenClaw execution layer allows complex automated tasks to be processed locally. Even when offline, your backup schedules and AI file organization continue seamlessly. Say goodbye to monthly cloud subscriptions and high latency.
- [Dual 10GbE & USB4 Ultra-Connectivity] Experience server-class speeds with dual 10GbE ports and a 40Gbps USB4 interface. It enables multi-user real-time collaboration on large project files directly from the NAS, ensuring zero-lag editing for creative studios and production teams.
- [Open-Source ZimaOS for Total Privacy] Running on the fully open-source ZimaOS, NEXUS ensures your data stays physically on-premise with no backdoors. It acts as a "Digital Fortress" for privacy-conscious families and small businesses who demand absolute data sovereignty.
That is the original published design, not a complete bill of materials for the refreshed system. IBM’s refresh account says GPU capacity roughly doubled but does not state the updated node configuration or total number of GPUs. IBM also reported less than 5% virtualization overhead per node in its original design, while making GPU, CPU, network, and storage capabilities available inside virtual machines.
What is Vela, and who can use it?
Vela is IBM Research’s AI-optimized, cloud-native supercomputer, hosted in IBM Cloud. IBM said it had been in operation since May 2022 and supported data preprocessing, model training, fine-tuning, deployment, and product incubation. IBM Research used it for foundation-model work and to bring watsonx.ai online.
Vela is enterprise research infrastructure, not a retail computer or a generally available product with a public price. The cited material does not establish that customers can reserve or rent capacity on Vela itself, nor does it provide current pricing. Its role in IBM Cloud should not be taken to mean that the specific research system is an on-demand IBM Cloud instance.
Rank #3
- 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.
Can an organization build a Vela-like system on premises?
Yes. IBM’s 2024 technical note described a Vela-derived on-premises cloud-native AI supercomputer design, distinct from the Vela system hosted in IBM Cloud. It is designed to scale from dozens to hundreds or thousands of NVIDIA H100 GPUs and uses RDMA-enabled Ethernet, IBM Storage Scale, OpenShift Container Platform, and OpenShift AI, alongside pre-built containers, models, and APIs for elastic access.
The first phase of this design went live at Phoenix Technologies in Switzerland in mid-August 2024 through a collaboration involving IBM, Red Hat, Phoenix, and Dell. This demonstrates an on-premises deployment path based on Vela’s approach; it does not mean Vela itself moved out of IBM Cloud or that every deployment uses the same scale and configuration.
Quick Recap
What the published results do—and do not—show
- They show IBM’s reported system changes: GPU-direct RDMA over Ethernet, higher reported throughput, lower reported latency, greater GPU capacity, and faster automated failure diagnosis.
- They do not establish an independent benchmark: the published figures are IBM-reported, and the cited accounts do not provide enough test methodology to reproduce or generalize them.
- They do not establish a public product offer: the sources describe IBM Research infrastructure and a Vela-derived on-premises design, not a public Vela price or general access terms.
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.
