Key Findings
NVIDIA has updated the hardware shipping specifications for the Rubin GPU as production ramps up in the fourth quarter of 2026. The revised specs show a reduction in both HBM4 memory bandwidth and scale-up NVLink bandwidth, offset by an increase in scale-out networking bandwidth. Compute throughput remains unchanged at 50 PFLOPS for FP4 inference and 35 PFLOPS for FP4 training.
Technical Details
The following adjustments have been made to the Rubin GPU specifications compared to preliminary targets:
- HBM4 Memory Bandwidth: Reduced from 22TB/second to 19.2TB/second. This represents a data rate drop from 10.75Gbps/pin to 9.375Gbps/pin (an 11% loss).
- NVLink Bandwidth: Reduced from 3.6TB/second to 3.0TB/second, a 17% reduction.
- Scale-out Networking Bandwidth: Increased from 400GB/second to 450GB/second.
These changes propagate to the Vera Rubin NVL72 rack. Cumulative GPU memory bandwidth for the rack has dropped from 1,580TB/second to 1400TB/second, and system-wide NVLink bandwidth has decreased from 260TB/second to 216TB/second. Conversely, bidirectional networking bandwidth for the NVL72 has increased from 28.8TB/second to 32.4TB/second. GPU memory capacity remains at 20.7TB of HBM4.
Background and Context
NVIDIA previously noted that preliminary specifications were subject to change. For context, the revised 9.375Gbps/pin rate remains higher than the official HBM4 specification target of 8Gbps/pin and significantly higher than AMD’s MI455X, which runs HBM4 stacks at 7.6Gbps/pin. The source suggests the downward revision may be due to challenges in memory controller speeds, stack speeds, or yields.
Significance and Outlook
While the compute hardware and CPUs are shipping at spec, the reduction in memory and NVLink bandwidth may impact the AI performance of the NVL72 rack compared to original projections. However, the boost in scale-out networking bandwidth provides a performance advantage for larger deployments that rely heavily on networking across multiple systems.
Source: https://www.servethehome.com/nvidia-updates-rubin-gpu-specs-as-first-production-hardware-arrives/
Get our weekly technology intelligence — free
Receive an infographic that lets you judge at a glance whether each field’s analysis report is worth reading.
Subscribe Free — Weekly Tech Intelligence
By subscribing, you’ll receive Troy-Technical’s weekly technology intelligence newsletter.
- Your email and selected fields are used only to deliver the newsletter.
- We never share your information with third parties.
- You can unsubscribe anytime via the link in each email.
See our Privacy Policy for details.
Takes about a minute · Unsubscribe anytime
