NVIDIA's Vera Arm CPU claims up to 3.67x faster storage data-path processing than x86 chips from Intel and AMD, per @rohanpaul_ai. The chip powers the BlueField-4 STX Storage Processor.
Key facts
- 3.67x: claimed speed lead over x86 for storage data-path ops
- 3.29x: claimed compression throughput advantage for Vera
- BlueField-4 STX: NVIDIA's storage processor hosting Vera
- Targets Intel and AMD's decades-old storage CPU market
- Agentic AI: thousands of agents multiply storage operations
NVIDIA is taking direct aim at Intel and AMD's decades-long grip on storage processing. According to @rohanpaul_ai, the company's Vera Arm CPU handles the routine but critical work storage systems perform — encrypting, checksumming, compressing, and rebuilding data — up to 3.67x faster than the x86 chips that have traditionally owned that job.
The claim centers on the BlueField-4 STX Storage Processor, where Vera runs the storage data path. [The source says] Vera hits 3.29x higher compression throughput than x86 CPUs. This is the unglamorous layer that locks, squeezes, and verifies data before it lands on a drive — work that happens before a GPU ever touches the data.
Why agentic AI changes the math
The competitive stakes have shifted because of agentic AI. [Per @rohanpaul_ai], a single agent step can fire several storage operations, and with thousands of agents running concurrently, the load on storage controllers has exploded. What was once a quiet background job — a controller compressing and encrypting on spare cores — is now a bottleneck.
NVIDIA's pitch is that offloading this work to a purpose-built Arm CPU in the BlueField-4 STX delivers a measurable speed advantage. The 3.67x figure is a vendor claim, not an independent benchmark; [the source does not disclose] the specific x86 CPU compared, the workload mix, or the test methodology. That matters — compression and encryption throughput vary widely with data type and algorithm.
Still, the strategic signal is clear. NVIDIA is extending its data-center dominance beyond GPUs into the storage data path, a market Intel and AMD have served for decades. If the 3.67x claim holds under scrutiny, it pressures both incumbents on a core competency.
What to watch
The source does not provide a release timeline for BlueField-4 STX or pricing. Watch for third-party benchmarks from storage vendors (like Pure Storage or NetApp) that validate or challenge the 3.67x figure, and for Intel or AMD counter-claims on their next-gen data-center CPUs.
[Updated 06 Aug via tomshardware]
Musk's endorsement extends beyond terrestrial data centers: he confirmed SpaceX will exclusively use Nvidia GPUs 'because they are the best,' and that an optimized Vera Rubin NVL72 rack will be launched into space next year [per Tom's Hardware]. This marks the first public confirmation of Nvidia's AI hardware operating in orbit, potentially for on-orbit inference or training. The announcement adds a new dimension to Nvidia's dominance, suggesting its architecture will handle AI workloads beyond Earth, a market Intel and AMD have not targeted. It also implies Nvidia's roadmap remains on track for the Vera Rubin platform.
