How oxide cuts data center power consumption in half
Oxide’s claim to cut data center power use roughly in half by redesigning racks around a centralized DC bus bar and larger, slower fans prompts debate over how novel and impactful this really is compared to existing telco and Open Compute designs. Commenters weigh the trade‑offs between efficiency gains, new single‑point‑of‑failure modes, and the realities of integrating such tightly engineered systems versus buying commodity x86 servers. The thread also examines Oxide’s business choices—Illumos-based stack, lack of GPUs, full‑rack “cloud computer” positioning—and whether its premium, opinionated hardware can gain traction beyond large enterprises given cloud alternatives and homelab interest.
Apple/ARM vs x86 in the data center
- Some wonder why Apple doesn’t sell rack-mount M‑series servers, citing strong perf/W for data centers.
- Others argue Apple already uses Apple Silicon internally for specific privacy/security workloads but that this is a niche use case.
- Skeptics note AMD’s latest Epyc chips are extremely efficient and hard to beat, and that Apple is unlikely to open its chips or platform to the general server market.
- Consumer macOS is described as an unreliable server OS historically, though some issues have been fixed.
Oxide’s hardware design and power savings claims
- Key savings are attributed to: shared high‑efficiency rectifiers feeding a DC bus bar, and larger, slower fans with less airflow restriction.
- Some doubt “12x” fan energy reduction, but Oxide staff report that default fan speeds already overcool, so they run at low RPM.
- Oxide confirms future generations will fit existing racks, emphasizing rack‑scale design and reuse.
Power distribution, redundancy, and failure modes
- Debate over whether a shared DC bus and power shelf are a single point of failure vs. 70 individual PSUs.
- Pro‑bus‑bar side: fewer, higher‑quality rectifiers with N+1 (or more) redundancy are preferable; bus bars are “dumb copper” and very reliable.
- Critics note DC protection and rack‑wide faults, but others counter that at 48V and with proper fusing/rectifiers this is manageable.
- Comparison to traditional dual‑PSU, dual‑PDU setups highlights that those also have systemic failure risks and capacity pitfalls.
Relation to OCP and telco DC practices
- Commenters point out DC bus bars and centralized rectifiers have long existed in telco and Open Compute designs.
- Distinction: OCP gear is hard to buy and integrate for ordinary enterprises; Oxide’s pitch is a turnkey, vendor‑supported rack.
Software stack, Illumos, and security
- Oxide uses Illumos under the hypervisor and services; customers run standard VMs/containers on top.
- Some worry about relying on a niche OS and speculative‑execution mitigations; others respond that Oxide explicitly owns and ships all patches, with a single-vendor responsibility model.
Market fit, pricing, GPUs, and homelab interest
- Current product targets large organizations; prices and minimum scale don’t fit fast‑growing startups or homelabs.
- Many would like a smaller or cheaper system or just Oxide’s control plane/BMC on commodity servers; Oxide says that wouldn’t meet their design goals and they lack bandwidth for loss‑leader lines.
- Lack of GPUs is noted as a gap given AI demand; Oxide acknowledges this and plans to address it later.
Energy and climate framing
- Some agree data center efficiency matters; others see the “data centers use X% of world power” framing as weak, arguing that overall workload value and avoiding wasteful workloads (e.g., some LLM uses) matter more.