Volume 343, Issue 3 · 6 articles from 3 issues
Oxide Computer raises a $445M Series D led by Eclipse, with AMD joining as a strategic investor, to fund a large order backlog.
Oxide Computer says it became profitable enough to pay income tax this year and now has a backlog of orders that far exceeds supply.
To secure inventory, fulfill orders and keep accepting new demand, it raised a $445 million Series D led by Eclipse, with existing investors USIT, Riot Ventures and Jane Street taking large pieces and Atreides Management joining as a new investor.
AMD also joined as a strategic investor, reflecting Oxide's early bet on AMD EPYC and their joint platform enablement work.
Oxide's Product Assurance team introduces Voxel, a virtual lab that emulates a full Oxide rack on a single machine for testing.
Oxide explains that rack-scale hardware is hard to come by for testing because completed racks go quickly to customers.
Its Product Assurance team needs to replicate diverse customer network topologies, run the real control plane and current software, and represent hardware such as switches and microcontrollers without special test releases.
Beyond its small four-sled racklettes and simulated Omicron environments, Oxide built Virtual Oxide Labs, or Voxel, a successor to the earlier a4x2 virtual labs that packs a high-fidelity emulated rack into a single piece of hardware.
Oxide's Bryan Cantrill explains how a hardware root of trust anchors attestation from silicon up to running VMs.
In an Oxide FAQ Friday video, Bryan Cantrill explains that Oxide's hardware root of trust is a secure NXP LPC55 chip that can keep a secret and is uniquely programmed with a device ID at the factory.
It verifies that the software it loads is signed by Oxide, then attests to what runs on the service processor, which in turn attests to the host CPU's software.
Because Oxide owns the whole stack from root of trust through service processor, host OS and control plane, it can carry those attestations up to running VMs, letting a VM verify the stack beneath it.
Details are in Oxide's RFD 605.
In STAC-M3's baseline Antuco suite, L tick database on an Oxide VM set a storage-efficiency record and led cloud deployments.
STAC's baseline Antuco suite measures performance under fixed conditions, and in it the L tick database 1.0 on a single 128-vCPU Oxide OxVM virtual machine achieved the highest storage efficiency of any published Antuco result, at 364%, or 1.86 times the previous best.
It beat all disclosed results on the 1T YRHIBID-2 benchmark by 1.81 times and was faster than all other cloud-based deployments in 11 of 17 mean-response-time tests.
Against a similar q-language stack on NVMe SSDs, it won 9 of 17 tests, including an 8.4x speedup on THEOPL.
Full reports and configurations are available to STAC subscribers.
An audited STAC-M3 Shasta run of L Labs' tick database on one Oxide VM beat all public results in 16 of 17 response-time tests.
STAC audited a stack running L Labs PBC's L tick database 1.0 on a single Oxide OxVM cloud virtual machine with 128 vCPUs, 1.25TiB of RAM and five 3.44TB Oxide local NVMe disks, running Ubuntu 24.04.
In the Shasta suite, which allows data to be preloaded into memory, the stack outperformed all publicly disclosed results in 16 of 17 mean-response-time benchmarks.
Highlights include a 13.7x speedup on the 10T VOLCURV benchmark and 10.7x on 100T VWAB-12D-NO versus the next-best kdb+ results, plus the top result on the optional NBBO benchmark.
STAC-M3 is the industry standard for tick-analytics stacks.
Oxide and L Labs set 18 audited STAC-M3 records on a single VM on the Oxide Cloud Computer, with speedups up to 13.7x.
Oxide Computer and L Labs set 18 STAC-M3 world records on a single virtual machine on the Oxide Cloud Computer.
STAC-M3 is the industry-standard benchmark for tick database platforms, and the results were audited by STAC.
The setup set records in 16 of 17 Shasta response-time metrics, with speedups up to 13.7x over the previous best.
In the storage-bound Antuco suite, L 1.0 achieved the highest storage efficiency of any published result, at 364%.