Reported by StorageReview SSD/NVMe/存储评测与行业新闻 on 2026-10-13. Summarized and analyzed by XAIPART Editorial.
What happened
On Yotta 2026's show floor, StorageReview's Brian sat down with Chris Sullivan, Director of Research and Academic Computing in Oregon State University's College of Earth, Ocean, and Atmospheric Sciences, to talk about field-deployed AI storage.S1
For the NSF-funded regional class research vessel program, Sullivan and team sized an onboard data center in the 30kW to 50kW range and chose Solidigm D5-P5336 drives after a validation pass on a Dell PowerEdge XE7745 with four NVIDIA RTX Pro 6000 GPUs.S1
The podcast disclosed several concrete figures: the ship costs $150 million, a ten-day cruise runs about $1 million, the plankton experiment generates about 100TB of captured data in five days, the U.2 QLC drives delivered roughly 55GB/s reads and 15GB/s writes, and the forest acoustic program ingests about 250TB every six to eight weeks from a ranger station that holds roughly 300TB of local capacity on two 122TB Solidigm drives.S1
Sullivan also described 4,000 autonomous recording units in the field, a convolutional neural net that identifies 135 species, and a grant cycle that assumes 18-month hardware refreshes while the industry has compressed to three to six, plus the Ocean Observatories Initiative data that Oregon State holds for 25 years minimum, expected to remain for 50 to 100.S1
| Metric | Reported value | Source |
|---|---|---|
| Research vessel cost | $150 million | [S1] |
| Cost per ten-day cruise | ~$1 million | [S1] |
| Plankton data per 5-day cruise | ~100TB | [S1] |
| U.2 QLC read throughput (validation) | ~55GB/s | [S1] |
| U.2 QLC write throughput (validation) | ~15GB/s | [S1] |
| Ranger station ingest cadence | ~250TB every 6 to 8 weeks | [S1] |
| Ranger station local capacity | ~300TB (2x 122TB) | [S1] |
| Forest acoustic recorders deployed | ~4,000 | [S1] |
| CNN species identification scope | 135 species | [S1] |
| Grant-cycle assumed hardware refresh | 18 months vs. 3 to 6 industry cycle | [S1] |
| OOI data retention target | 25 years minimum, 50 to 100 expected | [S1] |
Background timeline
- - StorageReview published its prior write-up of Oregon State's "AI at Sea" research vessel data center.
- - StorageReview recorded Podcast #154 with Chris Sullivan on the Yotta 2026 show floor in Las Vegas.
- Earlier 2026 dates not disclosed - Sullivan co-published the convolutional neural net that identifies 135 species from forest audio.
- - Oregon State expects delivery of the first NSF-funded regional class research vessel that Sullivan has been specifying for the past year.
XAIPART take
Technical impact
The validation result is the headline technical takeaway for storage designers: a single U.2 QLC tier pulled ~55GB/s reads and ~15GB/s writes, enough to feed six to eight concurrent plankton processing streams, with Sullivan noting disk, not GPU, was the limiter on the test bed.S1 That is a useful anchor for anyone sizing QLC tiers for inference ingest at the edge, where power and cooling - not rack units - define the budget.XAIPART analysis
The 122TB D5-P5336 fits because Solidigm also kept a 30TB SKU in the same family for the deployment that needed throughput per drive rather than raw capacity; Sullivan framed the trade-off as energy savings on storage once GPU power was already committed.S1 The eight-drive, one-petabyte config retired a full rack of spinning disk and trimmed the surrounding hardware, not just the drives themselves.XAIPART analysis
Market impact
For procurement, this is a real-world reference design for buyers weighing QLC vs. nearline HDD at the edge: a liquid-cooled Supermicro GPU node on a standard 15A circuit, two drive slots, ~300TB raw, fed by ~250TB of field ingest every six to eight weeks, with no air conditioning available.S1 That package should be tracked by anyone sourcing Solidigm D5-P5336 122.88TB and 30TB SKUs or comparable high-density QLC.XAIPART analysis
The grant-cycle mismatch Sullivan flagged - 18-month assumed refresh versus three-to-six-month industry cycles - is a sourcing risk that applies to every academic and federal buyer writing multi-year hardware quotes today, and it directly affects how aggressively QLC TLC roadmaps should be designed in.XAIPART analysis
Action
Design engineers evaluating edge AI storage should treat the 55GB/s read / 15GB/s write U.2 QLC result as a benchmark for shipboard and remote-station inference ingest, then size capacity against the workload's per-cycle ingest rather than per-day.XAIPART analysis
Procurement teams sourcing the Solidigm D5-P5336 family should confirm both 122.88TB and 30TB SKU availability with their franchised distributor and ask about lead times, since the source describes a multi-month shipbuilding and grant cycle that cannot tolerate 12-week-plus storage slip.XAIPART analysis
XAIPART Forecast
XAIPART expects that if Solidigm D5-P5336 122.88TB lead times at franchised distributors do not stabilize below 8 weeks by Q2 2027, the Oregon State regional class research vessel program will be forced to qualify a second 100TB-class QLC source, since the December 2026 ship delivery and the 18-month-vs-3-to-6-month grant cycle cannot absorb long storage slip.XAIPART analysis
Facts are as reported by the cited source. The design-in, sourcing and forward-looking statements above are XAIPART's analysis, not the source's claims. Uncertainty: storage pricing, distributor stocking, and QLC roadmap cadence are not disclosed by the source; the 55GB/s and 15GB/s figures are from a single validation pass, not a published datasheet.
Corrections: none at publication. This brief will be updated with a timestamp if the source corrects its report.
Alternatives and lifecycle
| Part | Relation | Lifecycle | Basis |
|---|---|---|---|
| — | No alternative on file NEEDS DATA | — | Requires match_type before any predecessor label may appear |
“Predecessor” and “alternative” labels come only from a recorded match type, a lifecycle status, or the source text. Where no match is recorded, no label is shown. Our own listings are not a market reference.
Affected parts
No catalog match. This story does not map to any part currently listed on XAIPART. We do not infer matches, and we do not show a placeholder list.
Cost pressure by dimension
| Dimension | FRED series ID | Unit | Lag | As of |
|---|---|---|---|---|
| Input cost | PCOPPUSDM, PALUMUSDM, PNICKUSDM | USD/mt | 96 d | |
| Supply/demand | PCU334413334413, IPG3344S | index(1982=100) / index(2017=100) | 46 d | |
| FX & energy | DEXCHUS, DCOILBRENTEU, DHHNGSP | rate / USD/bbl / USD/MMBtu | 6 d |
No free authoritative interface for: gold · silver · rare earth · polysilicon · lithium carbonate · market share — listed rather than approximated.
How this was produced
Process evidence — fetch, source sanity, model and gate
| Source fetched | StorageReview SSD/NVMe/存储评测与行业新闻 · www.storagereview.com · 16,168 chars · fetch status success |
|---|---|
| Source sanity (S1) | ok — passed before any model call (no s1_ block recorded) |
| Model / prompt | MiniMax-M3 · prompt Blog Agent - News Summary (v2.8-final) · 10,608 tokens · 15.7s · run completed |
| Gate | verdict pass · 6/6 applicable checks passed · spec v2.9 · C10 ratio 0.015633 · C10 rounds 0 |
| Time contract | 3 <time datetime> values in this brief — 0 violations (+08:00 or date-only) |
Every row above is produced by the pipeline, not written by hand.
Comments
Be the first to comment.
🔐 Please sign in to post a comment.
Don't have an account? Sign up