How to verify genuine server CPUs before buying in bulk

Quick answer: Verifying genuine server CPUs before buying in bulk means decoding the Intel sSpec (e.g. SRM7K) at ark.intel.com or the AMD OPN against the ordered SKU, inspecting the IHS for crisp laser etching, and testing a sample on an OEM board with lscpu, a 30-minute all-core load test and full-population training. Q-spec ES/QS units are never retail parts.

A container of "Xeon Gold" processors arrives at 30% below market price. The trays look right, the packaging is convincing — and three weeks later your build line discovers half the units are remarked engineering samples. This scenario costs the secondary server market millions every year, and it is almost entirely preventable with a 15-minute verification routine.

Here is the checklist our lab runs on every inbound CPU batch, and the same checks you should demand from any wholesale supplier.

1. Decode the sSpec or OPN first

Every genuine Intel tray CPU carries an sSpec number (five characters, e.g. SRM7K) laser-etched on the integrated heat spreader. AMD tray units carry an OPN (e.g. 100-000000452). Before anything else:

• Intel: enter the sSpec at ark.intel.com — it must resolve to the exact model, stepping and SL-code you ordered.
• AMD: the OPN's first segments identify model and stepping; the full code must match AMD's product brief for that SKU.
• Mismatch = stop. A valid-looking model number with a wrong sSpec is the single most common remarking trick.

Two details make this lookup faster in practice. On Intel trays the sSpec appears twice — etched on the IHS and printed on the tray label — and the two must agree before you even open ARK. Once there, compare three fields on the spec page: processor number, stepping and spec code. On AMD units, a trailing WOF suffix marks retail-boxed packaging; bulk tray units ship without it. If the seller's datasheet shows a different OPN than your quote, ask for a photo of the actual tray label before paying.

2. Inspect the IHS markings

The metal heat spreader is hard to fake well. Under bright light, check:

3. Power it on a real board

Paper checks catch remarking; electrical tests catch everything else. For any batch above a few units, validate a sample on a genuine OEM platform (Dell, HPE, Lenovo, Supermicro):

1. POST and inventory: the BIOS must report the exact model string and core count — no "Unknown CPU" or downgraded core counts.
2. CPUID / stepping: on Linux, lscpu or dmidecode -t processor exposes family, model and stepping; compare against the sSpec record.
3. Load test: 30+ minutes of sustained all-core load (stress-ng, Prime95) — watch for thermal throttling, WHEA errors or core shutdowns that indicate binning fraud.
4. Memory training: full-population DIMM training should complete at rated speed on the first or second attempt.

On Windows hosts, CPU-Z and HWiNFO expose the same identity data. In CPU-Z, the "Specification" field must read the full retail model string — a generic "Genuine Intel(R) CPU @ 2.00GHz" string, an "(ES)" tag or a blank stepping line all point to an engineering sample, whatever the IHS claims. HWiNFO's CPU window lists the exact stepping and flags ES/QS units directly; compare the value against the ARK record from step 1. One digit of difference invalidates the batch, not just the unit.

4. Know your ES, QS and retail markings

Engineering samples are not fraudulent — but they are not retail parts, and they must never be priced as such:

• Intel ES: sSpec starting with Q (Q-spec), or confidential markings — no warranty, no long-term supply guarantee.
• Intel QS: closer to production, still Q-spec — acceptable for lab validation only.
• AMD ES: OPN marked with ES codes or "ENG SAMPLE" laser marks.
• Retail / tray: SR-prefix (Intel) or production OPNs — the only units that belong in production servers.

"Qualification samples are for your lab, not your customers. Any supplier mixing ES units into a retail tray order — at any ratio — has told you everything about their quality system."

— Huazhe incoming-inspection procedure

5. Structure the deal to cap your risk

• Sample first: order 1–2 units for validation before any volume commitment.
• Test report in the PO: make a signed per-batch test report a contractual deliverable, not a courtesy.
• Matched stepping clause: for cluster builds, specify same-stepping / same-CPUID allocation in writing.
• Warranty terms: up to 2 years with advance replacement — remarkers rarely survive a warranty conversation.

Two clauses do most of the work in practice. First, serial-level traceability: the invoice should list the per-unit sSpec or OPN (and serials where available), so a quarantined unit maps back to a line item instead of an anonymous tray. Second, a written rejection threshold — any ES/QS unit or any failed load test voids the batch, not just the affected unit. Suppliers who push back on either clause usually know something about their sourcing that you do not.

Where to buy tested CPUs in bulk

The cheapest quote that skips the inspection steps above is not cheaper. Our server CPU wholesale stock list covers Xeon 6 and EPYC 9005 down to legacy LGA3647 / SP3 platforms — every batch validated on OEM boards and shipped with a signed per-batch test report. Send your BOM and we quote within 24 hours. The same verify-before-you-pay discipline applies to accelerator batches: GPU-on hours, aggregate ECC history and sustained thermals play the role the sSpec plays for CPUs.

Notes from the Zhongshan desk

Most remarked CPUs we see are not crude fakes — they are genuine silicon re-labelled to a higher SKU. A typical telling detail is the etching font: factory laser marks are thin, uniform in depth and sit dead-centre in the IHS field, while re-etched lines come out slightly bolder, with edges that catch raking light differently. Alignment is the second giveaway. The re-marked line of text often drifts a fraction of a millimetre off the axis of the original markings, because the stencil was aligned to the spreader, not to the layout underneath.

We have also seen ES and QS units laser-etched with production-looking sSpec codes. The etching is clean, but the CPUID gives the game away — the chip reports an identity that matches no production record for that spec code, or reports no retail spec code at all.

Our first-line visual check, before any unit reaches a board, takes under two minutes per tray: raking light across the IHS, a 10× loupe on the sSpec characters, then a corner wipe with isopropyl alcohol. Anything that smears or darkens is quarantined for board-level verification or returned.

FAQ

Is a pulled (used) server CPU safe to buy? Yes — if it comes from a supplier who tests on OEM boards, discloses per-unit health data and warrants the unit. Pulled enterprise CPUs often have years of service life left; the risk is in the seller's process, not the silicon.

Can software alone detect a remarked CPU? CPU-Z, lscpu and HWiNFO read the CPUID the processor reports, which a sophisticated re-mark may fake. Physical IHS inspection plus board-level validation is the only complete check.

What should a batch test report include? At minimum: serial / sSpec list per unit, platform tested on, BIOS version, load-test duration and result, memory training result, and thermal readings. Ours ship with every batch.