What Is Still Current, What Is Withdrawn, and What Replaced It

Written and maintained by Haink's infrastructure team · Platform status verified against Lenovo Press product guides, 29 August 2026 · authorized-channel, MTM and serial verified

The question that brings people to a page like this is specific: there is an SR630 in rack four, nobody remembers when it was bought, and someone needs to know whether it is a problem. Two separate facts answer that, and they come from two different places.

The first is platform status — whether the model is still sold, superseded, or withdrawn, and what replaced it. That is published, stable, and mapped below. The second is your machine's end-of-service date, which is specific to its machine type and model, differs within a single product family, and changes if extended service was purchased. That one we deliberately do not republish.

Why we do not publish an end-of-service date table

Because it would be wrong, and being wrong about this is expensive.

End-of-service dates in the ThinkSystem line are set per machine type and model, not per product name. Two servers that both say "SR650" on the bezel can carry different machine types with different dates. Extended service coverage, where it was bought, moves the date again. And a table published once and left to age is exactly how a refresh gets planned around the wrong quarter — which is a real cost, because the difference between "this rack is covered until next year" and "this rack has been uncovered for six months" is an incident away from being someone's board paper.

Reseller sites republish these tables constantly and rarely date them. Treat any end-of-service table you find, including on supplier sites, as a starting hypothesis rather than a fact. The authoritative answer takes about two minutes to obtain.

How to get your actual dates

1. Collect machine type and model, or serials, for the fleet — not model names. On ThinkSystem hardware the machine-type-model is on the pull-out tab and in the management controller inventory; XClarity can export it for the whole estate at once.
2. Run them through Lenovo's end-of-service date lookup for servers, storage and networking.
3. Read which milestone you are looking at. End of sale and end of service are different events, usually years apart: a platform can be unavailable to buy while remaining fully supported. Confirm the definition on the vendor page rather than assuming.
4. Check whether extended coverage was purchased on any of the machines. It is common, it is not visible from the model name, and it changes the plan.
5. Sort the estate by date, not by rack. This is the step that changes projects — see the planning section below.

Platform status map

This is the part that is stable enough to publish. Status verified against Lenovo Press product guides on the date in the byline.

Rack servers, Intel

Rack servers, AMD

A point that catches people reading a version number as a date: the AMD ThinkSystem platforms — SR635 V3, SR645 V3, SR655 V3, SR665 V3 and others — are current, carrying 5th Gen EPYC "Turin" processors — up to 160 cores per socket on the SR665 V3. A V3 label on an AMD quote is not a previous-generation quote. We unpack that in ThinkSystem V3 vs V4.

ThinkAgile hyperconverged

Two things worth knowing before planning a ThinkAgile refresh. From the V4 series onward Lenovo dropped the Integrated System / Appliance / Certified Node nomenclature on the VX line, so a purchasing template that specifies a "VX Certified Node" will not match a current quote. And the FX V4 models are validated for VMware, Microsoft and Nutanix stacks on one hardware model, which changes the calculation if the hypervisor decision is not settled. Both are covered in which ThinkAgile series for which hypervisor.

Storage

The DE, DM and DG series remain the current ThinkSystem storage lines — DM and DG running NetApp ONTAP, DE being NetApp E-Series underneath. Lenovo has also introduced a DS Series of all-flash SAN arrays based on OEM'd NetApp ASA systems. If you are planning a storage refresh, the platform question is worth settling before the label question; we cover it in enterprise storage selection.

Planning a refresh around support cliffs, not the calendar

Once the estate is sorted by end-of-service date rather than by rack or by purchase order, three things usually become visible, and all three change the project:

1. The fleet does not expire together. Hardware bought in a single wave frequently carries dates spread across two or three years, because machine types differed and some machines got extended coverage. Refreshing it as one wave means retiring supported hardware early.
2. A tranche is already uncovered. Almost every estate we look at has a handful of machines past their date that nobody flagged, usually in a secondary site. Those set the urgency, not the average age of the fleet.
3. The right wave size is smaller than expected. Sequencing by cliff usually produces two or three smaller projects rather than one large one — which is easier to fund, easier to schedule against maintenance windows, and avoids the trap of buying capacity a year before it is needed.

The economics of the replacement itself — what a consolidation actually saves, and why the licence line does not fall with the node count — are worked through in server refresh and consolidation. How a project of that size is bought is in buying twenty to sixty servers as one project.

If you cannot replace a withdrawn platform yet

Sometimes the answer is that the budget is next year and the hardware has to keep running. That is a manageable position if it is a decision rather than an oversight:

• Hold spares for the specific failure modes. On a withdrawn platform, power supplies, fans and drives are the parts that end the machine. Holding them locally is cheaper than an emergency sourcing exercise on a platform nobody stocks any more.
• Verify anything you buy on the secondary market. Withdrawn platforms are where gray-market supply concentrates. Confirm machine type and serial before payment — see gray-market channel risks and verifying an in-stock offer.
• Move the workload, not the hardware, where you can. A withdrawn node running something non-critical is a different risk from the same node carrying a production database. Rebalancing workloads costs nothing and buys time.
• Write down the decision. Running out-of-support hardware knowingly, with compensating controls, is a defensible position. Running it unknowingly is not, and the difference matters at audit.

Frequently asked questions

Is the ThinkSystem SR650 end of life?

The original SR650 on Xeon SP Gen 1 and Gen 2 is a withdrawn product. The SR650 V2 is superseded, the SR650 V3 is the widely stocked previous generation and fully supported, and the SR650 V4 is current. The name alone does not tell you which one you have — check the machine type.

Where do I find my server's end-of-service date?

From Lenovo's end-of-service date lookup, using the machine type and model or the serial — not the model name. Dates differ between machine types within the same product family and change if extended service was purchased, which is why no published table is reliable for your specific machine.

Why won't you publish the dates on this page?

Because they are per machine type and model, they vary inside a single product family, extended coverage changes them, and a table that ages silently is how a refresh gets planned around the wrong quarter. The status map above is stable enough to publish; the dates are not, and getting yours takes about two minutes.

Is a V3 server end of life?

No. On the Intel side, V3 is the previous generation, widely stocked and supported. On the AMD side, V3 is the current generation — those platforms carry 5th Gen EPYC "Turin" processors. Do not read the version number as an age.

What replaced ThinkAgile SX?

Nothing, as a like-for-like refresh. ThinkAgile SX and SXM for Microsoft Azure Stack Hub are withdrawn, and the successor conversation is Azure Local, which is a platform migration rather than a hardware replacement. Budget it accordingly.

Can we keep running servers past end of support?

Technically yes, and plenty of estates do. Make it a decision rather than an oversight: hold spares for the parts that actually fail, verify anything bought on the secondary market by machine type and serial before payment, move critical workloads off the affected nodes, and document the position with its compensating controls.

Should we refresh the whole fleet at once?

Usually not. Hardware bought in a single wave often carries end-of-service dates spread across two or three years. Sorting by date rather than by rack typically produces two or three smaller, better-funded waves and avoids retiring machines that still have supported life in them.

Related

• MTM and part numbers — how to read the machine type you look the dates up with
• HPE Alletra explained — what happened to 3PAR, Primera and Nimble, and the lifecycle dates
• Model numbers decoded — making sense of the family names in the status map above
• ThinkSystem V3 vs V4 — why the version number is not an age, and which generation to specify
• Server refresh and consolidation — what the replacement actually saves, and what it does not
• Which ThinkAgile series for which hypervisor — if the refresh is hyperconverged
• Buying 20–60 servers as one project — how a refresh wave is actually purchased
• ITAD and the secondary market — what the retired fleet is worth, and when
• How to choose Lenovo ThinkSystem servers · stock, lead times and pricing

Sources

• Lenovo — end-of-service date lookup for servers, storage and networking (the authoritative source for per-machine dates)
• Lenovo Press — ThinkSystem SR630 (Xeon SP Gen 1/Gen 2) product guide · SR650 datasheet · SR530 · SR550 · SR570 (all marked withdrawn product)
• Lenovo Press — ThinkAgile HX 1U appliances and certified nodes, Xeon SP Gen 3 (withdrawn product)
• Lenovo Press — ThinkAgile SX for Microsoft Azure Stack Hub (withdrawn product)
• Lenovo Press — ThinkSystem servers with 5th Gen AMD EPYC (which V3 platforms are current)
• Lenovo Press — Introducing the ThinkSystem V4 servers