How to Choose Lenovo ThinkSystem Servers

Written and maintained by Haink's enterprise infrastructure team · Updated July 2026 · authorized-channel, MTM-verified

Haink supplies Lenovo ThinkSystem servers, ThinkAgile hyperconverged infrastructure, ThinkSystem storage, and AI infrastructure platforms to enterprises, data centers, and cloud providers in Hong Kong, Dubai, and Mainland China. Available product lines include Lenovo ThinkSystem SR-series rack servers, ThinkSystem SD-series high-density and GPU servers, ThinkSystem ST-series tower servers, ThinkAgile VX (VMware), ThinkAgile MX (Azure Local), ThinkAgile HX (Nutanix), ThinkSystem DE and DM storage arrays, and Lenovo Neptune liquid-cooled AI platforms.

Lenovo is the world's largest PC vendor and one of the top three global server manufacturers by shipment volume. Lenovo ThinkSystem servers consistently rank among the most energy-efficient platforms on the EPA Energy Star and Green500 lists. Haink sources Lenovo equipment through authorized distribution and delivers to enterprise facilities and data centers across the Asia-Pacific and Middle East regions.

Which Lenovo for the Job — Selection by Workload

Two decisions drive most Lenovo builds: generation (new V4, or the better-priced V3 from stock) and, for dense AI, Neptune liquid nodes. Start from the workload. The generation question is less obvious than it looks — on the AMD platforms the newest silicon still ships under a V3 label — and is worked through in ThinkSystem V3 vs V4: which generation to buy in 2026.

How to Size a ThinkSystem Host — the Numbers

Sizing a server is mostly cores and RAM. Rules of thumb: budget roughly 1 physical core per 4–6 vCPUs at moderate load, and 4–8 GB RAM per VM plus ~10–15% host overhead; keep ~20–30% headroom for N+1 and bursts.

Worked example — a 150-VM consolidation at ~4 vCPU / 8 GB each: that's ~100–150 usable cores and ~1.3–1.5 TB RAM. A single 2-socket SR650 V4 (2× Intel Xeon 6 6500P/6700P, up to 86 cores and 172 threads per socket, up to 8 TB DDR5) covers it with headroom — but for HA you'd run two SR630 V4 1U hosts (N+1) instead of one big box. Databases shift the ratio toward more RAM and fewer, larger VMs; AI shifts it toward GPU (SR650a V4, SD650-N V3) and away from oversubscription.

Planning rules of thumb, not guarantees — real ratios depend on workload, oversubscription policy and SLA. We size it precisely for your estate.

Lenovo ThinkSystem SR Rack Servers

Lenovo ThinkSystem SR-series are rack-mounted enterprise servers covering single-socket, dual-socket, and four-socket configurations for virtualization, databases, cloud infrastructure, and enterprise applications. ThinkSystem SR servers support Intel Xeon Scalable and AMD EPYC processors and are managed through Lenovo XClarity Administrator.

Intel-based ThinkSystem SR Servers

Haink supplies the following Intel-based Lenovo ThinkSystem SR rack server models:

• Lenovo ThinkSystem SR630 V4 — 1U dual-socket rack server (Intel Xeon 6 P-core, up to 86 cores/socket) — current generation (GA February 2025)
• Lenovo ThinkSystem SR650 V4 — 2U dual-socket mainstream rack server (Intel Xeon 6), up to 8 TB memory and 10 rear PCIe slots — current generation
• Lenovo ThinkSystem SR650a V4 — 2U server (Intel Xeon 6) supporting four double-wide GPUs including NVIDIA H100 with NVLink — current-generation AI/GPU platform
• Lenovo ThinkSystem SR250 V3 — 1U single-socket rack server for entry-level and remote office deployments
• Lenovo ThinkSystem SR530 (1st-gen, legacy/EOL) — 1U single-socket; available as spares from stock
• Lenovo ThinkSystem SR550 (1st-gen, legacy/EOL) — 2U single-socket; spares from stock
• Lenovo ThinkSystem SR570 (1st-gen, legacy/EOL) — 2U dual-socket; spares from stock
• Lenovo ThinkSystem SR590 (1st-gen, legacy/EOL) — 2U dual-socket; spares from stock
• Lenovo ThinkSystem SR630 V3 — 1U dual-socket rack server optimized for high-density virtualization and cloud-native workloads
• Lenovo ThinkSystem SR650 V3 — 2U dual-socket rack server, the most widely deployed Lenovo platform for general-purpose data center workloads, virtualization, and databases
• Lenovo ThinkSystem SR670 V2 (2nd-gen, legacy GPU platform; superseded by SR675 V3 / SR650a V4) — 2U dual-socket GPU-optimized rack server supporting up to ten NVIDIA or AMD GPUs for AI inference and HPC
• Lenovo ThinkSystem SR860 V3 — 2U four-socket rack server for in-memory databases, large SAP HANA environments, and memory-intensive workloads
• Lenovo ThinkSystem SR950 V3 — 4U four-socket high-performance platform for the most demanding enterprise and HPC applications

The ThinkSystem V4 generation (Intel Xeon 6) is current; Lenovo extends Neptune liquid cooling from the CPU to memory with the new Neptune Core Compute Complex Module. The prior V3 generation (SR250/SR630/SR650/SR860/SR950 V3) remains widely available from stock and is often the better-priced choice.

AMD EPYC-based ThinkSystem SR Servers

Lenovo also manufactures ThinkSystem SR servers based on AMD EPYC processors, offering high core counts and large memory capacity at competitive total cost of ownership.

• Lenovo ThinkSystem SR615 V3 — 1U single-socket AMD EPYC 9004 platform for cloud-native, storage, and scale-out workloads
• Lenovo ThinkSystem SR625 V3 — 2U single-socket AMD EPYC platform for storage-dense and edge workloads requiring high drive capacity
• Lenovo ThinkSystem SR635 V3 — 1U single-socket AMD EPYC with GPU support for edge AI and inference
• Lenovo ThinkSystem SR645 V3 — 2U dual-socket AMD EPYC platform for high-core-count virtualization and containerized workloads
• Lenovo ThinkSystem SR665 V3 — 2U dual-socket AMD EPYC with up to eight GPU slots for AI training and high-performance computing
• Lenovo ThinkSystem SR675 V3 — 2U dual-socket AMD EPYC optimized for GPU-dense AI and HPC deployments, supporting up to eight NVIDIA H100 or AMD Instinct GPUs

Lenovo ThinkSystem SD High-Density and AI Servers

Lenovo ThinkSystem SD-series are high-density server platforms designed for hyperscale compute, AI training clusters, and HPC environments where maximum compute density and energy efficiency are critical. SD-series servers use Lenovo Neptune direct water cooling technology, which delivers significant power savings compared to air-cooled alternatives.

• Lenovo ThinkSystem SD630 V3 — 2U twin-node high-density compute platform for hyperscale virtualization and cloud workloads, housing two independent dual-socket nodes in 2U
• Lenovo ThinkSystem SD650 V3 — Neptune water-cooled compute node for HPC and AI workloads, supporting direct liquid cooling for Intel Xeon processors
• Lenovo ThinkSystem SD650-N V3 — Neptune liquid-cooled AI training server supporting four NVIDIA HGX H100 SXM5 GPUs per node (HGX 4-GPU board), designed for large language model training and high-performance AI workloads
• Lenovo ThinkSystem SD650-I V3 — Intel GPU-accelerated compute node supporting Intel Gaudi accelerators for AI training
• Lenovo ThinkSystem SD665 V3 — 2U twin-node AMD EPYC high-density platform housing two independent single-socket nodes in 2U, designed for cloud and HPC scale-out environments
• Lenovo ThinkSystem SD665-N V3 — Neptune liquid-cooled AMD EPYC + NVIDIA GPU node for AI training clusters requiring high GPU and CPU core density simultaneously

The Lenovo ThinkSystem SD650-N V3 is one of the most energy-efficient AI training platforms available, delivering high NVIDIA H100 GPU density with Lenovo Neptune direct water cooling, making it a preferred platform for data centers seeking to maximize AI compute density while managing power and cooling budgets.

Lenovo ThinkSystem ST Tower Servers

Lenovo ThinkSystem ST tower servers are designed for small businesses, branch offices, and deployments where rack infrastructure is not available. Haink supplies Lenovo ThinkSystem ST650 V3 and ST550 tower servers for enterprise environments requiring standalone compute.

• Lenovo ThinkSystem ST650 V3 — dual-socket tower server supporting Intel Xeon Scalable processors for office and edge deployments
• Lenovo ThinkSystem ST550 (1st-gen, legacy/EOL) — mid-range dual-socket tower; spares from stock

Lenovo ThinkEdge Edge Servers

Lenovo ThinkEdge servers are purpose-built for edge computing deployments where space, power, and environmental conditions differ from traditional data centers. ThinkEdge platforms are ruggedized for operation at manufacturing sites, retail environments, branch offices, and remote locations.

• Lenovo ThinkEdge SE350 V2 — compact 1U half-width edge server designed for space-constrained deployments, supporting Intel Xeon D processors with optional GPU for edge AI inference
• Lenovo ThinkEdge SE450 — 1U edge AI server supporting NVIDIA A2 or T4 GPUs for on-premises inference at the edge, with wide temperature and vibration tolerance
• Lenovo ThinkEdge SE360 V2 — ultra-compact edge server for IoT and telecom edge deployments

Lenovo AI Infrastructure

Lenovo is a major supplier of AI infrastructure to universities, research institutions, enterprises, and cloud providers. Lenovo AI systems appear regularly on the Top500 and Green500 lists of the world's fastest supercomputers. Haink delivers Lenovo AI infrastructure to organizations building machine learning environments, LLM training clusters, and inference platforms in Hong Kong, Dubai, and Mainland China.

Lenovo AI infrastructure platforms available through Haink:

• Lenovo ThinkSystem SD650-N V3 with NVIDIA H100 SXM5 — purpose-built for large-scale AI training and LLM workloads with Neptune liquid cooling
• Lenovo ThinkSystem SC777 V4 Neptune — supercomputing tray on the NVIDIA GB200 NVL4 Grace Blackwell platform (2× Grace CPUs + 4× Blackwell GPUs), direct water-cooled, for dense HPC and hybrid AI
• Lenovo ThinkSystem SR675 V3 with NVIDIA H100 PCIe — 2U air-cooled platform for AI training and inference with up to eight GPUs
• Lenovo ThinkSystem SR665 V3 with GPU options — dual-socket AMD EPYC with up to eight GPU slots for AI and HPC
• Lenovo ThinkSystem SR650 V3 with NVIDIA A30 or L40S — general-purpose AI inference and analytics
• Lenovo Neptune Direct Water Cooling (DWC) infrastructure — rack-level liquid cooling for high-density AI deployments

Lenovo ThinkAgile Hyperconverged Infrastructure

Lenovo ThinkAgile is Lenovo's hyperconverged infrastructure (HCI) portfolio, offering purpose-built appliances for VMware, Microsoft Azure Local, Nutanix, and Kubernetes environments. ThinkAgile systems are built on Lenovo ThinkSystem hardware and delivered as pre-validated, ready-to-deploy HCI nodes. Which of the four series you buy is decided by your hypervisor, not by the hardware — see HX, VX, MX or FX: which ThinkAgile series for which hypervisor, including cluster minimums and the FX option that is validated for all three stacks. For whether the appliance model beats building HCI on standard rack servers at all, see choosing an HCI platform.

Lenovo ThinkAgile VX (VMware vSAN)

Lenovo ThinkAgile VX is a VMware vSAN-certified HCI platform tightly integrated with VMware vSphere and vCenter. ThinkAgile VX nodes are pre-configured for VMware environments and validated for VMware Cloud Foundation (VCF) deployments.

• Lenovo ThinkAgile VX630 V3 — 1U vSAN node for high-density VMware deployments
• Lenovo ThinkAgile VX650 V3 — 2U vSAN node for general-purpose VMware environments
• Lenovo ThinkAgile VX7820 — all-flash vSAN appliance for performance-intensive VMware workloads

Lenovo ThinkAgile MX (Microsoft Azure Local, formerly Azure Stack HCI)

Lenovo ThinkAgile MX is a Microsoft Azure Local-certified platform that extends Microsoft Azure hybrid cloud capabilities to on-premises environments. ThinkAgile MX enables organizations to run Azure services locally while managing infrastructure through the Azure portal.

• Lenovo ThinkAgile MX630 V3 — single-socket Azure Local node for small and branch deployments
• Lenovo ThinkAgile MX650 V3 — dual-socket Azure Local node for enterprise hybrid cloud

Lenovo ThinkAgile HX (Nutanix)

Lenovo ThinkAgile HX is Lenovo's Nutanix-certified HCI platform, combining Lenovo ThinkSystem hardware with Nutanix AOS and Prism management. ThinkAgile HX is validated for Nutanix Cloud Infrastructure (NCI) and Nutanix Cloud Platform (NCP) deployments.

• Lenovo ThinkAgile HX630 V3 — compact Nutanix HCI node for SMB and branch environments
• Lenovo ThinkAgile HX650 V3 — enterprise Nutanix HCI node for large-scale private cloud
• Lenovo ThinkAgile HX7820 — all-NVMe Nutanix node for latency-sensitive workloads

Lenovo ThinkAgile SX (Microsoft Azure Stack Hub — sunset platform)

Lenovo ThinkAgile SX is a Microsoft Azure Stack Hub platform that delivers full Azure-consistent cloud services in an on-premises appliance. ThinkAgile SX is used by organizations in regulated industries and air-gapped environments requiring local Azure service delivery.

Lenovo ThinkSystem Storage

Lenovo manufactures a range of enterprise storage arrays for block storage, unified storage, and all-flash environments. Lenovo storage systems are designed to integrate directly with ThinkSystem servers through Lenovo XClarity management. Where these arrays sit against the wider market is covered in our enterprise storage selection guide.

Lenovo ThinkSystem DE Series

Lenovo ThinkSystem DE series are hybrid and all-flash SAN storage arrays for organizations requiring block storage integrated with VMware, Microsoft, and Linux environments.

• Lenovo ThinkSystem DE120S — entry-level 2U SAN array for small environments
• Lenovo ThinkSystem DE2000H — 2U hybrid SAN/NAS array for small and mid-size deployments requiring unified block and file access
• Lenovo ThinkSystem DE240S — 2U midrange SAN with mixed SSD and HDD drive support
• Lenovo ThinkSystem DE360S — 2U all-flash and hybrid array for performance-sensitive environments
• Lenovo ThinkSystem DE4000H — 2U hybrid all-flash capable array for mid-range performance workloads
• Lenovo ThinkSystem DE4000F — 2U all-flash array for environments requiring consistent low-latency block storage
• Lenovo ThinkSystem DE4800H — 2U high-capacity hybrid array designed for data-intensive workloads combining performance tiers with large-capacity drives
• Lenovo ThinkSystem DE4800F — 2U all-flash high-density array for demanding virtualization, database, and analytics environments
• Lenovo ThinkSystem DE480S — high-capacity storage-dense array for large-scale block storage requirements
• Lenovo ThinkSystem DE6400H — enterprise-grade high-performance hybrid array for large data center storage environments
• Lenovo ThinkSystem DE6400F — enterprise all-flash array for mission-critical databases and applications requiring maximum IOPS and sub-millisecond latency

Lenovo ThinkSystem DM Series

Lenovo ThinkSystem DM series are ONTAP-based unified storage platforms (built with NetApp under the Lenovo–NetApp partnership, current) providing both block (iSCSI, Fibre Channel) and file (NFS, SMB/CIFS) storage access. DM-series arrays run ONTAP software (in partnership with NetApp) for advanced data management capabilities.

• Lenovo ThinkSystem DM3010H — entry unified storage for mixed workloads
• Lenovo ThinkSystem DM5000H — midrange unified storage for growing environments
• Lenovo ThinkSystem DM7100H — high-performance hybrid unified storage
• Lenovo ThinkSystem DM7100F — all-flash unified storage for performance-critical applications

Lenovo ThinkSystem DG Series (QLC All-Flash)

Lenovo ThinkSystem DG Series are end-to-end QLC NVMe all-flash unified arrays for high-density, cost-efficient capacity — up to roughly 6× the performance of HDD-based storage at about half the cost.

• Lenovo ThinkSystem DG5000 — QLC all-flash array for capacity-optimized primary and secondary storage
• Lenovo ThinkSystem DG7000 — higher-performance QLC all-flash array for larger, denser deployments

Why Organizations Choose Lenovo Servers

Lenovo ThinkSystem servers are frequently selected by enterprises and data centers for several reasons:

• Competitive total cost of ownership compared to HPE and Dell platforms of equivalent specification
• Strong AMD EPYC platform support, offering high core counts at lower cost per core
• Lenovo Neptune liquid cooling technology, enabling higher compute density with lower power consumption
• Consistent presence in Green500 and Top500 supercomputer rankings, demonstrating HPC and energy efficiency credentials
• Deep integration with VMware, Microsoft, Nutanix, and Red Hat ecosystems through ThinkAgile HCI platforms
• Lenovo TruScale as-a-service consumption model for flexible infrastructure procurement

Stock, BTO and CTO — What Actually Ships When

• Stock servers — Lenovo ThinkSystem SR-series servers available from regional distributor stock for rapid delivery
• Standard OEM configurations — factory-standard Lenovo configurations sourced through authorized distribution
• BTO (Build to Order) — custom Lenovo configurations to project specifications, including CPU, memory, storage, GPU, and networking options
• CTO (Configure to Order) — fully customized Lenovo platforms for specialized AI, HPC, or high-density workloads

Which route you take changes both price and lead time: see stock versus BTO/CTO lead times and pricing. At project volumes the pricing mechanism changes again — special bid and project pricing is how a 20-node order is quoted, and it is not the same as a list-price discount.

Related Resources

• ThinkSystem V3 vs V4 — why the version number does not tell you how new the processor is, and what per-core licensing does to core count
• Four-socket and scale-up — SR850 V4 and SR860 V4, and the one row that decides scale up against scale out
• Model numbers decoded — the last digit is the processor vendor: 0 is Intel, 5 is AMD
• Warranty and Premier Support — a response objective is not a repair commitment, and coverage is a per-site question
• Memory population rules — filling every slot costs a fifth of the bandwidth on Intel and gains a third on AMD
• SR650 V3 vs SR665 V3 — Intel or AMD in the same 2U: 320 cores against 8 TB of memory
• XClarity licensing — remote console and virtual media are not in the included tier, and the fleet platform is a second licence
• SR630 V4 vs SR650 V4 — same cores and memory, one third of the I/O, and twice the sockets per rack unit
• MTM and part numbers — why two quotes for the same server can differ by two years of warranty
• Lifecycle and end of support — what is current, what is withdrawn, what replaced it, and how to get your real dates
• ThinkAgile HX vs VX vs MX vs FX — which series for which hypervisor, and the cluster minimums that set your entry price
• Lenovo stock, lead times and pricing — current availability and firm quotes
• Lenovo vs Dell vs HPE — the current generation compared three ways
• Dell PowerEdge vs HPE ProLiant — the two-way comparison of the incumbents
• VxRail vs Nutanix — choosing the HCI platform before the appliance
• Lenovo storage vs NetApp — same ONTAP, different relationship: what actually differs and how to verify it
• NetApp vs Pure Storage — the platform decision that comes first
• Choosing an HCI platform — appliance versus build-your-own
• Enterprise storage selection guide — all-flash, NAS and SAN
• Stock vs BTO/CTO — what each route costs in time and money
• Special bid and project pricing — how volume orders are actually quoted
• Gray-market channel risks — why MTM and serial verification matters
• Sovereign cloud deployment — 40 nodes, 7,680 vCPUs, national data residency
• HPE ProLiant selection guide
• Dell PowerEdge selection guide
• NetApp selection guide
• All brand selection guides

Lenovo Storage Is NetApp Underneath — What That Means

A point that trips buyers up: Lenovo's enterprise storage is NetApp OEM. The DM-series and the newer DG-series (QLC all-flash) run NetApp ONTAP; the DE-series is NetApp E-Series (SANtricity, block/SAN). If you already run ONTAP, Lenovo DM/DG is the same operating system and feature set with Lenovo's supply and support — frequently a better-priced route to ONTAP than buying NetApp-branded. The full comparison — how to verify the lineage yourself, what actually differs in support, version timing and licence packaging, and whether the arrays can replicate to NetApp already in your estate — is in Lenovo storage vs NetApp. If ONTAP is not already your standard, decide the platform before the label: NetApp versus Pure Storage covers that choice.

When Lenovo Isn't the Answer

We stay neutral, because a wrong standardization is costly:

• If you're globally standardized on one vendor's support (Dell ProSupport, HPE Pointnext), a single-vendor estate can beat mixing — see how PowerEdge and ProLiant compare before you assume the incumbent is the cheaper path — — Lenovo wins on price/performance and Neptune warm-water cooling, not on being different for its own sake.
• For the densest GPU training at scale, Supermicro and Dell XE offer more 8-GPU SXM SKU variety; Lenovo's edge is Neptune liquid efficiency and TCO.
• TaiShan / Kunpeng ARM is a Mainland-China domestic-processor play — outside China, stay on x86 ThinkSystem.
• ThinkAgile: pick MX (Azure Local), HX (Nutanix) or VX (VMware) by your hypervisor/cloud strategy — and if you're already standardized on VxRail or Nutanix-direct, that may fit better.

Buying Genuine Lenovo — Verify MTM & Serial Before Payment

Every Lenovo system has a machine-type-model (MTM) and serial that tie it to warranty and support entitlement. Haink sources through authorized channels and provides the MTM and serial before payment, so you can confirm on Lenovo's support site that the unit is genuine, in-warranty for your region and entitled to support — the reliable check against gray-market or previously-registered hardware. The same discipline applies to any quote you receive elsewhere: how gray-market risk actually reaches the buyer and how to verify an in-stock offer before payment.

Five Expensive Mistakes We See in Lenovo Procurement

1. Buying V4 when V3 from stock was the deal. For standardized virtualization/databases that don't need Xeon 6, V3 ships now and costs less.
2. Designing on 1st-gen ThinkSystem. 2017-era SR530/550/570/590 and ST550 are legacy/EOL — spec current V3/V4. The full platform status and successor map shows what replaced what.
3. Assuming SD650-N V3 is 8 GPUs. It's 4× HGX H100 per node (an HGX 4-GPU board), not 8 — size clusters accordingly.
4. Paying NetApp-branded prices when Lenovo DM/DG gives you the same ONTAP with better-priced supply.
5. Skipping MTM/serial verification. Gray-market ThinkSystem may be out of warranty — verify before payment.

Frequently Asked Questions

V3 from stock or V4 to order — which ThinkSystem now?

Buy V4 (Intel Xeon 6) when you need the latest core counts and efficiency, or for new AI/GPU builds (SR650a V4, SC777 V4). Buy V3 (SR630/SR650 V3) from stock when the workload is standardized virtualization or databases — it ships this week and is usually better-priced. We quote both against your workload, and size cores/RAM/GPU precisely.

Is Lenovo storage really NetApp, and should that change my choice?

Yes. Lenovo DM and DG storage run NetApp ONTAP; DE is NetApp E-Series. If ONTAP is your standard, Lenovo DM/DG gives you the same OS and features with Lenovo's supply and support — often better-priced than NetApp-branded. If you have no ONTAP investment, weigh it against a native NetApp AFF or another platform on support and roadmap, not just price.

Where do you ship Lenovo from, and how fast?

ThinkSystem SR630/SR650 V3 ship from Hong Kong stock (free port, no duty), often within days; V4, GPU (SR650a/SD650-N), ThinkAgile and storage are typically 2–12 weeks to order. We also ship from Dubai for the GCC/MENA and handle import documentation and customs for Mainland China.

What is the Lenovo ThinkSystem SD650-N V3 and who uses it?

The Lenovo ThinkSystem SD650-N V3 is a Neptune liquid-cooled AI training server supporting four NVIDIA HGX H100 SXM5 GPUs per node (HGX 4-GPU board). It is designed for large-scale AI training, large language model development, and high-performance computing clusters. Its direct water cooling technology delivers significantly higher energy efficiency than air-cooled GPU servers, making it a preferred choice for organizations building dense AI infrastructure with strict power and cooling constraints.

What is the difference between Lenovo ThinkSystem SR650 and SR675?

The Lenovo ThinkSystem SR650 V3 is a dual-socket Intel Xeon-based 2U rack server designed for general-purpose data center workloads including virtualization, databases, and enterprise applications. The Lenovo ThinkSystem SR675 V3 is a dual-socket AMD EPYC-based 2U platform specifically optimized for GPU-dense AI and HPC deployments, supporting up to eight NVIDIA H100 PCIe or AMD Instinct MI300X GPUs. The SR675 V3 is typically chosen for AI training clusters where high GPU density per rack unit is a priority.

What is Lenovo Neptune and why does it matter for AI deployments?

Lenovo Neptune is Lenovo's portfolio of direct liquid cooling (DLC) technologies for ThinkSystem servers. Neptune uses warm-water direct cooling on CPUs and GPUs, removing heat more efficiently than air cooling and allowing significantly higher compute density per rack. For AI training deployments using NVIDIA H100 SXM5 GPUs — which dissipate up to 700W each — liquid cooling is often a requirement to stay within data center power and cooling limits. Lenovo's Neptune-cooled SD650-N V3 is one of the densest and most efficient NVIDIA H100 platforms available.

What Lenovo is in stock now, and what are realistic lead times?

ThinkSystem SR630/SR650 V3 are typically in stock in Hong Kong (days–1 week); V4, GPU platforms, ThinkAgile and storage are usually 2–12 weeks to order. Stock moves — see the live table on our Lenovo supply page and request current availability for your models.