Dell PowerEdge C6620 vs C6615
Dell PowerEdge C6620
Dell PowerEdge C6615
The PowerEdge C6620 and C6615 are both compute sleds for Dell's dense PowerEdge C6600 chassis, the shared-infrastructure platform built for HPC clusters, hyperscale, and large scale-out fleets. The difference is silicon and socket count. The C6620 is a dual-socket Intel Xeon sled, and the C6615 is a single-socket AMD EPYC sled. Both drop into the same 2U C6600 enclosure at up to four sleds each, share its power and cooling, and use the same Dell management stack, so the decision is not about which is broadly better. It comes down to per-node compute profile, core economics, and which CPU ecosystem the cluster standardizes on.
Side by side
| Dell PowerEdge C6620 | Dell PowerEdge C6615 | |
|---|---|---|
| Platform & sockets | Dual-socket (2P) Intel Xeon sled for the PowerEdge C6600 shared-infrastructure chassis | Single-socket (1P) AMD EPYC sled for the same PowerEdge C6600 chassis |
| Processor family | 4th/5th Gen Intel Xeon Scalable (Sapphire Rapids / Emerald Rapids), two per sled | AMD EPYC 9004 series (Genoa / Genoa-X / Bergamo), one per sled |
| Compute approach & cores | Two sockets per node deliver high aggregate cores, with strong per-core and high-frequency SKU choices plus Intel AMX/AVX-512 acceleration | One socket per node delivers very high core density from a single CPU (up to 96 cores on Genoa, up to 128 on Bergamo) with no inter-socket NUMA hop |
| Memory | DDR5 across two sockets; the second socket adds aggregate memory bandwidth and capacity per sled for bandwidth-bound jobs | DDR5 on 12 channels from one socket; strong single-socket bandwidth in a simpler, flatter memory domain |
| Shared chassis (C6600) | Half-width sled, up to four per 2U C6600; shares enclosure power and cooling for density and efficiency | Same half-width form factor, up to four per 2U C6600, sharing the same power and cooling; C6620 and C6615 sleds can even mix in one chassis |
| Best-fit workloads | Intel-tuned HPC and AI (AMX/AVX-512, oneAPI), memory-bandwidth-heavy simulation, and clusters standardized on Intel toolchains | Core-dense and throughput-bound HPC, cloud-native and containerized scale-out, and per-core-licensing-sensitive fleets |
| Density & licensing economics | Two sockets per node can raise per-socket software exposure, but fewer, bandwidth-rich nodes may cover a bandwidth-bound cluster | One high-core socket per node can lower per-socket licensing and pack more compute per rack U for scale-out building blocks |
| Management & service | iDRAC9 with Lifecycle Controller and OpenManage; standard PowerEdge firmware, security, and ProSupport/ProDeploy options | iDRAC9 with Lifecycle Controller and OpenManage; identical management stack and the same Dell service options |
Shop these now
Live configurations from our catalog with partner pricing. Add to your cart to request a firm quote, or build a full BOM.
Dell PowerEdge C6620
Dell PowerEdge C6615
Choose the Dell PowerEdge C6620 when the cluster is Intel-tuned or leans on the second socket
Pick the C6620 when workloads benefit from Intel-specific acceleration such as AMX for AI inference and AVX-512 for simulation, when ISV or HPC codes are validated and tuned on Xeon, or when the environment is already standardized on Intel drivers, oneAPI, and operational tooling. The dual-socket sled also matters for memory-bandwidth-bound jobs, since the second socket adds aggregate bandwidth and capacity per node, which can mean fewer nodes for the same result on certain solvers. It is the natural building block for shops that want their dense C6600 cluster to match an existing Intel software and support standard. Size the sled count and CPU SKUs against the real job mix in a /quote, and drop a full chassis loadout into a /bom for the cluster build.
Choose the Dell PowerEdge C6615 when core density and single-socket economics win
Pick the C6615 when one AMD EPYC per sled delivers all the cores, memory bandwidth, and PCIe Gen5 lanes a node needs, without the cost, power, and NUMA complexity of a second socket. Its high single-socket core counts (up to 96 on Genoa, up to 128 on Bergamo) make it excellent for throughput-bound and highly parallel HPC, cloud-native and containerized scale-out, and any fleet where per-core software licensing rewards packing more compute into fewer sockets. Because it is a single-socket design in the same dense C6600 chassis, it is often the cleaner, more repeatable scale-out building block. Model the per-core licensing and node count in a /bom, and confirm the configured sleds and chassis in a /quote before the cluster order.
Neither sled is strictly better, because they target different design goals inside the same dense platform. The C6620 is the dual-socket Intel node: strong per-core and high-frequency options, AMX and AVX-512 acceleration, and more aggregate memory bandwidth per sled for bandwidth-bound HPC and Intel-standardized clusters. The C6615 is the lean single-socket AMD node: outstanding core density, a flat single-socket memory domain, and per-core licensing economics that suit throughput-bound and scale-out fleets. As a reseller guideline, steer Intel-tuned, second-socket-bandwidth, or Intel-standardized customers to the C6620, and steer core-dense, throughput-per-watt, or licensing-sensitive scale-out to the C6615. Both share the C6600 chassis, iDRAC9, and OpenManage, so day-two operations look the same, and the two can even coexist in one enclosure. Final config drives both performance and price more than the model name, so map the workload and license model against a configured /quote, and build the full chassis and rack in a /bom. Uniqcli sells and supports both sides and can scope the right mix for the cluster.
Talk to a specialistFrequently asked
Can the C6620 and C6615 share the same chassis?
Yes. Both are half-width compute sleds for the same 2U PowerEdge C6600 shared-infrastructure chassis, which holds up to four sleds and shares power and cooling across them. That means you can run Intel C6620 sleds and AMD C6615 sleds in the same enclosure, which is useful when different parts of a cluster favor different silicon. If you want a mixed loadout, we can lay it out sled by sled in a /bom.
Which one is better for HPC?
It depends on the code. Bandwidth-bound solvers and Intel-tuned or AMX/AVX-512-accelerated workloads often favor the dual-socket C6620, since the second socket adds aggregate memory bandwidth and Intel-specific acceleration per node. Throughput-bound, highly parallel, and containerized workloads often favor the single-socket C6615 for its high core density and per-core licensing economics. Benchmark the actual application, then quote the sled and CPU mix that matches it.
Is the C6615 single-socket only, and does that limit it?
Yes, the C6615 is a single-socket AMD EPYC sled, while the C6620 is a dual-socket Intel sled. Single socket is not a weakness for many dense workloads, because one EPYC 9004 CPU offers up to 96 cores on Genoa or 128 on Bergamo, 12 DDR5 channels, and plenty of PCIe Gen5 lanes, with no inter-socket NUMA penalty. The trade-off is the per-node ceiling. For the highest aggregate memory bandwidth or capacity per node, the dual-socket C6620 has more room, so pick based on whether the workload wants denser nodes or bigger nodes.
Do they share the same management and warranty tooling?
Yes. Both use iDRAC9 with Lifecycle Controller and OpenManage Enterprise, along with the same firmware, update, and security tooling, so they slot into an existing Dell-managed cluster identically. They also share the same Dell ProSupport and ProDeploy service options. The decision is about hardware fit, sockets, core density, and CPU architecture, rather than day-two operations, which look the same across both sleds.
Are these available for federal HPC procurement?
Yes. PowerEdge C6600-class dense compute is well suited to federal and research HPC, and Uniqcli supports TAA-compliant configurations and federal buying paths including NASA SEWP V, GSA, and GPC purchases. Send the cluster requirements through a /quote and we will scope the compliant sled, chassis, and rack build, then package it as a /bom for the acquisition.
Build your Dell bill of materials.
Send us the requirement, the project, or an existing quote to beat. We come back with a validated, TAA-compliant Dell configuration and a real price, often below list.
[email protected] · Chicago, IL