Skip to content

Dear customers! Please send orders and inquiries to [email protected] or to Telegram @servix_uz instead of calling by phone: due to the high workload, the operator cannot answer all incoming calls.

Servix

Intel Xeon 6 or AMD EPYC 9005: which to choose for virtualization

Servix technical team · Updated

In brief. The difference between Intel Xeon 6 and AMD EPYC 9005 is smaller than the difference between a good and a bad configuration: the choice is dictated by per-core licensing. VMware counts a minimum of 16 cores per socket, so dense chips with 64–⁠128 cores are more cost-effective than machines with 16 cores each. Xeon 6 offers up to 128 P-cores and 12 memory channels, EPYC 9005 up to 192 Zen 5c cores at 155–⁠500 W. Midsize businesses hardly need 500 W flagships: an office server room supports 3–⁠5 kW per rack.

Choosing a processor for a virtualization cluster has long since stopped being an argument about clock speeds. What needs to be calculated is something else: how many virtual machines will fit on a node, what the licenses for those cores will cost, and whether the server room can handle the heat output of modern chips. Let’s look at how the current Intel and AMD lines differ and in which cases each of them is a sensible choice.

What the lines are

Intel Xeon 6 is the generation code-named Granite Rapids, introduced in September 2024. The line is divided into two families with different sockets. The 6900P series is designed for AP platforms: up to 128 performance cores per processor, twelve memory channels, up to 500 W per chip. The flagship 6980P has 128 cores and 256 threads, a base frequency of 2.0 GHz, 3.2 GHz with all cores loaded, and 504 MB of L3 cache. The 6700P series is designed for more mainstream SP platforms: eight memory channels, but with the ability to install two modules per channel.

Separately, Intel offers versions with efficient E-cores (Sierra Forest): they do not support multithreading, but on the AP platform they provide up to 288 cores per processor. This is a specialized option for large-scale uniform workloads such as web front ends and microservices, not a universal replacement for P-cores.

AMD EPYC 9005 is the fifth generation, code-named Turin. The line’s range is wider than its competitor’s: from 8 to 192 cores and from 155 to 500 W. There are two types of cores. Classic Zen 5 provides up to 128 cores and up to 512 MB of L3 cache. Dense Zen 5c, made on a 3 nm process, provides up to 192 cores and 384 MB of cache — this is the flagship 9965 with 384 threads. There are twelve memory channels, as in Intel’s top series.

The main argument of 2026: licenses are counted by cores

This is a case where the cost of software dictates the choice of hardware. Since VMware moved to a subscription model, licensing has been per core, and two minimums apply. The first: no fewer than sixteen cores per socket, even if the processor physically has twelve. The second, introduced in April 2025 and canceled later the same year: a minimum order of 72 cores per subscription. While it was in effect, a company with a single server on an eight-core processor still paid for 72 cores; now only the threshold of 16 cores per socket remains.

This leads to two practical conclusions. First: buying processors with fewer than sixteen cores for VMware makes no sense — you pay for sixteen anyway. Second: the fewer physical nodes for the same total capacity, the fewer cores are “lost” to minimums and the easier it is to stay within subscription limits. Dense processors with 64–⁠128 cores become more cost-effective than a set of machines with 16 cores each, even if the advantage in hardware price is not obvious.

The same logic applies to Windows Server: Datacenter and Standard licenses are counted by the server’s physical cores, with their own minimums. The project cost must be calculated together with licenses; otherwise, savings on the processor will turn into an overpayment for software that is many times larger.

Comparison by key parameters

Parameter

Intel Xeon 6 (Granite Rapids)

AMD EPYC 9005 (Turin)

Maximum cores per processor

128 P-cores (6900P series); up to 288 E-cores in the version with efficient cores

128 Zen 5 cores or 192 Zen 5c cores

L3 cache

Up to 504 MB (6980P)

Up to 512 MB for Zen 5, up to 384 MB for Zen 5c

Memory channels

12 for the 6900P series, 8 for the 6700P series

12 across the entire line

Memory speed

DDR5-6400, up to 8,800 MT/s with MRDIMM modules

DDR5-6400

TDP range

Up to 500 W on top models

155–⁠500 W

Multithreading

Yes on P-cores, no on E-cores

Yes across the entire line

The difference in specification figures is small: both platforms have reached twelve memory channels and half a kilowatt per socket. The difference shows in the details. Intel has a trump card in MRDIMM memory, which raises bandwidth where the workload is limited specifically by memory. AMD has a wider choice in the lower part of the line and the maximum core density in a single socket.

What this means for different tasks

Dense general-purpose virtualization. Cores per socket and memory per node win: the more virtual machines on one host, the fewer nodes, licenses and switch ports. Top EPYC models and top Xeon 6900P models are strong here. A practical guideline: if 60–⁠100 virtual machines with a medium load are planned per node, look at processors with 48 cores or more and 512 GB of memory or more.

Databases. Here, what matters is not the total number of cores but frequency, cache and memory latency. The large L3 cache of both platforms helps, but the choice is more often dictated by DBMS licensing: commercial databases are also licensed per core, and a processor with fewer fast cores may turn out cheaper to operate than a dense chip.

Mixed workloads of a midsize business. Flagships are overkill. The sensible choice is mid-range two-socket servers: HPE ProLiant DL380 Gen11 and Dell PowerEdge R760 with Intel processors, HPE ProLiant DL385 Gen11 and DL365 Gen11 with AMD EPYC. These platforms accept processors across a wide range, so the configuration can be precisely matched to licensing minimums.

The restriction people remember last: power and cooling

A 500 W processor is not just an electricity bill. A two-socket server with top chips, full memory and drives easily exceeds a kilowatt under load. A rack of ten such machines already requires more than ten kilowatts of supplied power and corresponding heat removal, while a typical office server room is designed for three to five kilowatts per rack.

So before choosing top-end processors, check three things: how many kilowatts are supplied to the room, what the UPS capacity is, taking the required runtime into account, and whether the cooling can cope with the additional heat load. It often turns out that two mid-range servers fit into the existing infrastructure, while one flagship does not. We covered a checklist for preparing the room in the article on an office server room.

How to choose in practice

Start not with the processor but with a calculation: how many virtual machines, how many cores and how much memory for each, and what headroom for growth and node failure. Divide the resulting number of cores among the nodes so that the cluster stays operational if one of them fails. Then check what the licenses for the resulting configuration will cost — and only after that choose a specific processor model.

On typical midsize business tasks, the difference between Intel and AMD platforms is smaller today than the difference between a good and a bad configuration of the same platform. The result depends far more on memory capacity and its distribution across channels, the type of drives, and how precisely the number of cores is matched to licensing minimums.

To select a configuration for a specific workload and calculate it together with licenses, write to us. Available platforms of both architectures are in the Serverssection: HPE ProLiant and Dell PowerEdge.

When a generation upgrade is justified

Replacing working servers with a new generation for the sake of the generation itself rarely pays off. It is worth calculating for three reasons. First: vendor support for the current machines has ended or is ending — then the risk of downtime because a spare part cannot be obtained outweighs the cost of the upgrade. Second: the nodes have hit the limit of memory or cores, and there is nothing to add — all slots are occupied. Third, and most significant: consolidation. If five old two-socket servers are replaced by two new ones, the savings come from three items at once — licenses, electricity and rack space — and the project pays for itself in a couple of years.

The opposite case also happens: if the workload is stable, support is in effect and there are free memory slots, it is cheaper to add memory and drives to the existing machines. An upgrade for the sake of a line in the specification is the worst purchasing scenario.

Related collections and tools: Server for virtualization, Server processors, Server configurator.

Questions and answers

Which is more cost-effective: two nodes with 32 cores each or one with 64?

In terms of licenses, usually one 64-core node, because minimums are counted per socket. In terms of fault tolerance, two nodes; otherwise, a server failure stops everything. In practice, a cluster is not built with fewer than two nodes, and beyond that you choose the minimum number of nodes at which the failure of one does not bring down the service.

Do midsize businesses need 500 W processors?

Almost never. Top models are designed for workloads with dozens of nodes, where every rack unit counts. For a company with two to four servers, the mid-range segment is more sensible: it fits into the existing power and cooling and costs noticeably less.

Is it worth buying processors with efficient E-cores?

Only for a specific scenario: many uniform lightweight threads without single-thread performance requirements — web front ends, cache servers, microservice containers. For a general-purpose virtualization cluster, which also runs frequency-sensitive applications, classic performance cores are the safer choice.

Related reading

HPE introduced ProLiant Gen12 on Intel Xeon 6

ProLiant Gen12 on Intel Xeon 6: more cores and memory channels, new iLO, firmware protection. When a generation change is justified and when Gen11 from the catalog is the better deal.