AMD's EPYC 9005 series, codenamed Turin, lands on the same SP5 socket as the Genoa 9004 parts before it - and that one fact reframes the whole buying decision. When a new generation drops into your existing platform, the question stops being can I afford the upgrade and becomes is the per-socket density worth refreshing for, or do I ride the platform I already have. Turin pushes core counts and per-core performance up with the Zen 5 architecture while keeping the memory and I/O platform familiar. For UK buyers, this is a consolidation and total-cost story more than a raw-speed one. Here is the wait-or-refresh take.
Same socket, new architecture
Turin's headline for anyone with an existing AMD estate is SP5 compatibility. Genoa (Zen 4) and Turin (Zen 5) share the platform, the memory subsystem family and much of the surrounding I/O, so a refresh is far less disruptive than a socket change - in many validated platforms it is a CPU and firmware step rather than a forklift. That lowers the barrier to a mid-life upgrade and makes a like-for-like density jump genuinely tempting.
Zen 5 raises instructions-per-clock over Zen 4, so per-core performance improves at a given frequency. Combined with higher top-end core counts in the dense parts, a single Turin socket can take on noticeably more work than the Genoa part it replaces - which is the lever that drives consolidation, not just benchmark bragging rights.
Core density and the consolidation case
EPYC's whole pitch has been core density per socket, and Turin extends it. The high-density parts reach very high core counts in a two-socket box, which means fewer servers, fewer sockets and fewer racks to do the same job. For UK organisations where rack space and power draw are hard limits - and increasingly where grid capacity itself is the constraint - that consolidation is where the money is.
The trade is the familiar per-core one. If your software is licensed per core (VMware, SQL Server, Oracle, Windows Server Datacenter), maximum core count is a recurring cost, not a free win, and you should reach for the higher-frequency, lower-core parts instead. If your stack is open-source and scales with threads, the dense parts let you collapse a lot of older hardware. Decide which camp you are in before you choose the SKU, as we set out in server configuration.
- •High-density Turin parts: collapse many older nodes where software is not per-core-licensed
- •High-frequency lower-core parts: best for per-core-licensed databases and virtualisation
- •SP5 compatibility: refresh is often a CPU + firmware step, not a platform replacement
Memory and I/O: 12 channels and PCIe 5.0
Turin keeps the generous SP5 memory and I/O envelope: twelve DDR5 channels per socket and a large complement of PCIe 5.0 lanes, with CXL support on the platform. Twelve channels is the quiet advantage for memory-bandwidth-bound workloads - analytics, large in-memory databases and dense virtualisation - because aggregate bandwidth scales with channel count, and AMD's channel count has been a real differentiator.
As ever, that bandwidth only materialises if every channel is populated and balanced; a lopsided DIMM layout throttles the platform regardless of CPU. The abundant PCIe 5.0 lanes also matter for all-NVMe storage and GPU-attached nodes, letting a single host carry more drives or accelerators. Plan the memory layout with our memory and RAM guidance so you do not strand bandwidth.
Turin versus Genoa versus Intel: how to frame it
Against Genoa, Turin is a clean step up in per-core performance and top-end density on a socket you may already own - so the decision is timing, not capability. Against Intel's current Xeon, the comparison is the long-running one: AMD tends to lead on core density and memory channels per socket, while Intel's Xeon 6 P-cores compete hard on per-core performance and bring accelerator and platform features of their own. Neither wins everywhere.
For a UK buyer the honest framing is workload-led. Core-dense, bandwidth-hungry, not-per-core-licensed workloads lean AMD; per-core-licensed latency workloads can go either way and should be costed across the whole software stack. We benchmark the realistic options rather than trusting a single headline number, and compare platforms in Dell PowerEdge servers where both vendors' silicon is available.
Wait or refresh: the practical answer
If you are on first-generation EPYC (Naples) or Rome, Turin is a large generational jump and the consolidation, power and density gains usually justify a refresh outright. If you are already on Genoa, the case is narrower: refresh when you are out of headroom, out of rack power, or chasing further consolidation to cut software and energy cost - otherwise the platform you have will serve happily, and SP5 means Turin will still be a low-friction upgrade later.
Either way, the platform longevity of SP5 is itself a buying argument: it protects the investment whether you jump now or next cycle. Pick the exact EPYC part with our processors guidance and build the validated host in our server configuration service.