A Technical Guide to Every CPU Generation in OpenMetal’s Lineup

You’re scanning the hardware catalog and the tiers make sense: Medium, Large, XL, but the CPU model numbers underneath them don’t mean much on their own. A Xeon Gold 6530 and a Xeon 6530P look almost identical at a glance. But they’re not the same chip, and the difference matters more than the naming suggests!

OpenMetal’s current catalog spans four distinct Intel Xeon generations. Knowing which generation actually powers a given server tier, and what that generation does or doesn’t support, is the difference between picking hardware that fits your workload and picking hardware that almost fits.

Why Generation Matters More Than Core Count Alone

Two servers with the same core count can perform meaningfully differently if they’re built on different silicon generations. Cache size, memory bandwidth, instruction set extensions, and power efficiency all shift generation to generation, often more than clock speed alone suggests.

For workloads that depend on confidential computing specifically, generation is a non-negotiable: Intel TDX requires 5th Gen Xeon Scalable (Emerald Rapids) or newer, so anything built on an older generation is out of consideration regardless of core count or RAM.

Granite Rapids: The Current v5 Lineup

OpenMetal’s v5 servers, Medium v5, Large v5, and XL v5, along with the RTX Pro 6000 and H200 GPU servers, all run on Intel’s Xeon 6 platform, codenamed Granite Rapids, built on the Intel 3 process node.

What Changed From Previous Generations

Granite Rapids brings a substantial cache increase over the previous generation, DDR5-6400 memory for meaningfully higher bandwidth, and a lower TDP per socket despite higher sustained clocks, a result of the smaller process node. Per Intel’s own published specifications, the Xeon 6517P carries 72MB of L3 cache at 190W TDP, compared to the Xeon Gold 6526Y’s 37.5MB of cache at 195W TDP in the equivalent Large v4 tier, a genuine near-doubling of cache at a comparable power envelope. AMX (Advanced Matrix Extensions) support is also standard across the generation, relevant for workloads doing matrix-heavy computation.

The cache increase isn’t just a bigger number on a spec sheet. L3 cache determines how often the CPU has to leave its fastest storage tier and reach out to main memory, and that round trip costs real cycles every time it happens. Workloads with large, frequently-accessed working sets, in-memory databases, analytics engines scanning large datasets, containerized services juggling many concurrent connections, all benefit directly from more cache sitting closer to the core, independent of raw clock speed or core count.

This is part of why OpenMetal positions Medium v5 specifically for containerized databases, mid-scale Kubernetes, and multi-threaded analytics: 96MB of total L3 cache across the dual-socket configuration gives those workload types meaningfully fewer cache misses than the previous generation at the same tier.

Medium v5, Large v5, and XL v5 Differ in More Than Size

Medium v5 runs the Xeon 6505P, 12 cores per socket at 2.2 GHz base, 4.1 GHz turbo. Large v5 steps up to the Xeon 6517P, 16 cores per socket at a notably higher 3.2 GHz base clock, which matters for workloads more sensitive to single-thread performance than raw core count. XL v5 runs the Xeon 6530P, 32 cores per socket at 2.3 GHz base and 4.1 GHz turbo, the highest core density in the generation.

Mapped to real deployments, this plays out as three genuinely different fits rather than a single tier scaled up three times:

• Medium v5’s cache-per-core advantage and moderate core count suit a mid-sized containerized application or a development and staging environment that needs production-grade hardware without production-scale core density.
• Large v5’s higher clock speed fits workloads where a small number of threads need to run as fast as possible, a latency-sensitive API layer, a database primary handling synchronous writes, blockchain validator software where per-block processing speed matters more than parallel throughput.
• XL v5’s core count is built for genuine horizontal scale, a hosted private cloud control plane serving many tenant VMs at once, or a CI/CD fleet running dozens of build jobs in parallel where aggregate throughput matters more than any single job’s latency.

TDX eligibility within v5 isn’t purely a generational question. XL v5 ships with all memory channels fully populated and is TDX-ready out of the box. Large v5 supports TDX at the silicon level but, like Large v4 in the previous generation, needs a RAM configuration change to reach full channel population before TDX can actually activate.

Emerald Rapids: Most of the v4 Lineup

This is where the naming gets less intuitive. Most of OpenMetal’s v4 tier, Large v4, XL v4, and XXL v4, runs on 5th Gen Intel Xeon Scalable, codenamed Emerald Rapids, on the Intel 7 process node. This is a mature, well-proven platform, and it’s the generation floor for TDX eligibility.

“Mature” here means something concrete. Emerald Rapids has had roughly two years of production deployment across the industry by the time Granite Rapids became broadly available, which translates into a deeper base of validated software compatibility, well-worn kernel and hypervisor support, and driver stacks that have had longer to shake out edge cases. For teams running specific software with narrow, tested compatibility lists, database engines, storage backends, licensed enterprise software, that track record carries real weight independent of the raw performance numbers, since running on a newer platform occasionally means being an early adopter of compatibility fixes you’d rather not be.

Large v4 vs. XL v4 vs. XXL v4

Large v4 runs the Xeon Gold 6526Y, 16 cores per socket. XL v4 and XXL v4 both step up to the Xeon Gold 6530, 32 cores per socket, the same chip across both tiers. The real difference between XL v4 and XXL v4 isn’t the CPU at all, it’s RAM and storage capacity, with XXL v4 offering a larger memory and drive footprint on identical processors for workloads that need the extra headroom without a faster or slower chip underneath them.

That distinction matters for sizing decisions specifically. A workload that’s core-bound but not memory-hungry, a compute-heavy render farm, a CI/CD fleet running many parallel but modest jobs, gets no benefit from XXL v4 over XL v4, since the processor doing the actual work is identical. A workload that’s memory-bound, a large in-memory cache layer, a database with a big working set that benefits from staying entirely in RAM, an analytics job processing datasets that don’t fit comfortably in XL v4’s ceiling, is exactly where the extra headroom on XXL v4 earns its higher price rather than paying for compute you won’t use.

TDX Availability Varies by Tier Even Within Emerald Rapids

XL v4 and XXL v4 both ship TDX-active out of the box, with all memory channels populated at 1TB or 2TB depending on tier. Large v4 supports TDX at the silicon level but needs a RAM upgrade, filling the remaining DIMM slots to reach full channel population, to actually activate it, a real option but one that’s rarely cost-effective compared to simply starting with an XL v4 if TDX is a firm requirement.

Sapphire Rapids: Medium v4, the Outlier

Medium v4 is genuinely different from the rest of the v4 lineup. It runs the Xeon Silver 4510, which is 4th Gen Intel Xeon Scalable, codenamed Sapphire Rapids, one generation behind the Emerald Rapids chips powering the rest of the v4 tier. This is a meaningful distinction worth being precise about: Medium v4 isn’t TDX-capable at any RAM configuration, not because of a DIMM population issue like Large v4, but because Sapphire Rapids doesn’t meet Intel’s generational requirement for TDX at all.

For workloads that don’t need confidential computing, this doesn’t matter much in practice. Sapphire Rapids remains a solid, capable platform for general-purpose workloads, and Medium v4 is a genuinely cost-effective entry point into the v4 lineup.

In practice this tier fits the workloads that make up the bulk of most infrastructure footprints even when they don’t get the same attention as the specialized ones: self-hosted CI/CD runners, small to mid-sized web applications, development and staging environments that mirror production without needing production-scale hardware, and internal tooling that runs continuously but doesn’t demand the newest silicon to do its job well. The distinction matters specifically the moment TDX becomes a requirement, at which point Medium v4 is off the table regardless of how the rest of its specs look.

Ice Lake: The Oldest Generation Still Offered

XL v3 and the XL v2 series (v2, v2.1, v2.2) run on 3rd Gen Intel Xeon Scalable, codenamed Ice Lake, older still than Sapphire Rapids. These aren’t TDX-capable at any configuration, and they represent the value end of OpenMetal’s current catalog rather than the performance end.

Ice Lake’s DDR4 memory, rather than the DDR5 used across every newer tier, is the most consequential technical difference here, and it’s worth being clear about which workloads actually notice. Memory-bandwidth-bound workloads, large in-memory datasets, memory-heavy analytics, feel the DDR4-to-DDR5 gap directly. Workloads that are CPU-bound rather than memory-bound, batch processing jobs, static content serving, many general-purpose application backends, don’t push memory bandwidth hard enough for the difference to matter in practice.

For teams running that second category of workload, where cost predictability matters more than having the newest silicon, this generation remains a legitimate, supported option rather than something being phased out from under existing customers.

A Decision Framework by Requirement

• Need TDX confidential computing: stay within Emerald Rapids or newer. XL v4, XXL v4, and XL v5 are TDX-active out of the box. Large v4, Large v5, and Medium v5 can reach TDX eligibility with the right RAM configuration.
• Need the newest platform and highest memory bandwidth: the v5 lineup (Granite Rapids) is the current generation across the board, with the largest cache-per-socket and fastest DDR5-6400 memory available in the catalog.
• Need per-core speed over core count: Large v5’s notably higher base clock on fewer cores than XL v5 fits workloads more sensitive to single-thread latency than raw parallelism.
• Need a cost-effective general-purpose option without a TDX requirement: Medium v4 (Sapphire Rapids) or the Ice Lake-based XL v3/v2 series remain solid, supported choices where confidential computing isn’t a factor.

Getting Started

Current specs and pricing across every generation and tier are on our bare metal pricing page. For a broader look at choosing between hardware generations for your specific deployment, see our guide to choosing between OpenMetal’s five hardware generations, and for the specifics of TDX eligibility across the catalog, see our confidential computing infrastructure page.

FAQ

What’s the difference between OpenMetal’s v4 and v5 server generations?

The v5 lineup runs on Intel’s Xeon 6 platform (Granite Rapids), the current generation, with more cache, faster DDR5-6400 memory, and lower power draw per socket. Most of the v4 lineup runs on 5th Gen Xeon Scalable (Emerald Rapids), one generation older, with Medium v4 specifically running an even older 4th Gen (Sapphire Rapids) chip.

Practically, the cache and memory bandwidth gains translate into fewer round trips to main memory for workloads with large working sets, in-memory databases, analytics jobs, containerized services under heavy concurrent load, which shows up as lower latency and higher throughput on identical core counts rather than a number that only matters on a spec sheet. The lower power draw per socket also means more thermal headroom for sustained turbo clocks under load, and it’s a real, if secondary, factor in total cost of ownership at scale, since power and cooling are ongoing costs that track directly with TDP.

The CPU generation isn’t the only thing that changed. Storage moved too: v5 servers ship with Micron 7500 MAX NVMe, while v4 servers use the previous Micron 7450 MAX generation, an independent upgrade path that affects sequential and random I/O performance regardless of which CPU generation sits alongside it. Memory speed also isn’t uniform across v4: it ranges from DDR5-4400 up to DDR5-5200 depending on tier, while every v5 server standardizes on DDR5-6400, so the memory bandwidth gap between an entry-level v4 server and its v5 equivalent is often larger than the gap between two different v4 tiers.

Is every v4 server the same CPU generation?

No. Large v4, XL v4, and XXL v4 all run Emerald Rapids (5th Gen) chips. Medium v4 is the exception, running a Sapphire Rapids (4th Gen) chip, one generation behind the rest of the v4 tier.

Which OpenMetal servers support Intel TDX?

TDX requires 5th Gen Xeon Scalable (Emerald Rapids) or newer, plus full memory channel population. XL v4, XXL v4, and XL v5 are TDX-active out of the box. Large v4, Large v5, and Medium v5 can reach TDX eligibility with a RAM configuration change. Medium v4 and any Ice Lake-based server (XL v3, XL v2 series) cannot support TDX at any configuration, since they don’t meet the generational requirement.

Why would I choose Large v5 over XL v5 if it has fewer cores?

Large v5 trades core count for clock speed, 16 cores per socket at a notably higher base clock than XL v5’s 32 cores at a lower base clock. This favors workloads where per-core performance and lower latency matter more than maximum parallel throughput, certain database engines, latency-sensitive services, and single-threaded-heavy application logic.