How to Read Intel Xeon and AMD EPYC Model Numbers
Intel and AMD server CPU names look like alphabet soup until you know the pattern. They are actually structured codes: every digit and suffix carries a specific meaning. Once you know the decode, you can read any Xeon or EPYC part number in under 10 seconds.
The 30-second version
Every segment of the model number has a job. The green box — the generation digit — is the one that matters most.
Intel Xeon Gold 6338AMD EPYC 9654★ Same generation digit = same platform era. Two chips with the same generation digit compete in the same market, use the same socket, and support the same memory. That’s the comparison unit that matters for procurement.
Intel Xeon decoded
The tier system
Xeon Scalable has four named tiers. The tier appears in the product name and is also reflected in the first digit of the 4-digit model number.
| Tier | Max cores | Socket support | Typical buyer |
|---|---|---|---|
| Bronze | 10 | Single-socket | Entry workloads, cost-constrained deployments |
| Silver | 24 | Single or dual-socket | Mid-range general-purpose servers |
| Gold 5xxx | 28 | Dual-socket | Standard enterprise dual-socket builds |
| Gold 6xxx | 28 | Dual-socket, full memory bandwidth | Memory-intensive workloads — database, analytics |
| Platinum | 40 | Up to 8-socket | Mission-critical, scale-up NUMA systems |
The 4-digit model number
The model number (e.g. 6338) is a 4-digit code that follows the tier name. Each digit position carries a specific meaning:
- First digit — encodes the tier (5 = Gold 5xxx, 6 = Gold 6xxx, 8 = Platinum, etc.)
- Second digit — the generation. This is the most important digit when comparing chips across vendors or builds.
- Last two digits — position within the generation. Higher generally means more cores, more cache, or higher base clock.
Example: 6338 = Gold 6xxx tier → 3rd generation (Ice Lake) → position 38 in that generation.
Generation decoder
The second digit of the model number tells you the platform generation — socket, memory type, and the competitive era the chip belongs to.
| 2nd digit | Codename | Year | Socket | Memory |
|---|---|---|---|---|
| 1 | Skylake-SP | 2017 | LGA3647 | DDR4-2666 |
| 2 | Cascade Lake | 2019 | LGA3647 | DDR4-2933 |
| 3 | Ice Lake | 2021 | LGA4189 | DDR4-3200 |
| 4 | Sapphire Rapids | 2023 | LGA4677 | DDR5-4800 |
| 5 | Emerald Rapids | 2024 | LGA4677 | DDR5-5600 |
| 6 | Granite Rapids | 2024–25 | LGA4710 / LGA7529 | DDR5-6400 |
Suffix letters
An optional suffix after the model number indicates a specialized variant. Most chips on the secondary market are unsuffixed.
| Suffix | Meaning |
|---|---|
| N | Network-optimized — extra PCIe lanes, QuickAssist acceleration |
| T | Thermal / long-life — extended temperature range, industrial deployments |
| L | Large memory — expanded NUMA support beyond standard configuration |
| M | Media — integrated media transcode or QuickAssist |
| Y | Speed Select — configurable TDP and frequency bins |
| P | Cloud / IaaS — ISA-locked for cloud platform use |
| V | VM density — larger L3 cache, NUMA-optimized for virtualization |
| H | HBM — High Bandwidth Memory variant |
| + | Refresh — minor die or firmware revision within the same generation |
Intel dropped the tier system for Xeon 6. There is no Bronze, Silver, Gold, or Platinum. Instead, Xeon 6 uses core architecture as the primary differentiator: P-core (Performance-core, high single-thread throughput) vs E-core (Efficient-core, high density and throughput at lower TDP). The 4-digit model numbering continues, but there is no tier name in the product string. If you see “Xeon 6” without a tier, look for the P or E suffix.
AMD EPYC decoded
The series system
The first digit of the model number identifies the EPYC series — which maps to a specific socket and market segment.
| Prefix | Series | Socket | Market |
|---|---|---|---|
| 7xxx | Naples, Rome, Milan, Milan-X | SP3 | Mainstream datacenter (2017–2022) |
| 9xxx | Genoa, Bergamo, Turin | SP5 | Current-generation datacenter (2022–present) |
| 8xxx | Siena | SP6 | Edge, telco, dense cloud |
| 4xxx | Ryzen-derived | AM5 | Budget / single-socket workstation class |
The 4-digit model number
Each digit of the EPYC model number encodes a specific dimension of the chip:
- First digit — series (7 = SP3 mainstream, 9 = SP5 current-gen)
- Second digit — generation within the series. The key digit for cross-chip comparison.
- Third digit — performance tier within the generation. Higher = more cores.
- Fourth digit — position within the performance tier.
Example: 9654 = 9xxx series (SP5) → 6th generation (Genoa) → perf tier 5 → position 4.
Generation decoder
The second digit (and sometimes the third) identifies the AMD generation. The table uses the prefix pattern since AMD’s generation encoding varies slightly between 7xxx and 9xxx series.
| Model prefix | Codename | Year | Socket | Memory |
|---|---|---|---|---|
| 7xx1 | Naples | 2017 | SP3 | DDR4-2666 |
| 7xx2 | Rome | 2019 | SP3 | DDR4-3200 |
| 7xx3 | Milan / Milan-X | 2021 | SP3 | DDR4-3200 |
| 9xx4 | Genoa | 2022 | SP5 | DDR5-4800 |
| 97x4 | Bergamo | 2023 | SP5 | DDR5-4800 |
| 9xx5 | Turin | 2024 | SP5 | DDR5-6000 |
| 99x5 | Turin Dense | 2025 | SP5 | DDR5-6400 |
x = wildcard digit; prefix shows the pattern, not an exact match.
Suffix letters
EPYC suffixes are less common than Xeon suffixes, but each one meaningfully changes the chip’s trade-off profile.
| Suffix | Meaning |
|---|---|
| F | High frequency — boosted base and boost clocks, fewer cores |
| P | Single-socket only — reduced inter-socket logic, lower cost |
| X | 3D V-Cache — stacked L3 cache for latency-sensitive workloads (gaming, EDA, CFD) |
| H | HPC-optimized — high core count, tuned for parallel throughput |
How to compare chips
When are two chips directly comparable?
Two chips are directly comparable when they share the same generation digit and the same competitive market: same socket generation, same memory type, same platform era. A Xeon Gold 6338 (Ice Lake, gen 3) competes with an EPYC 7xx3 (Milan, gen 3) — both are 2021 platform chips, both DDR4, both targeting the same workloads. Comparing them to a Sapphire Rapids or Genoa chip is an apples-to-oranges comparison.
Cross-generation caveats
IPC gains between successive generations typically run 20–40%. A newer chip with a lower model number can outperform an older chip with a higher model number. Clock speed and core count comparisons across generations are misleading. Always compare within the same generation first, then evaluate generation-to-generation upgrades separately.
Intel vs AMD rough equivalents
These pairings are directional — they competed in the same market window, not necessarily at the same price point or workload.
| Intel Xeon | AMD EPYC | Era |
|---|---|---|
| Ice Lake (3xxx — gen 3) | Milan / Milan-X (7xx3) | 2021–22 · LGA4189 / SP3 · DDR4 |
| Sapphire Rapids (4xxx — gen 4) | Genoa (9xx4) | 2022–23 · LGA4677 / SP5 · DDR5 |
| Emerald Rapids (5xxx — gen 5) | Genoa-X / early Turin (9xx4–5) | 2024 · LGA4677 / SP5 · DDR5 |
| Granite Rapids (6xxx — gen 6) | Turin Dense (99x5) | 2024–25 · LGA4710 / SP5 · DDR5 |
Older generations (Skylake / Naples, Cascade Lake / Rome) are not shown — they are retired from active procurement in most environments.
How this connects to pricing
Chips from the same generation compete for the same workloads, so their secondary market prices move together. A Gen 4 Xeon and a Genoa EPYC trade in the same demand pool. Chips from different generations are in different markets entirely — a Gen 3 chip does not compete with Gen 5, and its price is driven by different buyers and different inventory dynamics.
When you see a price signal on MarketSignalIndex, the generation digit tells you which demand pool that price belongs to. A wide spread within a generation usually means mixed core counts trading together. A wide spread across generations means you’re looking at different markets, not volatility.
Browse live pricing for Intel Xeon and AMD EPYC processors