Quick Verdict
A strong mid-range workstation CPU with excellent IPC, modern I/O, and AMX-based AI acceleration, best suited for professionals who need more than desktop cores but don’t require 60+ core monsters.
Overview
Launch
2026
Status
LaunchedGeneration
Xeon 600 (Granite Rapids-WS)
Market
Workstation / Server
The Intel Xeon 638 is a 16-core, 32-thread workstation and server processor in Intel’s Xeon 600 Granite Rapids-WS family, built on the Intel 3 process with Redwood Cove P-cores, 72 MB of L3 cache, and 80 PCIe 5.0 lanes. It targets professional workloads like 3D rendering, simulation, and AI inference where high single-thread clocks and decent multi-thread density matter more than maximum core count.
Intel’s Xeon 638 sits in the middle of the new Xeon 600 Granite Rapids-WS workstation stack. It combines 16 Redwood Cove P-cores, 32 threads, a 3.2 GHz base and 4.
8 GHz turbo, 72 MB of shared L3, and 80 PCIe 5.0 lanes. Compared to previous Xeon W parts, it delivers substantially better single-thread performance, much faster DDR5, and native FP16 AMX acceleration for AI workloads.
It’s best for professional 3D, simulation, and AI tasks where you need more than a desktop CPU but don’t need the 60–86 core flagships.
Specifications
Performance
Expected to perform very well in professional productivity apps (rendering, simulation, data analysis) thanks to 16 high-IPC cores and fast DDR5/PCIe 5.0, but no official productivity scores are available.
Suitable for small-to-medium virtualization hosts with its 32 threads, 2 TB memory support and VT-x/VT-d, but no standardized virtualization scores exist for this SKU.
Not targeted at gaming; no official gaming benchmarks. Single-thread performance is strong, but workstation-focused GPUs and platform tuning matter more than CPU here.
180 W TDP is moderate for a 16-core workstation CPU on Intel 3; efficiency is better than older 14 nm Xeons but not as strong as low-power EPYC or mobile parts.
- •No integrated graphics
- •Not validated for gaming workloads by Intel
- •Gaming performance will depend heavily on GPU and platform tuning
- •AMX with FP16/BF16/INT8 acceleration
- •FP16 native support reduces overhead vs prior Xeon generations
- •Still limited compared to dedicated AI accelerators or high-end GPUs for large models
Architecture
Intel 3
Process Node
Granite Rapids-WS
Codename
16C / 32T
Core Config
72 MB
L3 Cache
180 W
TDP
Architecture Overview
Granite Rapids-WS is Intel’s workstation-oriented implementation of the Granite Rapids server architecture, using Redwood Cove P-cores on the Intel 3 process. It shifts from the hybrid P/E-core approach of some client chips to a homogeneous P-core design optimized for sustained professional workloads.
CPU Design
The Xeon 638 has 16 Redwood Cove performance cores with Hyper-Threading, giving 32 threads. Each core has 112 KB L1 cache (64 KB instruction, 48 KB data) and 2 MB L2 cache, with a 72 MB shared L3 slice across the die, totaling 104 MB cache.
Memory Subsystem
It integrates a quad-channel DDR5 memory controller supporting DDR5-6400 RDIMMs and MRDIMMs (up to 8000 MT/s on some Xeon 600 SKUs), with ECC and up to 2 TB capacity on this 4-channel model.
PCIe & I/O
80 PCIe 5.0 lanes are routed directly from the CPU, enabling multi-GPU setups and fast NVMe or accelerator attachments without the lane restrictions of mainstream desktop platforms.
Overclocking
Intel offers unlocked X-series SKUs in the Xeon 600 family; the Xeon 638 is listed by third parties as having an unlocked multiplier, allowing enthusiast-style tuning, though this is officially at the user’s risk and with warranty implications.
- Move from Intel 7 to Intel 3 process technology
- Higher DDR5 data rates (6400 MT/s RDIMM, up to 8000 MT/s MRDIMM vs 4800 MT/s)
- Larger L2 and L3 caches per core
- AMX FP16 acceleration and AVX-512-FP16 support
- Simplified Xeon 600 branding instead of split W-2500/W-3500 stacks
Key Highlights
- 16 high-IPC Redwood Cove cores with 4.8 GHz turbo
- 80 PCIe 5.0 lanes for multi-GPU and fast storage
- Quad-channel DDR5-6400 with ECC and MRDIMM support
- AMX with FP16 for improved AI inference performance
- Unlocked multiplier for overclocking on supported platforms
- Modern Intel 3 process improves performance per watt vs older Xeons
- Only four memory channels versus eight on higher Xeon 600 SKUs
- 180 W TDP and 216 W max turbo can stress compact cooling solutions
- No integrated graphics, requiring a discrete GPU
- Higher platform cost than mainstream desktop CPUs
- Workstation pricing may not justify upgrades for users with existing Sapphire Rapids-WS systems
History
The Xeon 638 belongs to Intel’s Xeon 600 Granite Rapids-WS generation, launched in early 2026 as a unified workstation brand replacing the earlier Xeon W-2500 and W-3500 Sapphire Rapids-WS families. This generation brings the Redwood Cove P-core architecture from Meteor Lake to the workstation space on Intel’s 3 node, with substantially larger caches, DDR5-6400 and MRDIMM support, and up to 128 PCIe 5.0 lanes.
Intel positions Xeon 600 as a major architectural shift rather than a simple refresh, targeting professionals who need more cores and memory bandwidth than desktop platforms provide, but without stepping up to full datacenter Xeons. The 638 fills the mid-range, 4-channel, 80-lane slot, offering a balance of clocks, core count, and I/O that suits many 3D, simulation, and AI workloads.
Improvements over Previous Generation
- Move from Intel 7 to Intel 3 process technology
- Higher DDR5 data rates (6400 MT/s RDIMM, up to 8000 MT/s MRDIMM vs 4800 MT/s)
- Larger L2 and L3 caches per core
- AMX FP16 acceleration and AVX-512-FP16 support
- Simplified Xeon 600 branding instead of split W-2500/W-3500 stacks
Alternatives & Competitors
Should You Buy It?
Recommended for the right buyer
Professional workstation or small server needing 16–32 threads, strong per-core performance, and lots of PCIe 5.0 connectivity, but not extreme core counts or eight-channel memory.
Avoid if…
- You need eight memory channels or 128 PCIe lanes (consider Xeon 600 8-channel SKUs)
- You want maximum multi-threaded performance for rendering or simulation (consider 24–86 core Xeon 600 or Threadripper PRO)
- Your workload is purely gaming or light desktop use (a Core Ultra or Ryzen is better value)
Use Cases
Interesting Facts
Xeon 638 is part of Intel’s first unified Xeon 600 workstation brand, which replaces the previous Xeon W-2500 and W-3500 split.
It uses the same Redwood Cove P-core architecture first introduced in Meteor Lake mobile processors, but on an Intel 3 server-optimized node.
Despite being a “638” model, it has 16 cores and 72 MB L3, matching the cache density of higher-core Granite Rapids-WS dies.
The Xeon 600 platform supports MRDIMM memory up to 8000 MT/s, though the 4-channel mainstream SKUs like 638 cap official RDIMM speeds at 6400 MT/s.
Intel and partners have already set multiple overclocking world records using Xeon 600 X-series CPUs on W890 motherboards.
This CPU targets the “mainstream” 4-channel segment of the Xeon 600 family, with 80 PCIe lanes; 8-channel “expert” SKUs offer 128 lanes.
Granite Rapids-WS inherits CXL 2.0 support from the server Granite Rapids line, though CXL usage is still emerging in workstations.
The Xeon 638’s 4.8 GHz max turbo matches the highest boost clocks across the entire Xeon 600 workstation lineup.
Its 180 W TDP sits between the lower-power 12-core 634 (150 W) and the 20-core 656 (210 W).
Intel officially positions Xeon 600 as a “full architecture overhaul” rather than a refresh, with major changes to cache, memory, and I/O.
People Also Ask
Is the Intel Xeon 638 good for gaming?
It can run games if paired with a discrete GPU, but it’s not optimized or validated for gaming. Gamers are better served by mainstream desktop CPUs like Core Ultra or Ryzen, which offer similar or better gaming performance at lower cost.
How many PCIe lanes does the Xeon 638 have?
80 PCIe 5.0 lanes directly from the CPU, intended for multi-GPU, NVMe and accelerator cards in workstations and servers.
Does the Xeon 638 support DDR5?
Yes, it supports quad-channel DDR5 up to 6400 MT/s RDIMMs and MRDIMMs, with ECC and up to 2 TB of memory.
Can you overclock the Intel Xeon 638?
The Xeon 638 is listed by third-party databases as having an unlocked multiplier, and the Xeon 600 X-series family is officially unlocked. Overclocking is supported on W890 motherboards but is at the user’s own risk and may affect warranty.
What is the difference between Xeon 638 and Xeon 654?
Xeon 654 has 18 cores, a 200 W TDP, and is an 8-channel DDR5 SKU with 128 PCIe lanes, while Xeon 638 is a 4-channel, 80-lane, 180 W part. The 654 is better for memory-bound workloads; the 638 is cheaper and sufficient if you don’t need eight channels.
Is the Xeon 638 better than a Threadripper PRO for workstations?
It depends on the workload. Threadripper PRO 7000/9000 series offers more cores and memory bandwidth on some models, while Xeon 638 offers strong per-core performance, AMX FP16 acceleration, and a modern platform. For heavily threaded rendering and simulation, high-core Threadripper often wins; for mixed or latency-sensitive workloads, Xeon 638 can be competitive.
Does the Xeon 638 have integrated graphics?
No, it requires a discrete GPU.
What socket does the Xeon 638 use?
FCLGA4710, part of the W890 workstation platform.
How much cache does the Xeon 638 have?
72 MB of shared L3 cache, plus 2 MB L2 per core (32 MB total L2), for 104 MB total cache.
Is the Xeon 638 suitable for virtualization?
Yes. With 32 threads, support for up to 2 TB of ECC memory, and Intel VT-x/VT-d, it’s a capable platform for small-to-medium virtualization hosts.
Frequently Asked Questions
What platform does the Intel Xeon 638 use?
It uses the Intel W890 chipset and FCLGA4710 socket, forming the new Xeon 600 workstation platform.
Is the Xeon 638 multi-socket capable?
No, Intel lists it as 1S-only (single socket).
What memory speeds does the Xeon 638 support?
Official DDR5-6400 RDIMMs and MRDIMMs (depending on SKU and platform configuration), with ECC.
Does the Xeon 638 support Intel vPro?
Yes, it is eligible for Intel vPro Enterprise, offering manageability and security features for business environments.
Can I use the Xeon 638 in a desktop PC?
Technically possible on a compatible W890 workstation motherboard, but it’s optimized for professional workloads and server-style cooling; a desktop Core Ultra or Ryzen is usually more practical for general use.
What cooling is recommended for the Xeon 638?
A high-end tower air cooler or 280–360 mm AIO liquid cooler is recommended, given the 180 W TDP and 216 W max turbo power.
Does the Xeon 638 support AVX-512?
Yes, it supports AVX-512 and AVX-512-FP16 instructions, which is important for some HPC and AI workloads.
How does the Xeon 638 compare to older Xeon W-3500 CPUs?
It offers higher IPC, faster DDR5, more PCIe 5.0 lanes, and AMX FP16 acceleration, but you’ll need a new W890 motherboard; the performance per watt and feature set are significantly improved.
Is the Xeon 638 a good choice for AI inference?
It’s a solid CPU-based AI inference platform thanks to AMX FP16/BF16/INT8 acceleration, but for large models, a dedicated AI accelerator or high-end GPU will still outperform it.
What is the maximum turbo power of the Xeon 638?
Intel lists a maximum turbo power of 216 W, which the motherboard and cooler should be designed to handle.