HomeCXL Course
⚡ FREE COURSE · THE MEMORY FABRIC OF THE AI ERA

Compute Express Link (CXL)

The coherent interconnect that lets CPUs, accelerators, and memory pools share memory across a link — breaking the AI memory wall through pooling and disaggregation. The most talked-about data-center technology of the decade, taught in depth and free.

▶ Start with Day 1 — What is CXL?
15 DaysCXL.io / .cache / .memMemory PoolingCXL 1.1 → 4.0AI Memory WallFree

Who this is for: Data-center architects, VLSI and SoC engineers, systems-software developers, and anyone trying to understand why "memory disaggregation" and "CXL" appear in nearly every AI-infrastructure discussion today. No prior CXL experience needed — but a working grasp of PCIe and CPU memory helps.

What you'll learn: The complete CXL story — from the coherent-interconnect fundamentals and the three sub-protocols, through the spec evolution from CXL 1.1 to 4.0, all the way to memory pooling, disaggregation, the software tiering stack, and the real products shipping today. Every technical claim in this course is verified against primary sources before publishing.

The Problem CXL Exists to Solve

A CPU can only physically attach so much memory to its own DIMM slots. Meanwhile, in real data centers, an average of ~6% of memory is "stranded" — allocated to a socket but unused because that socket's cores are already fully scheduled — rising to as much as 25% as VM and container density increases. You're paying for DRAM that no workload can reach. At the same time, large AI models are hard-limited by memory capacity and bandwidth: the "memory wall." CXL attacks both problems at once, by making memory a shared, coherent, expandable resource reachable over a link rather than a fixed quantity bolted to one CPU.

Stranded-memory figures per published data-center memory utilization studies.

CONVENTIONAL — MEMORY CAPTIVE TO EACH CPU CPU A DRAM (full use) CPU B DRAM STRANDED CPU A can't borrow CPU B's idle memory — it's wasted WITH CXL — SHARED, POOLED MEMORY CPU A CPU B CXL CXL Shared Memory Pool Either CPU draws capacity on demand — nothing stranded

What CXL Actually Is

CXL is an open, cache-coherent interconnect standard that runs on top of the PCIe physical layer. It reuses PCIe's electrical signaling and connectors, then layers cache-coherent protocols on top — which is precisely what plain PCIe cannot do. That coherence is the whole point: it lets a device and a host agree on the state of shared memory in hardware, so memory attached to a device can behave, to software, like memory attached to the CPU itself.

The Three Sub-Protocols — The Heart of CXL

📁
CXL.ioConfiguration & I/ODiscovery, enumeration, register access, DMA — functionally similar to PCIe, and the mandatory baseline every CXL device supports.
💾
CXL.cacheDevice caches host memoryLets an accelerator coherently cache the host's memory, so it can work on shared data without constant round-trips.
🧰
CXL.memHost accesses device memoryLets the CPU access memory attached to a device — the basis of memory expanders and pooling.
Three Sub-Protocols, One Shared Physical Link CXL.io config & I/O CXL.cache device caches host mem CXL.mem host accesses device mem CXL multiplexed onto one link PCIe Physical Layer PCIe 5.0 (32 GT/s) → 6.0 (64 GT/s) → 7.0 (128 GT/s)

Which sub-protocols a device speaks defines what kind of CXL device it is:

Device TypeSub-ProtocolsWhat It Is
Type 1CXL.io + CXL.cacheAccelerator with a cache but no host-attached memory (e.g. a SmartNIC or coherent accelerator)
Type 2CXL.io + CXL.cache + CXL.memAccelerator with its own directly-attached memory, kept coherent with the host (e.g. a GPU/accelerator with local memory)
Type 3CXL.io + CXL.memMemory expander — presents DRAM or persistent memory to the CPU over CXL.mem. The workhorse of memory expansion and pooling.

The Spec Evolution — Where the Real Capabilities Live

CXL's power grew dramatically across versions. Each generation rides a newer PCIe physical layer and adds fundamentally new capabilities, not just more speed:

VersionReleasedPCIe Base / SpeedHeadline Capability
CXL 1.0 / 1.12019PCIe 5.0 / 32 GT/sMemory expansion, coherent accelerator caching, device-host memory sharing
CXL 2.02022PCIe 5.0 / 32 GT/sSwitching, hot-plug, memory pooling, and IDE (link encryption & integrity)
CXL 3.02023PCIe 6.0 / 64 GT/s (PAM4)Fabric-level memory sharing (multiple hosts, one region, coherent), multi-level switching, 256B flit
CXL 3.12023PCIe 6.0 / 64 GT/sFabric scale-out via Port-Based Routing (PBR), Trusted-Execution security (TSP) for confidential computing
CXL 4.0Nov 2025PCIe 7.0 / 128 GT/sBandwidth doubling, bundled ports supporting up to 1.5 TB/s connections

Versions, dates, and PCIe base speeds per CXL Consortium specifications and public disclosures.

POOLING — CXL 2.0 Host 1 Host 2 Region → owned by ONE host Reassignable, but one owner at a time SHARING — CXL 3.0 Host 1 Host 2 Same region, BOTH hosts Simultaneous, hardware-coherent access

The single most important distinction to remember: CXL 2.0 introduced pooling — a memory region can be dynamically reassigned between hosts, but only one host owns it at a time. CXL 3.0 introduced sharing — multiple hosts can coherently access the same region simultaneously, with hardware maintaining coherence. That jump from pooling to sharing is what turns CXL from a memory-expansion feature into a genuine memory fabric.

Why This Course Matters Right Now

🧠
The AI Memory WallLarge-model training and inference are memory-bound. CXL pooling and tiering are among the most-discussed answers in AI infrastructure today.
💰
Real TCO ImpactEliminating stranded memory and over-provisioning has a direct, measurable cost benefit — which is why hyperscalers are moving on it.
📈
Zero Free DepthCXL is covered in scattered vendor whitepapers and spec PDFs. A structured, verified, beginner-first course does not exist elsewhere for free.
PHASE 1

Foundations — Coherence, Protocols & Device Types (Days 1–5)

Day 1What is CXL? The memory wall and why coherent interconnect mattersLive Day 2CXL & PCIe — how CXL rides the PCIe physical layerLive
Day 3The Three Sub-Protocols — CXL.io, CXL.cache, CXL.mem in depthSoon
Day 4Device Types — Type 1, Type 2, and Type 3 explainedSoon
Day 5Coherency & Bias Modes — host bias vs device biasSoon
PHASE 2

Spec Evolution & Architecture (Days 6–10)

Day 6CXL 1.1 — memory expansion and the first coherent devicesSoon
Day 7CXL 2.0 — switching, hot-plug, memory pooling & IDE securitySoon
Day 8CXL 3.0 — the fabric, multi-level switching & memory sharingSoon
Day 9CXL 3.1 & 4.0 — PBR fabric scale-out, TSP security, 128 GT/sSoon
Day 10The FLIT Format — 68B vs 256B flits, FEC, and link encryptionSoon
PHASE 3

Deployment & Industry Reality (Days 11–15)

Day 11Memory Pooling & Disaggregation — solving stranded memorySoon
Day 12CXL for AI — breaking the memory wall, tiered memory for LLMsSoon
Day 13The Software Stack — OS/kernel memory tiering & page migrationSoon
Day 14Real Products & Ecosystem — controllers, switches, the CXL landscapeSoon
Day 15🎉 CXL vs Alternatives & the Roadmap AheadSoon

Pair with the Chiplets & UCIe Course for the complete picture of how modern chips and memory fabrics connect — UCIe links chiplets inside the package, CXL links memory and accelerators across the system.