SemiconductorsFEATURE

CXL: The Cache-Coherent Interconnect Standard Built on PCIe

N
NathanTechnology Editor · Technical Lead
Published · Updated
CXL, announced in early 2019, layers coherency protocols on PCIe 5.0, reaching 32 GT/s per direction; CXL 2.0 adds switching for memory pooling.

What is CXL (Compute Express Link), and when was it announced?

CXL Consortium defines Compute Express Link (CXL) as an industry-supported cache-coherent interconnect for processors, memory expansion, and acceleratorsCITE:E1. The interconnect now known as CXL was announced in early 2019CITE:E6.

How does CXL build on PCIe's physical and electrical interfaces?

CXL builds upon the physical and electrical interfaces of PCIe, adding protocols on top that establish coherencyCITE:E2. This means CXL does not replace PCIe's wiring and signaling — it reuses that foundation and layers coherency behavior above itCITE:E2.

What protocol layers does CXL define, and what does each cover?

CXL defines three distinct protocols: CXL.io, CXL.cache, and CXL.memCITE:E5. These three protocols together make up the CXL specification, running over the shared PCIe-based physical linkCITE:E5CITE:E2.

What are the performance specifications for CXL 1.1 and CXL 2.0?

CXL 1.1 and CXL 2.0 both use the PCIe 5.0 physical layerCITE:E3. The following figures summarize the throughput this physical layer delivers:

SpecificationValue
Physical layer usedPCIe 5.0CITE:E3
Data transfer rate32 GT/sCITE:E3
Maximum throughput per directionUp to 64 GB/sCITE:E3
Standard first announcedEarly 2019CITE:E6

How does CXL's cache coherency improve efficiency and lower costs?

CXL Consortium states that CXL technology maintains memory coherency between the CPU memory space and memory on attached devicesCITE:E4. This coherency allows resource sharing that the consortium says delivers higher performance, reduced software stack complexity, and lower overall system costCITE:E4.

What does CXL 2.0's switching support mean for memory pooling?

CXL 2.0 supports switching to enable memory poolingCITE:E7. Switching is the specific mechanism CXL 2.0 adds beyond CXL 1.1 to make pooled memory possibleCITE:E7CITE:E3.

What this means

Across these sources, a consistent technical picture emerges: CXL keeps PCIe 5.0's physical layer and 32 GT/s (up to 64 GB/s per direction) signalingCITE:E3, adds three protocols — CXL.io, CXL.cache, and CXL.mem — on top of itCITE:E5, uses that combination to keep CPU and device memory coherentCITE:E4, and in the 2.0 revision extends this into switching-enabled memory poolingCITE:E7. The standard was announced in early 2019CITE:E6, and the coherency and pooling capabilities described here are what the consortium and technical publishers point to as its performance, complexity, and cost rationaleCITE:E4.

📊 Evidence

FAQ

What is CXL (Compute Express Link), and when was it announced?

CXL Consortium defines Compute Express Link (CXL) as an industry-supported cache-coherent interconnect for processors, memory expansion, and acceleratorsCITE:E1…

How does CXL build on PCIe's physical and electrical interfaces?

CXL builds upon the physical and electrical interfaces of PCIe, adding protocols on top that establish coherencyCITE:E2.

What protocol layers does CXL define, and what does each cover?

CXL defines three distinct protocols: CXL.io, CXL.cache, and CXL.memCITE:E5.

What are the performance specifications for CXL 1.1 and CXL 2.0?

CXL 1.1 and CXL 2.0 both use the PCIe 5.0 physical layerCITE:E3. The following figures summarize the throughput this physical layer delivers:

📎 Sources

  1. computeexpresslink.org
  2. rambus.com
  3. synopsys.com

Related data

Author's TakeNathan

The fact that CXL keeps the PCIe 5.0 physical layer while adding CXL.io, CXL.cache, and CXL.mem shows the consortium chose compatibility over a clean-sheet redesign, which is why the interconnect can reach 32 GT/s (up to 64 GB/s per direction) without new host silicon. The more consequential detail is that CXL 2.0 support for switching enables memory pooling, since pooling is what lets multiple hosts share disaggregated memory rather than each host being locked to its own attached capacity. That is the indicator worth tracking: how many CXL 2.0 deployments actually turn on switching-based pooling, versus how many merely use CXL 1.1 point-to-point memory expansion.

N
NathanTechnology Editor · Technical Lead

Related

BRIEF

CNA Launches Taiwan's First News MCP Tool, AskCNA, Priced at NT$200 a Month

Central News Agency (中央社) launched CNA MCP on August 31, 2026, Taiwan's first news tool built on Anthropic's Model Context Protocol (released November 2024), letting AI agents such as Claude, ChatGPT, and Grok retrieve and cite its archives in real time. The tool integrates nearly 5 million newswire stories, 3.5 million photos, and open data from about 150 government agencies, priced at NT$200 a month with an early-bird bonus-quota plan, and received funding from Google Taiwan's nDX Digital Innovation Grant Program.

EffectStory 編輯部 ·
BRIEF

Sony Music and Warner Chappell Sue Anthropic Over Alleged 'Brazen Campaign' of Copyright Theft

Sony Music Publishing and Warner Chappell, joined by other music publishers, sued Anthropic and co-founders Dario Amodei and Benjamin Mann in the U.S. District Court for the Northern District of California, alleging illegal torrenting, scraping, and downloading of copyrighted lyrics and sheet music. The publishers seek up to $150,000 per work and $25,000 per instance of stripped copyright data, a total that could reach several billion dollars. The filing follows Anthropic's earlier $1.5 billion settlement in the Bartz case.

EffectStory 編輯部 ·
BRIEF

Why Anthropic Turned to Nscale and Lambda for $45B and $35B GPU Compute Deals

Anthropic has assembled compute capacity across at least four NVIDIA-linked providers: a $35 billion contract with Lambda tied to a Hut 8-built Texas data center, a $45 billion, six-year deal with Nscale for a West Virginia campus running NVIDIA Vera Rubin systems, a $10 billion contract with startup Volta in Norway, and a reported (unconfirmed) tenancy at Riot Platforms' Rockdale, Texas site. NVIDIA sits inside nearly every arrangement — as investor, lessor, or chip supplier.

EffectStory 編輯部 ·