DCT

7:24-cv-00244

Advanced Cluster Systems Inc v. Advanced Micro Devices Inc

Log In
Key Events
Amended Complaint

I. Executive Summary and Procedural Information

  • Parties & Counsel:
  • Case Identification: 7:24-cv-00244, W.D. Tex., 12/08/2025
  • Venue Allegations: Plaintiff alleges venue is proper in the Western District of Texas because Defendant AMD maintains a regular and established place of business in Austin, Texas, including an 870,000-square-foot campus with over 3,500 employees, where it allegedly develops, makes, uses, sells, and tests the accused infringing products.
  • Core Dispute: Plaintiff alleges that Defendant’s high-performance computing processors and accelerators, including its EPYC and Instinct product lines, infringe five patents related to methods for enabling cluster computing.
  • Technical Context: The technology concerns software frameworks that allow applications designed for single processors to utilize the parallel processing power of a multi-node computer cluster, a key capability in the high-performance computing (HPC) and artificial intelligence (AI) markets.
  • Key Procedural History: The filing is a Corrected First Amended Complaint, indicating an original complaint was previously filed and subsequently amended by agreement of the parties. No prior litigation, licensing, or administrative proceedings concerning the patents-in-suit are mentioned in the complaint.

Case Timeline

Date Event
2006-06-13 Earliest Priority Date for all Patents-in-Suit
2019-06-25 U.S. Patent No. 10,333,768 Issued
2019-08-01 AMD EPYC 7002 Series Processors Launch Alleged
2021-09-21 AMD MI250 and MI250X Accelerators Testing Alleged
2021-11-01 AMD MI250X and MI250 Accelerators Offered for Sale Alleged
2022-03-22 AMD MI210 Accelerator Launch Alleged
2023-01-24 U.S. Patent No. 11,563,621 Issued
2023-01-31 U.S. Patent No. 11,570,034 Issued
2023-11-07 U.S. Patent No. 11,811,582 Issued
2023-11-11 AMD MI300 Series Accelerators Testing Alleged
2023-12-06 AMD Announces Availability of MI300X and MI300A Products
2024-06-25 U.S. Patent No. 12,021,679 Issued
2024-10-10 AMD Announces Availability of MI325 Accelerators
2025-06-12 AMD Announces Availability of MI350 Series Accelerators
2025-12-08 Corrected First Amended Complaint Filed

II. Technology and Patent(s)-in-Suit Analysis

U.S. Patent No. 10,333,768 - Cluster Computing

The Invention Explained

  • Problem Addressed: The patent's background describes that many software applications were not designed to leverage the power of computer clusters because they were written to run on a single processing node or used a restrictive "master-slave" architecture that was inefficient for tasks requiring inter-node data sharing during computation (Compl. ¶33-34; ’768 Patent, col. 1:21-32, 1:45-67).
  • The Patented Solution: The invention is a computing framework that enables software originally designed for a single node (a "kernel") to operate within a multi-node cluster (’768 Patent, Abstract). It introduces "cluster node modules" that act as intermediaries; each module communicates with a corresponding single-node kernel and also communicates directly with all other cluster node modules in a peer-to-peer network, allowing the otherwise isolated kernels to function collectively as a high-performance cluster (’768 Patent, col. 6:11-21; Fig. 2).
  • Technical Importance: This approach provided a method to unlock supercomputing-level performance for mainstream software applications without requiring them to be rewritten for parallel execution (Compl. ¶35).

Key Claims at a Glance

  • The complaint asserts independent claim 1 and dependent claims 4, 26, 29, 30, 31, and 33 (Compl. ¶56, ¶58-59).
  • Independent Claim 1 recites a computer cluster with a plurality of nodes, each having a hardware processor, and includes the following essential elements:
    • A mechanism for the nodes to communicate results of mathematical expression evaluations with each other using a peer-to-peer architecture.
    • Each node is configured to access code for a "single-node kernel" capable of evaluating mathematical expressions.
    • A configuration of at least a first, second, and third node, where the first node distributes calls, the second node receives a call and communicates its result to the third node, and the third node uses that result to perform a second evaluation before communicating its own result back to the first node.
    • The first node is configured to return the final result to a user interface.
    • Nodes are configured to accept instructions, communicate them to other nodes via the peer-to-peer mechanism, and pass them to their respective single-node kernels.

U.S. Patent No. 11,563,621 - Cluster Computing

The Invention Explained

  • Problem Addressed: The patent addresses the inability of conventional software programs, which often include a "kernel" designed for a single computer, to benefit from the advantages of multi-node computer clusters (Compl. ¶33; ’621 Patent, col. 1:24-35).
  • The Patented Solution: The patent describes a system where cluster node modules create a peer-to-peer communication layer over a plurality of processing nodes (’621 Patent, Fig. 2). This allows software kernels, which are designed to function independently, to intercommunicate and collaborate on complex tasks, effectively transforming a group of individual computers into a cohesive computing cluster without modifying the kernels themselves (’621 Patent, Abstract; col. 2:8-18).
  • Technical Importance: The invention makes supercomputing performance accessible to mainstream software by providing a framework to parallelize workloads across multiple processors (Compl. ¶35).

Key Claims at a Glance

  • The complaint asserts independent claim 1 and numerous dependent claims (Compl. ¶65, ¶68).
  • Independent Claim 1 recites a computer cluster with a plurality of nodes and includes the following essential elements:
    • A mechanism for the nodes to communicate mathematical evaluation results with each other using a peer-to-peer architecture.
    • A first node with a user interface and a first single-node kernel, configured to interpret user instructions and distribute calls.
    • A second node configured to receive calls from the first, execute a first mathematical evaluation, and communicate the result to a third node.
    • A third node configured to receive the result from the second node, execute a second mathematical evaluation using that result, and communicate its own result back to the first node.
    • The first node is configured to return the result of the second evaluation to the user interface.

U.S. Patent No. 11,570,034 - Cluster Computing

  • Technology Synopsis: This patent, part of the same family, describes a system for adapting single-node software kernels to a multi-node environment. It provides a framework using peer-to-peer communicating "cluster node modules" to enable parallel processing for applications not originally designed for it (Compl. ¶33-35).
  • Asserted Claims: Claims 1, 2, 3, 8, 10, 24, 25, 27, and 28 are asserted, with claim 1 being independent (Compl. ¶74, ¶77).
  • Accused Features: The complaint alleges that AMD's Server, Accelerator, and EPYC products incorporate the patented cluster computing technology (Compl. ¶74).

U.S. Patent No. 11,811,582 - Cluster Computing

  • Technology Synopsis: Continuing the themes of the patent family, this patent discloses a method for enabling peer-to-peer communication among otherwise independent software kernels in a computer cluster. This allows for the distribution and parallel execution of complex computational tasks (Compl. ¶33-35).
  • Asserted Claims: Claims 1, 3, 4, 6, 7, 8, 9, 10, 12, 13, 14, 15, 16, 17, 18, 19, and 20 are asserted, with claim 1 being independent (Compl. ¶83, ¶86).
  • Accused Features: The accused AMD Server, Accelerator, and EPYC products are alleged to infringe by implementing the claimed cluster computing systems and methods (Compl. ¶83).

U.S. Patent No. 12,021,679 - Cluster Computing

  • Technology Synopsis: This patent describes a computer cluster architecture where individual processing nodes, each running a single-node kernel, are networked to communicate directly with one another. This peer-to-peer model is presented as an advancement over master-slave grid computing architectures (Compl. ¶33-35).
  • Asserted Claims: Claims 1, 2, 3, 8, 17, and 18 are asserted, with claim 1 being independent (Compl. ¶92, ¶95).
  • Accused Features: The complaint accuses AMD's Server, Accelerator, and EPYC products of practicing the patented technology (Compl. ¶92).

III. The Accused Instrumentality

Product Identification

  • The accused instrumentalities are collectively identified as the "Accused Products," which encompass three categories: "Accused AMD Server Products," "Accused AMD Accelerator Products," and "Accused EPYC Products" (Compl. ¶53). Specific product lines named include AMD EPYC 7002, 7003, 9004, and 9005 Series processors, and AMD Instinct MI210, MI250, MI250X, MI300X, MI325X, MI355X, MI300A, and MI350A accelerators, as well as server platforms incorporating them (Compl. ¶50-52).

Functionality and Market Context

  • The Accused Products are high-performance processors and accelerators designed for data centers, servers, and supercomputers (Compl. ¶50). The complaint alleges they are used for computationally intensive tasks in the High Performance Compute (HPC) and AI/ML markets (Compl. ¶6). The complaint highlights their use in systems such as the "Frontier" exascale supercomputer, positioning them as critical components in the advanced computing market (Compl. ¶6). The core of the infringement allegation is that these products, when used in multi-node cluster configurations, implement the patented peer-to-peer communication architecture to achieve parallel processing performance (Compl. ¶35, ¶55).
    No probative visual evidence provided in complaint.

IV. Analysis of Infringement Allegations

The complaint references claim-chart exhibits (Exhibits F-J) that demonstrate infringement but does not attach them. The narrative infringement theory is summarized below.

Narrative Infringement Theory Summary

  • The complaint alleges that AMD's Accused Products, when configured and used in a computer cluster, directly infringe the patents-in-suit (Compl. ¶55, ¶64, ¶73, ¶82, ¶91). The theory suggests that the combination of AMD's hardware (e.g., multiple EPYC processors or Instinct accelerators) and enabling software (such as drivers and libraries) creates a system that maps onto the elements of the asserted claims. In this alleged configuration, individual AMD processors or accelerators function as the claimed "nodes," each running a "single-node kernel." The complaint contends these nodes communicate directly with one another in a "peer-to-peer architecture" to collaboratively execute complex computational tasks, mirroring the specific multi-step evaluation and communication sequence recited in the independent claims (Compl. ¶35). For example, the complaint alleges these products are used in a manner that practices the claimed method of a first node distributing a task, a second node executing a part of it and passing an intermediate result to a third node for further processing, before the final result is returned (Compl. ¶56, ¶65, ¶74, ¶83, ¶92).

Identified Points of Contention

  • Scope Questions: A central question may be whether the term "single-node kernel," which the patent specification exemplifies with high-level application software like Mathematica, can be construed to read on the firmware, drivers, or software libraries (e.g., AMD's ROCm platform) that manage the operation of the accused hardware accelerators and processors. The interpretation of "peer-to-peer architecture" may also be contested, raising the question of whether it is limited to the specific software-based "cluster node module" implementation described in the patents or if it can broadly cover hardware-level interconnects like AMD's Infinity Fabric.
  • Technical Questions: The complaint alleges infringement but relies on un-provided exhibits for detailed proof. A technical question for the court will be whether discovery produces evidence that the accused AMD products, in operation, actually perform the specific multi-step, inter-dependent computational sequences required by the claims (e.g., the first node distributing calls, the second node executing and passing a result to the third, and the third executing a subsequent step using that result).

V. Key Claim Terms for Construction

  • The Term: "single-node kernel"

  • Context and Importance: This term is fundamental to the patents' claimed invention, which focuses on adapting software designed for a single node to a cluster environment. The infringement case may depend on whether the software stack operating on each accused AMD processor or accelerator can be defined as a "single-node kernel." Practitioners may focus on this term because its construction could determine whether the patents apply to modern HPC hardware architectures or are limited to the specific software application context described in the specification.

  • Intrinsic Evidence for Interpretation:

    • Evidence for a Broader Interpretation: The specification defines a kernel as an "interpreter that executes instructions" and can "manage at least some hardware resources" (’768 Patent, col. 1:30-34). This language could support an interpretation that covers any software layer on a processing unit that executes instructions.
    • Evidence for a Narrower Interpretation: The specification consistently and repeatedly uses Wolfram Research's Mathematica® software as the primary example of a package containing a "kernel that is designed to communicate with a single node" (’768 Patent, col. 1:41-44). This repeated, specific example may support a narrower construction limited to kernels of high-level, interpreted software applications rather than the integrated firmware and drivers of a GPU or CPU.

  • The Term: "peer-to-peer architecture"

  • Context and Importance: The patents distinguish the invention from prior art "master-slave" grid computing systems by enabling direct, peer-to-peer communication between all nodes. The infringement allegation requires showing that the accused AMD products implement this specific architecture.

  • Intrinsic Evidence for Interpretation:

    • Evidence for a Broader Interpretation: The patent's background contrasts the invention with grid computing where "slave nodes communicated only with the master node" and not directly with other slave nodes (’768 Patent, col. 9:45-53, referring to background section). This could support a broad definition where any architecture allowing direct communication between any two non-master nodes qualifies as "peer-to-peer."
    • Evidence for a Narrower Interpretation: The detailed description explains that the architecture is implemented via "cluster node modules" that provide "MPI calls and/or advanced cluster functions" (’768 Patent, col. 6:11-15). This may support an argument that the term requires the specific software-module-based implementation disclosed, rather than any direct hardware interconnect.

VI. Other Allegations

  • Indirect Infringement: The complaint alleges induced infringement for all five patents. It asserts that AMD knowingly encourages infringement by providing customers, system builders, and partners with "marketing materials, technical specifications, whitepapers, datasheets, user manuals, and development and testing information" that instruct them to build and use the accused products in infringing cluster configurations (Compl. ¶58, ¶67, ¶76, ¶85, ¶94).
  • Willful Infringement: Willfulness is alleged for all five patents. The basis for willfulness is AMD’s alleged knowledge of the patents and its infringement "as of the filing of the Original Complaint," suggesting the allegation is based on continued infringement after the lawsuit was initiated (Compl. ¶57, ¶66, ¶75, ¶84, ¶93).

VII. Analyst’s Conclusion: Key Questions for the Case

  • A core issue will be one of definitional scope: can the term "single-node kernel," rooted in the patents' examples of high-level application software (Mathematica), be construed to cover the integrated firmware, drivers, and software libraries that are fundamental to the operation of modern, high-performance hardware like AMD's GPUs and CPUs?
  • A central architectural question will be one of functional correspondence: does the communication architecture of the accused AMD products, enabled by technologies like Infinity Fabric, perform the specific functions of the patents' "cluster node module" software overlay, or is there a fundamental mismatch in the operational layer and method of achieving peer-to-peer communication?
  • A key evidentiary question will be one of operational proof: beyond high-level architectural similarities, does discovery reveal evidence that the accused products, when operating, execute the specific multi-step, inter-dependent computational sequences recited in the asserted claims, such as the three-node communication and execution loop that forms the basis of the independent claims?