DCT

7:24-cv-00244

Advanced Cluster Systems Inc v. Advanced Micro Devices Inc

Log In

I. Executive Summary and Procedural Information

  • Parties & Counsel:
  • Case Identification: 7:24-cv-00244, W.D. Tex., 09/26/2024
  • Venue Allegations: Plaintiff alleges venue is proper in the Western District of Texas because Defendant maintains a regular and established place of business in Austin, Texas, including an 870,000 square foot campus, and has committed acts of infringement within the district.
  • Core Dispute: Plaintiff alleges that Defendant’s high-performance computing accelerators and server products infringe five patents related to enabling software designed for a single computer to operate efficiently across a multi-node computer cluster.
  • Technical Context: The technology concerns parallel computing architectures that allow individual processing nodes in a cluster to communicate directly with one another, a method crucial for scaling performance in modern artificial intelligence and supercomputing tasks.
  • Key Procedural History: The complaint notes that Plaintiff began commercializing related technology as early as 2008. No prior litigation, licensing history, or post-grant proceedings involving 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 Issues
2021-09-21 Alleged Making/Using/Testing of Accused MI250/MI250X Accelerators
2021-11-01 Alleged Offer for Sale of Accused MI250/MI250X Accelerators
2022-03-22 Alleged Making/Using/Testing of Accused MI210 Accelerator
2023-01-24 U.S. Patent No. 11,563,621 Issues
2023-01-31 U.S. Patent No. 11,570,034 Issues
2023-11-07 U.S. Patent No. 11,811,582 Issues
2023-11-11 Alleged Testing of Accused MI300 Series Accelerators
2023-12-06 Announced Availability of Accused MI300 Series Accelerators
2024-06-25 U.S. Patent No. 12,021,679 Issues
2024-09-26 Complaint Filed

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

U.S. Patent No. 10,333,768: Cluster Computing (issued Jun. 25, 2019)

The Invention Explained

  • Problem Addressed: The patent addresses the problem that many computer application programs, particularly those using a computational "kernel" like Wolfram Mathematica, are designed to run on only a single processor or "node" (Compl. ¶33; ’768 Patent, col. 1:21-32). This prevents them from leveraging the power of computer clusters. Prior art solutions like "grid computing" relied on a rigid master-slave architecture where slave nodes could only communicate with the master, not with each other, limiting efficiency for complex, interdependent tasks (Compl. ¶34; ’768 Patent, col. 1:45-64).
  • The Patented Solution: The invention provides a software layer, described as a "cluster node module," that runs on each node of a cluster alongside the application's original single-node kernel (’768 Patent, col. 5:35-40). These modules intercept communications and manage a peer-to-peer network, allowing all the single-node kernels to communicate directly with one another as if they were part of a unified, multi-node application (’768 Patent, Fig. 2; col. 6:11-25). This architecture makes supercomputing performance accessible to software not originally designed for it (’768 Patent, col. 2:10-13).
  • Technical Importance: This approach enables legacy or single-node-architected software to be parallelized without rewriting the core application, by creating a peer-to-peer communication fabric among otherwise isolated program instances (Compl. ¶35).

Key Claims at a Glance

  • The complaint asserts independent claims 1 and 35 (Compl. ¶53).
  • Independent Claim 1 requires:
    • A computer cluster with a plurality of nodes, each with a hardware processor.
    • An initialization process for the cluster.
    • Each node having access to a single-node kernel for evaluating mathematical expressions.
    • A mechanism for peer-to-peer communication of results between nodes.
    • A specific three-node configuration (first, second, third) where the second node executes a task and sends its result to the third node, which uses that result for a subsequent execution before communicating back to the first.
  • The complaint reserves the right to assert additional claims (Compl. ¶59).

U.S. Patent No. 11,563,621: Cluster Computing (issued Jan. 24, 2023)

The Invention Explained

  • Problem Addressed: The ’621 Patent, from the same family as the ’768 Patent, addresses the identical technical problem: enabling single-node software kernels to benefit from multi-node cluster architectures by overcoming the limitations of master-slave communication models (Compl. ¶¶ 33-34; ’621 Patent, col. 1:21-47).
  • The Patented Solution: The solution is materially the same as that of the ’768 Patent. It discloses a system of "cluster node modules" that create a peer-to-peer communication network, allowing otherwise isolated, single-node kernels to exchange data directly (’621 Patent, Abstract; Fig. 2). A key feature highlighted in the claims is a data structure within a cluster node module for storing messages received from other nodes, which facilitates the asynchronous, peer-to-peer data exchange (’621 Patent, col. 9:55-62).
  • Technical Importance: The invention provides a method for retrofitting high-performance, peer-to-peer parallelism onto software applications that were originally built for serial, single-processor execution (Compl. ¶35).

Key Claims at a Glance

  • The complaint asserts independent claim 1 (Compl. ¶62).

  • Independent Claim 1 requires:

    • A computer cluster with first, second, and third processors.
    • Associated first, second, and third kernels configured to translate commands for execution on their respective processors.
    • Associated first, second, and third "cluster node modules" that communicate with each other and their respective kernels.
    • The first cluster node module comprising a data structure for storing messages originating from the second and third cluster node modules.

  • The complaint reserves the right to assert additional claims (Compl. ¶68).
    Multi-Patent Capsules for Additional Patents-in-Suit

  • U.S. Patent No. 11,570,034: Cluster Computing (issued Jan. 31, 2023)

    • Technology Synopsis: Continuing the same technical theme, the ’034 Patent describes a system for adding cluster functionality to single-node applications. It details a peer-to-peer architecture where "cluster node modules" enable direct communication between computational kernels, bypassing the limitations of traditional master-slave models (Compl. ¶¶ 33-35).
    • Asserted Claims: Independent claims 1 and 30 (Compl. ¶71).
    • Accused Features: The complaint alleges that AMD's Instinct Accelerators and associated server products, when operating in multi-node configurations, embody the claimed peer-to-peer cluster computing methods (Compl. ¶¶ 44-50, 70).

  • U.S. Patent No. 11,811,582: Cluster Computing (issued Nov. 7, 2023)

    • Technology Synopsis: This patent also relates to methods for enabling single-node software kernels to operate in a parallel, multi-node environment. The invention centers on "cluster node modules" that facilitate a peer-to-peer communication network, allowing for efficient, direct data exchange between nodes performing interdependent computational tasks (Compl. ¶¶ 33-35).
    • Asserted Claims: Independent claim 1 (Compl. ¶80).
    • Accused Features: The complaint alleges that the architecture of AMD's Instinct accelerators, which are designed for large-scale, multi-node deployment in servers for AI and HPC tasks, infringes the claimed cluster communication system (Compl. ¶¶ 44-50, 79).

  • U.S. Patent No. 12,021,679: Cluster Computing (issued Jun. 25, 2024)

    • Technology Synopsis: As the most recently issued patent in the asserted family, the ’679 Patent continues to claim systems for parallelizing single-node applications. The core technology involves "cluster node modules" that create a peer-to-peer communication fabric, enabling kernels on different nodes to collaborate directly on complex computations rather than communicating through a central master node (Compl. ¶¶ 33-35).
    • Asserted Claims: Independent claim 1 (Compl. ¶89).
    • Accused Features: Plaintiff alleges that AMD’s server and accelerator products, particularly the MI200 and MI300 series, implement the claimed peer-to-peer cluster computing architecture to achieve high performance (Compl. ¶¶ 44-50, 88).

III. The Accused Instrumentality

Product Identification

  • The accused instrumentalities are "the Accused AMD Server Products and Accused AMD Accelerator Products" (Compl. ¶50). This specifically includes AMD’s Instinct™ MI210, MI250, MI250X, MI300X, and MI300A accelerators, as well as server and workstation products that incorporate them (Compl. ¶49).

Functionality and Market Context

  • The accused products are high-performance data center GPUs (accelerators) designed for artificial intelligence (AI) and high-performance computing (HPC) workloads (Compl. ¶6). These accelerators are typically deployed in multi-node server clusters where numerous accelerators work in parallel to solve large computational problems (Compl. ¶¶ 13, 48).
  • The complaint alleges that these products are designed to be interconnected in a manner that allows for direct communication between accelerator nodes to perform tasks, a key functionality for training large AI models and running complex scientific simulations (Compl. ¶¶ 35, 42). The complaint alleges AMD began making, using, and testing products with the infringing technology at least as early as September 2021 with the MI250 series accelerators (Compl. ¶44).

IV. Analysis of Infringement Allegations

No probative visual evidence provided in complaint.

The complaint references claim chart exhibits that were not provided. The following summaries are based on the narrative infringement theory presented in the complaint and the public patent documents.

’768 Patent Infringement Allegations

Claim Element (from Independent Claim 1) Alleged Infringing Functionality Complaint Citation Patent Citation
A computer cluster comprising: a plurality of nodes, wherein each of the plurality of nodes comprises a hardware processor... The accused instrumentality comprises servers and workstations containing multiple AMD Instinct accelerators, which function as nodes in a computer cluster. ¶48 col. 9:39-41
a mechanism for the nodes to communicate results of mathematical expression evaluation with each other using a peer-to-peer architecture... The accused accelerators, when deployed in a cluster, are allegedly configured to communicate directly with each other to exchange data and intermediate results for parallel processing tasks. ¶35 col. 12:31-41
wherein the plurality of nodes comprises: a first node... a second node... a third node... a second node is configured to receive calls from the first node, execute at least a first mathematical expression evaluation, and communicate a result... to a third node... When used for HPC or AI tasks, the accused accelerators allegedly operate in a peer-to-peer manner where one accelerator (second node) performs a computation and passes its result directly to another accelerator (third node) for a subsequent, dependent computation. ¶¶ 35, 48 col. 9:35-62
  • Identified Points of Contention:
    • Scope Questions: A central question may be whether the term "single-node kernel," which is described in the patent in the context of a software application like Mathematica, can be construed to read on the firmware, drivers, or hardware architecture of the accused AMD accelerators.
    • Technical Questions: The infringement allegation appears to depend on how the accused accelerators operate when clustered. A key factual question will be whether the communication between AMD accelerators in a server is functionally equivalent to the claimed "peer-to-peer architecture" managed by "cluster node modules," or if it operates on a fundamentally different technical principle.

’621 Patent Infringement Allegations

Claim Element (from Independent Claim 1) Alleged Infringing Functionality Complaint Citation Patent Citation
A computer cluster comprising: a first processor; a second processor; and a third processor. The accused servers contain multiple processors and/or accelerators, which function as the claimed first, second, and third processors in a cluster. ¶48 col. 7:1-9
a first kernel... a first cluster node module... a second kernel... a second cluster node module... The complaint alleges that the accused products' hardware, firmware, and software drivers collectively perform the function of the claimed kernels and cluster node modules, enabling communication between processors. ¶¶ 35, 49-50 col. 9:22-62
wherein the first cluster node module comprises a data structure in which messages originating from the second and third cluster node modules are stored. The software or firmware controlling the primary accused accelerator (first node) allegedly includes buffers or queues to store data packets received directly from other accelerators (second and third nodes) in the cluster. ¶35 col. 11:59-67
  • Identified Points of Contention:
    • Scope Questions: A likely point of dispute will be the construction of "kernel" and "cluster node module." The defense may argue these terms are limited to the specific software-based embodiments described in the specification (e.g., a Mathematica kernel and a software module), while the plaintiff may argue for a broader functional interpretation that covers the hardware and firmware of the accused accelerators.
    • Technical Questions: The complaint does not provide sufficient detail for analysis of the specific data structures used in the accused products. Discovery will be needed to determine whether the accused products contain a "data structure" that performs the storage function for peer-to-peer messages in the manner required by the claim.

V. Key Claim Terms for Construction

  • The Term: "cluster node module"

    • Context and Importance: This term appears in the independent claims of multiple asserted patents and is the central component of the inventive solution. Its construction will be critical, as the infringement case depends on mapping this term, described in the patent as a software entity, onto the functionality of AMD's hardware, firmware, and driver stack.
    • Intrinsic Evidence for Interpretation:
      • Evidence for a Broader Interpretation: The claims do not explicitly limit the module to software. Language describing the module functionally, such as being "configured to communicate with the single-node kernel and other cluster node modules" (’768 Patent, col. 9:63-65), might support an interpretation based on capability rather than specific implementation.
      • Evidence for a Narrower Interpretation: The detailed description repeatedly discusses the "cluster node module" in the context of a software module that works with a software "kernel" like Mathematica (’768 Patent, Fig. 2; col. 5:35-40). The specification states, "FIG. 2 is a block diagram showing relationships between software modules running on one embodiment of a computer cluster" (’768 Patent, col. 4:1-3), which may support limiting the term to a software implementation.

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

    • Context and Importance: This term distinguishes the invention from the prior art master-slave model. The infringement analysis will turn on whether the communication protocol between AMD's accelerators in a cluster meets the definition of "peer-to-peer" as used in the patents.
    • Intrinsic Evidence for Interpretation:
      • Evidence for a Broader Interpretation: The patent contrasts the term with grid computing where "slave nodes communicated only with the master node, and not directly with other slave nodes" (Compl. ¶34). This suggests "peer-to-peer" could broadly mean any architecture that allows direct communication between non-master nodes.
      • Evidence for a Narrower Interpretation: The specification describes the architecture as one where "Communications can occur between any two or more cluster node modules... and not just between 'adjacent' kernels" (’768 Patent, col. 6:11-14). This could be construed to require an all-to-all communication capability, which may or may not be present in the accused products depending on their specific interconnect topology (e.g., ring vs. mesh).

VI. Other Allegations

  • Indirect Infringement: The complaint alleges induced infringement based on AMD's publication of "marketing materials, technical specifications, whitepapers, datasheets, user manuals, and development and testing information" that allegedly instruct and encourage customers and system builders to integrate and use the accused accelerators in an infringing multi-node configuration (Compl. ¶55, ¶64, ¶73, ¶82, ¶91).
  • Willful Infringement: Willfulness is alleged on the basis that AMD has knowledge of the patents-in-suit and their infringement at least as of the filing date of the complaint. The complaint alleges that any continued infringement by AMD will be deliberate and willful (Compl. ¶54, ¶63, ¶72, ¶81, ¶90).

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

  • A core issue will be one of definitional scope: can the terms "kernel" and "cluster node module," which are described in the patents' specification primarily as distinct software components interacting with applications like Mathematica, be construed to cover the integrated hardware, firmware, and driver architecture of the accused high-performance accelerators?
  • A central evidentiary question will be one of technical implementation: what is the specific mechanism by which AMD's clustered accelerators communicate? The case will likely require a detailed technical comparison between the "peer-to-peer architecture" enabled by the claimed software modules and the actual hardware interconnect and software protocols (e.g., AMD's Infinity Fabric and ROCm software) used in the accused systems.
  • A third key question will be one of temporal scope and damages: given the different issue dates of the five patents, which range from June 2019 to June 2024, and the various launch dates of the accused products starting in 2021, the determination of which products are accused of infringing which patents, and for what period, will be a complex factual matter critical for assessing potential damages.