Quantum computer simulators are an indispensable tool for prototyping quantum algorithms and verifying the functioning of existing quantum computer hardware. The current largest quantum computers feature more than one thousand qubits, challenging their classical simulators. Statevector quantum simulators are challenged by the exponential increase of representable quantum states with respect to the number of qubits, making more than fifty qubits practically unfeasible. A more appealing approach for simulating quantum computers is adopting the tensor network approach, whose memory requirements fundamentally depend on the level of entanglement in the quantum circuit, and allows simulating the current largest quantum computers. This work investigates and evaluates the CUDA-Q tensor network simulators on an Nvidia Grace Hopper system, particularly the Matrix Product State (MPS) formulation. We compare the performance of the CUDA-Q state vector implementation and validate the correctness of MPS simulations. Our results highlight that tensor network-based methods provide a significant opportunity to simulate large-qubit circuits, albeit approximately. We also show that current GPUaccelerated computation cannot fully utilize GPU efficiently in the case of MPS simulations.

Memory management across discrete CPU and GPU physical memory is traditionally achieved through explicit GPU allocations and data copy or unified virtual memory. The Grace Hopper Superchip, for the first time, supports an integrated CPU-GPU system page table, hardware-level addressing of system allocated memory, and cache-coherent NVLink-C2C interconnect, bringing an alternative solution for enabling a Unified Memory system. In this work, we provide the first in-depth study of the system memory management on the Grace Hopper Superchip, in both in-memory and memory oversubscription scenarios. We provide a suite of six representative applications, including the Qiskit quantum computing simulator, using system memory and managed memory. Using our memory utilization profiler and hardware counters, we quantify and characterize the impact of the integrated CPU-GPU system page table on GPU applications. Our study focuses on first-touch policy, page table entry initialization, page sizes, and page migration. We identify practical optimization strategies for different access patterns. Our results show that as a new solution for unified memory, the system-allocated memory can benefit most use cases with minimal porting efforts.

Matrix multiplication is a core computational part of deep learning and scientific workloads. The emergence of Matrix Cores in high-end AMD GPUs, a building block of Exascale computers, opens new opportunities for optimizing the performance and power efficiency of compute-intensive applications. This work provides a timely, comprehensive characterization of the novel Matrix Cores in AMD GPUs. We develop low-level micro-benchmarks for leveraging Matrix Cores at different levels of parallelism, achieving up to 350, 88, and 69 TFLOPS for mixed, float, and double precision on one GPU. Using results obtained from the micro-benchmarks, we provide a performance model of Matrix Cores that can guide application developers in performance tuning. We also provide the first quantitative study and modeling of the power efficiency of Matrix Cores at different floating-point data types. Finally, we evaluate the high- level programmability of Matrix Cores through the rocBLAS library in a wide range of matrix sizes from 16 to 64K. Our results indicate that application developers can transparently leverage Matrix Cores to deliver more than 92% peak computing throughput by properly selecting data types and interfaces.

Disaggregated memory breaks the boundary of monolithic servers to enable memory provisioning on demand. Using network-attached memory to provide memory expansion for memory-intensive applications on compute nodes can improve the overall memory utilization on a cluster and reduce the total cost of ownership. However, current software solutions for leveraging network-attached memory must consume resources on the compute node for memory management tasks. Emerging off-path smartNICs provide general-purpose programmability at low-cost low-power cores. This work provides a general architecture design that enables network-attached memory and offloading tasks onto off-path programmable SmartNIC. We provide a prototype implementation called SODA on Nvidia BlueField DPU. SODA adapts communication paths and data transfer alternatives, pipelines data movement stages, and enables customizable data caching and prefetching optimizations. We evaluate SODA in five representative graph applications on real-world graphs. Our results show that SODA can achieve up to 7.9x speedup compared to node-local SSD and reduce network traffic by 42 % compared to disaggregated memory without SmartNIC offloading at similar or better performance.

The conventional model of resource allocation in HPC systems is static. Thus, a job cannot leverage newly available resources in the system or release underutilized resources during the execution. In this paper, we present Kub, a methodology that enables elastic execution of HPC workloads on Kubernetes so that the resources allocated to a job can be dynamically scaled during the execution. One main optimization of our method is to maximize the reuse of the originally allocated resources so that the disruption to the running job can be minimized. The scaling procedure is coordinated among nodes through remote procedure calls on Kubernetes for deploying workloads in the cloud. We evaluate our approach using one synthetic benchmark and two production-level MPI-based HPC applications - GRO-MACS and CM1. Our results demonstrate that the benefits of adapting the allocated resources depend on the workload characteristics. In the tested cases, a properly chosen scaling point for increasing resources during execution achieved up to 2x speedup. Also, the overhead of checkpointing and data reshuffling significantly influences the selection of optimal scaling points and requires application-specific knowledge.

Memory disaggregation has recently been adopted in data centers to improve resource utilization, motivated by cost and sustainability. Recent studies on large-scale HPC facilities have also highlighted memory underutilization. A promising and non-disruptive option for memory disaggregation is rack-scale memory pooling, where node-local memory is supplemented by shared memory pools. This work outlines the prospects and requirements for adoption and clarifies several misconceptions. We propose a quantitative method for dissecting application requirements on the memory system from the top down in three levels, moving from general, to multi-tier memory systems, and then to memory pooling. We provide a multi-level profiling tool and LBench to facilitate the quantitative approach. We evaluate a set of representative HPC workloads on an emulated platform. Our results show that prefetching activities can significantly influence memory traffic profiles. Interference in memory pooling has varied impacts on applications, depending on their access ratios to memory tiers and arithmetic intensities. Finally, in two case studies, we show the benefits of our findings at the application and system levels, achieving 50% reduction in remote access and 13% speedup in BFS, and reducing performance variation of co-located workloads in interference-aware job scheduling.