Skip to content

Search is only available in production builds. Try building and previewing the site to test it out locally.

Partitions

Our cluster, daisyfield, consists of multiple partitions featuring diverse processor architectures, including ARM, x86, GPUs, and specialized accelerators. This heterogeneous setup allows benchmarking across a wide range of hardware configurations.

The following page provides an overview of the available partitions, along with their corresponding names, which must be referenced in workflows.

Partition CPU Accelerators Description
bellis4 ARM Cortex A-72 Coral Edge TPU (USB) Raspberry Pi 4
bellis5 ARM Cortex A-76 HAILO 8L Raspberry Pi 5
tansy ARM Cortex A-78 NVIDIA Ampere NVIDIA Jetson Orin Nano Super
zinnia AMD EPYC 9554 NVIDIA A16, Tenstorrent Wormhole N150, Axelera AI Metis
chamomile Intel Xeon Silver 4516Y+ NVIDIA A10, Tenstorrent Wormhole N150, Axelera AI Metis
aster (soon) Ampere Altra Max M128-30 NVIDIA L4, Tenstorrent Wormhole N150, Axelera AI Metis

NOTE: We’re currently working on bringing up the Axelera AI Metis and HAILO 8L processors.

Available Performance Metrics

Partition - bellis4

MetricKeyDescription
Floating-Point Operations
(Double Precision)
flop_dpCounts the total number of double-precision floating-point operations executed by the benchmark.
Floating-Point Operations
(Single Precision)
flop_spCounts the total number of single-precision floating-point operations executed by the benchmark.
Floating-Point Operations per Second
(Double Precision)
flops_dpIndicates the rate of double-precision floating-point operations executed per second, measuring computational throughput.
Floating-Point Operations per Second
(Double Precision)
flops_spIndicates the rate of single-precision floating-point operations executed per second by the benchmark.
Memory Volume (DRAM)memory_volumeMeasures the total amount of data transferred between main memory (DRAM) and CPU caches.
Memory Bandwidthmemory_bandwidthRepresents the average rate of data transferred between DRAM and processor cores, indicating memory subsystem utilization.
Operational Intensity
(Double Precision)
operational_intensity_dpIndicates the ratio of double-precision floating-point operations to memory operations, describing whether the code is compute-bound or memory-bound.
Operational Intensity
(Single Precision)
operational_intensity_spIndicates the ratio of single-precision floating-point operations to memory operations, characterizing the compute versus memory demands of the algorithm.
RuntimeruntimeRecords the total execution time of the benchmark from start to finish.

Partition - bellis5

MetricKeyDescription
Floating-Point Operations
(Double Precision)
flop_dpCounts the total number of double-precision floating-point operations executed by the benchmark.
Floating-Point Operations
(Single Precision)
flop_spCounts the total number of single-precision floating-point operations executed by the benchmark.
Floating-Point Operations per Second
(Double Precision)
flops_dpIndicates the rate of double-precision floating-point operations executed per second, measuring computational throughput.
Floating-Point Operations per Second
(Double Precision)
flops_spIndicates the rate of single-precision floating-point operations executed per second by the benchmark.
Memory Volume (DRAM)memory_volumeMeasures the total amount of data transferred between main memory (DRAM) and CPU caches.
Memory Bandwidthmemory_bandwidthRepresents the average rate of data transferred between DRAM and processor cores, indicating memory subsystem utilization.
Operational Intensity
(Double Precision)
operational_intensity_dpIndicates the ratio of double-precision floating-point operations to memory operations, describing whether the code is compute-bound or memory-bound.
Operational Intensity
(Single Precision)
operational_intensity_spIndicates the ratio of single-precision floating-point operations to memory operations, characterizing the compute versus memory demands of the algorithm.
RuntimeruntimeRecords the total execution time of the benchmark from start to finish.

Partition - tansy

MetricKeyDescription
Floating-Point Operations
(Double Precision)
flop_dpCounts the total number of double-precision floating-point operations executed by the benchmark.
Floating-Point Operations
(Single Precision)
flop_spCounts the total number of single-precision floating-point operations executed by the benchmark.
Floating-Point Operations per Second
(Double Precision)
flops_dpIndicates the rate of double-precision floating-point operations executed per second, measuring computational throughput.
Floating-Point Operations per Second
(Double Precision)
flops_spIndicates the rate of single-precision floating-point operations executed per second by the benchmark.
Memory Volume (DRAM)memory_volumeMeasures the total amount of data transferred between main memory (DRAM) and CPU caches.
Memory Bandwidthmemory_bandwidthRepresents the average rate of data transferred between DRAM and processor cores, indicating memory subsystem utilization.
Operational Intensity
(Double Precision)
operational_intensity_dpIndicates the ratio of double-precision floating-point operations to memory operations, describing whether the code is compute-bound or memory-bound.
Operational Intensity
(Single Precision)
operational_intensity_spIndicates the ratio of single-precision floating-point operations to memory operations, characterizing the compute versus memory demands of the algorithm.
RuntimeruntimeRecords the total execution time of the benchmark from start to finish.

Partition - zinnia

MetricKeyDescription
Floating-Point Operations
(Double Precision)
flop_dpCounts the total number of double-precision floating-point operations executed by the benchmark.
Floating-Point Operations
(Single Precision)
flop_spCounts the total number of single-precision floating-point operations executed by the benchmark.
Floating-Point Operations per Second
(Double Precision)
flops_dpIndicates the rate of double-precision floating-point operations executed per second, measuring computational throughput.
Floating-Point Operations per Second
(Double Precision)
flops_spIndicates the rate of single-precision floating-point operations executed per second by the benchmark.
Memory Volume (DRAM)memory_volumeMeasures the total amount of data transferred between main memory (DRAM) and CPU caches.
Memory Bandwidthmemory_bandwidthRepresents the average rate of data transferred between DRAM and processor cores, indicating memory subsystem utilization.
Operational Intensity
(Double Precision)
operational_intensity_dpIndicates the ratio of double-precision floating-point operations to memory operations, describing whether the code is compute-bound or memory-bound.
Operational Intensity
(Single Precision)
operational_intensity_spIndicates the ratio of single-precision floating-point operations to memory operations, characterizing the compute versus memory demands of the algorithm.
RuntimeruntimeRecords the total execution time of the benchmark from start to finish.

Partition - chamomile

MetricKeyDescription
Floating-Point Operations
(Double Precision)
flop_dpCounts the total number of double-precision floating-point operations executed by the benchmark.
Floating-Point Operations
(Single Precision)
flop_spCounts the total number of single-precision floating-point operations executed by the benchmark.
Floating-Point Operations per Second
(Double Precision)
flops_dpIndicates the rate of double-precision floating-point operations executed per second, measuring computational throughput.
Floating-Point Operations per Second
(Double Precision)
flops_spIndicates the rate of single-precision floating-point operations executed per second by the benchmark.
Memory Volume (DRAM)memory_volumeMeasures the total amount of data transferred between main memory (DRAM) and CPU caches.
Memory Bandwidthmemory_bandwidthRepresents the average rate of data transferred between DRAM and processor cores, indicating memory subsystem utilization.
Operational Intensity
(Double Precision)
operational_intensity_dpIndicates the ratio of double-precision floating-point operations to memory operations, describing whether the code is compute-bound or memory-bound.
Operational Intensity
(Single Precision)
operational_intensity_spIndicates the ratio of single-precision floating-point operations to memory operations, characterizing the compute versus memory demands of the algorithm.
RuntimeruntimeRecords the total execution time of the benchmark from start to finish.