Table 10.
Raw performance results with the use of vector instructions (AVX512) are presented beyond the scope of the iDASH competition. These experiments were conducted on a distinct hardware setup featuring a more contemporary processor, specifically, the Intel® Xeon® Platinum 8480+. This table illustrates the scaling of latency and throughput for the workloads as the number of working cores varies
| Latency (s) | Throughputa (q/s) | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| # Cores | # Cores | |||||||||||
| Method | 1 | 2 | 4 | 8 | 16 | 32 | 1 | 2 | 4 | 8 | 16 | 32 |
| “Unsupervised method” section | 10.01 | 5.99 | 3.7 | 1.74 | 1.21 | 1.3 | 40 | 66 | 108 | 229 | 308 | 329 |
| “Clustering-based supervised method” section | 12.52 | 7.32 | 4.41 | 2.02 | 1.29 | 1.21 | 32 | 54 | 90 | 197 | 310 | 330 |
| “Linear regression method” section | 6.44 | 4.08 | 2.66 | 1.08 | 0.75 | 0.85 | 62.1 | 98 | 150 | 370 | 472 | 531 |
Disclaimer: Performance varies by use, configuration and other factors. Learn more at www.Intel.com/PerformanceIndex.
aDecimal digits were truncated to leave only the integer part for a more conservative measurement