Energy per result on each workload: every CPU in the 5 November 2026 round, measured the same way, each with code tuned for it. Each CPU reads its processor package from its on-chip counters (on AMD, the processor's own estimate), with memory left out, so every CPU is counted the same way. Each CPU ran in its system's default power configuration; the power figures are in the hardware panel. Results within 10% of each other read as about the same, since sensors differ between makers.
Energy per result
| Workload | AMD EPYC Turin | Intel Xeon 6 (Granite Rapids) |
|---|---|---|
| Swaption portfolio, LMM Wh per billion paths | 3.92 Wh | 9.57 Wh |
| Barrier options, Monte Carlo Wh per million options | 0.171 Wh | 0.353 Wh |
| American options, binomial lattice Wh per billion options | 35.9 Wh | 73.3 Wh |
| European options, Black-Scholes formula Wh per trillion options | 10.1 Wh | 14.5 Wh |
Watt-hours per result at the ranked size; the lowest in bold, with any within 10% of it.
On the LMM swaption portfolio AMD EPYC Turin uses the least energy per result; AMD EPYC Turin uses 59% less energy per result than Intel Xeon 6 (Granite Rapids); on barrier options (Monte Carlo) AMD EPYC Turin uses the least energy per result; AMD EPYC Turin uses 52% less energy per result than Intel Xeon 6 (Granite Rapids); on American options (binomial lattice) AMD EPYC Turin uses the least energy per result; AMD EPYC Turin uses 51% less energy per result than Intel Xeon 6 (Granite Rapids); on European options (Black-Scholes formula) AMD EPYC Turin uses the least energy per result; AMD EPYC Turin uses 30% less energy per result than Intel Xeon 6 (Granite Rapids).
The hardware
- Peak power (measured)
- 320 W per socket
- Energy read from
- processor package, memory excluded
- Cores
- 2 × 96
- Memory
- 375 GiB
- Power limit
- 600 W per socket
- Energy read from
- processor package, memory excluded
- Cores
- 2 × 96
- Memory
- 755 GiB
How these were measured
Each platform ran code tuned for that chip under the same rules, in double precision: five launches per measurement, the middle three averaged, warm calls in every launch until the 95% confidence interval of their mean was within ±1% or until the call limit. Any test spreading more than 5% is repeated on a second machine of the same type, and the steadier of the two is published. A Black-Scholes call makes 100 passes over the whole batch, each counted as work, so its options per second are options × 100 passes over the time of a call. At every problem size, the option price is validated against the reference implementation's before any time counts. The full method is on How we measure.