Platforms · Measurement record
Recordturin-2026-11-05Outputs validatedFixed
AMD EPYC Turin
Measured on 5 November 2026 on a whole machine, with the published method. 26 double-precision measurements of the tuned code for this platform: five launches each, every launch recorded.
- Published method
- Run and checked by Xcelerit
- Suite version: November 2026 round
- Double precision
- 2 measurements repeated on a second machine
- Peak power (measured)
- 320 W per socket
- Processors
- AMD EPYC 9005 (Turin), 2 × 96 cores
- Memory
- 375 GiB
- Operating system
- Ubuntu 24.04.5 LTS
- Compiler
- GCC 16.2 -O3 -ffast-math -fopenmp -march=znver5 -mprefer-vector-width=512
- Libraries
- oneMKL
- Placement
- OMP_PLACES=threads OMP_PROC_BIND=close, all threads
Results at a glance
The sizes the results rank at. Time is the mean of the middle three of five launches; throughput is the problem size over that time (Black-Scholes: options × 100 passes, the work of one call). Sizes in K and M are binary (1K = 1,024); throughput is per second, in decimal units. Tuned code: tuned for this chip under the same rules. Select a row to see the launches behind it.
| Workload | Size | Time (ms) | Throughput (per s) | Energy | Within-launch CI | Launch spread | |
|---|---|---|---|---|---|---|---|
| Swaption portfolio, LMM | 1,024K paths | 23.85 | 43,966,151 | 3.923 Wh/bn | ±1.00% | 3.63% | Inspect |
| Barrier options, Monte Carlo | 16,000 options | 15.32 | 1,044,209 | 0.171 Wh/M | ±0.73% | 0.46% | Inspect |
| American options, binomial lattice | 512K options | 106.03 | 4,944,545 | 35.89 Wh/bn | ±0.99% | 1.14% | Inspect |
| European options, Black-Scholes formula | 128M options | 773.63 | 17,348,996,757 | 10.12 Wh/tn | ±0.75% | 0.56% | Inspect |
All 26 measurements
| Workload | Code | Size | Time (ms) | Within-launch CI | Launch spread | Average warm calls per launch | Published from | Other machine (ms) |
|---|---|---|---|---|---|---|---|---|
| Swaption portfolio, LMM | Tuned code | 16K | 0.544 | ±0.75% | 0.19% | 3,679 | – | – |
| Swaption portfolio, LMM | Tuned code | 32K | 0.872 | ±0.85% | 1.27% | 2,748 | – | – |
| Swaption portfolio, LMM | Tuned code | 64K | 1.597 | ±0.73% | 0.45% | 1,253 | – | – |
| Swaption portfolio, LMM | Tuned code | 128K | 2.999 | ±0.90% | 1.50% | 1,927 | – | – |
| Swaption portfolio, LMM | Tuned code | 256K | 5.881 | ±0.93% | 1.90% | 373 | – | – |
| Swaption portfolio, LMM | Tuned code | 512K | 11.948 | ±1.00% | 3.29% | 300 | – | – |
| Swaption portfolio, LMM | Tuned code | 1,024K | 23.850 | ±1.00% | 3.63% | 141 | – | – |
| Swaption portfolio, LMM | Tuned code | 2,048K | 50.113 | ±0.94% | 0.52% | 128 | machine 2 | machine 1: 47.269 (-5.7%, launch spread 6.5%) |
| Barrier options, Monte Carlo | Tuned code | 2,000 | 2.134 | ±0.74% | 0.19% | 938 | – | – |
| Barrier options, Monte Carlo | Tuned code | 4,000 | 4.015 | ±0.69% | 0.14% | 499 | – | – |
| Barrier options, Monte Carlo | Tuned code | 8,000 | 7.808 | ±0.82% | 0.34% | 256 | – | – |
| Barrier options, Monte Carlo | Tuned code | 16,000 | 15.323 | ±0.73% | 0.46% | 131 | – | – |
| Barrier options, Monte Carlo | Tuned code | 32,000 | 30.508 | ±0.55% | 0.19% | 66 | – | – |
| American options, binomial lattice | Tuned code | 16K | 3.459 | ±0.63% | 0.27% | 579 | – | – |
| American options, binomial lattice | Tuned code | 32K | 6.701 | ±0.73% | 0.34% | 299 | – | – |
| American options, binomial lattice | Tuned code | 64K | 13.225 | ±0.93% | 0.91% | 161 | – | – |
| American options, binomial lattice | Tuned code | 128K | 26.554 | ±1.00% | 0.64% | 123 | – | – |
| American options, binomial lattice | Tuned code | 256K | 52.643 | ±0.99% | 0.57% | 49 | – | – |
| American options, binomial lattice | Tuned code | 512K | 106.034 | ±0.99% | 1.14% | 36 | – | – |
| American options, binomial lattice | Tuned code | 1,024K | 211.774 | ±0.96% | 0.31% | 14 | – | – |
| American options, binomial lattice | Tuned code | 2,048K | 421.141 | ±0.60% | 0.71% | 10 | – | – |
| European options, Black-Scholes formula | Tuned code | 16M | 54.827 | ±1.00% | 11.66% · noisy | 560 | machine 1 | machine 2: 53.920 (-1.7%, launch spread 12.5%) |
| European options, Black-Scholes formula | Tuned code | 32M | 131.270 | ±1.00% | 1.49% | 235 | – | – |
| European options, Black-Scholes formula | Tuned code | 64M | 368.206 | ±0.62% | 1.44% | 30 | – | – |
| European options, Black-Scholes formula | Tuned code | 128M | 773.634 | ±0.75% | 0.56% | 49 | – | – |
| European options, Black-Scholes formula | Tuned code | 256M | 1,645.124 | ±0.71% | 0.15% | 30 | – | – |
Outputs validated before speed
A time only counts if the output is right. At every problem size, the option price the tuned code computes is validated against the reference implementation's on the same machine.
| Workload | Against the reference code | Result |
|---|---|---|
| Swaption portfolio, LMM | Agrees to 14 significant digits | Pass |
| Barrier options, Monte Carlo | Agrees to 14 significant digits | Pass |
| American options, binomial lattice | Identical to double precision | Pass |
| European options, Black-Scholes formula | Identical to double precision | Pass |
Binomial and Black-Scholes are deterministic: the tuned code gives the reference code's prices. Monte Carlo: on GPUs the barrier and LMM codes generate their random numbers differently, so their prices differ only by sampling: within 4 standard errors of the reference implementation's, and fed the same random numbers they agree with the reference implementation's to within 1e-9 (relative).
Every launch behind one measurement
Pick a workload and a size. Each dot is one launch: the mean of its warm calls, run until the 95% confidence interval was within ±1% or until the call limit. The fastest and slowest launch are dropped and the middle three averaged into the result.
Zoom: drag a box on the chart or use the slider under it; reset brings it back.
| Launch | Warm calls | Warm mean (ms) | Range (ms) | Used |
|---|
How time grows with the problem
Time against size on log scales, for every size measured. The dashed line is perfect linear scaling from the smallest size: points under it mean the machine was not yet full at small sizes.
What the record holds
The measurement record behind this page: every launch, the machine and the output validation, and the results as a table.
Full record (JSON) Results (CSV)
Free to cite with attribution: Xcelerit Finance Benchmarks, record turin-2026-11-05.
SHA-2565e0a5d5ba620b6f020950fb56d6472d1ca89f513b374ff61a8ebcd7bb1300b43
Records are fixed: once published, a record is never edited. A correction is published as a new record, linked from the old one. Checksums of every record