StrategyQuant X · Real benchmark
StrategyQuant X Benchmark: Strategies per Hour on a Real Server
How fast should StrategyQuant X be? Vendors rarely publish numbers, so here are ours: five weeks of real strategy generation and robustness checks on one dedicated server, taken straight from the logs SQX writes by itself.
Quick answer: on an Intel Core i9-13900 server with 16 threads and 7 years of data, StrategyQuant X generated about 172,000 strategies per hour on M30, 352,000 on H1, 676,000 on H2 and 1,467,000 on H4. A Walk Forward check took about 98 seconds per strategy and Randomize Data about 136 seconds.
Main numbers
Five weeks of StrategyQuant X on one server
The server ran from August 4 to September 9, 2026. Fifteen projects started each other in a chain (how the pipeline works). Every number on this page is calculated from the SQX project logs (global_log_*.log), 11,778 logged tasks in total.
| Result | Value |
|---|---|
| Strategies generated and tested | 130,747,057 |
| Accepted by the Builder filters | 374,397 (0.29%) |
| Builder runs | 944 |
| Time spent on generation (Build) | 190.6 hours |
| Time spent on robustness checks (Retest) | 501.8 hours |
| Average generation speed (all generated / all Build time) | 686,000 per hour |
| Median speed across Builder runs | 487,000 per hour |
| Peak speed in one run | 3,198,886 per hour |
| Median time per strategy | 7 ms |
| Average per day | about 3.6 million strategies |
| Reached final storage after all checks | 669 strategies |
Test setup
Server and workload
Server
- Intel Core i9-13900 (Raptor Lake), L2 16 MB, L3 36 MB
- 16 threads visible to Windows and SQX: 8 performance cores with Hyper-Threading. All results are for these 16 threads.
- 64 GB DDR5-4800 (2 × 32 GB)
- 2 × 1.92 TB NVMe SSD, Samsung PM9A3
- Windows Server 2019 Standard
- StrategyQuant X build 144, Java heap
-Xmx56g, ParallelGC - SQX used 15 threads, one was left to the system
Workload
- Genetic evolution, population 200
- MetaTrader 5 backtest engine (hedged)
- About 7 years of data, October 2017 to October 2024, with in-sample and two out-of-sample parts
- Builder on M1 base precision, tick check as a separate stage
- Timeframes M30, H1, H2, H4
- Long only, short only, long and short
- Forex, metals and index CFDs
- 29 instruments of tick and M1 history on disk, about 92 GB
The cycle of every project: Build, OOS, Trade Manipulation, SPP, Randomize Data, Walk Forward, Tick Retest, Final Storage, then restart. During the whole run SQX had 56 GB of heap and never ran out of memory.
Generation speed
Strategies per hour by timeframe
Timeframe is the main factor. A higher timeframe has fewer bars, so each backtest is shorter. On the same 7 years of data M30 was about 8.5 times slower than H4. This is the table to compare with your own machine.
| Timeframe | Builder runs | Strategies generated | Accepted | Accepted % | Build hours | Strategies per hour | Accepted per hour |
|---|---|---|---|---|---|---|---|
| M30 | 135 | 8,328,416 | 54,000 | 0.65% | 48.4 | 172,000 | 1,116 |
| H1 | 387 | 20,275,251 | 153,600 | 0.76% | 57.6 | 352,000 | 2,666 |
| H2 | 214 | 18,838,113 | 85,454 | 0.45% | 27.9 | 676,000 | 3,067 |
| H4 | 208 | 83,305,277 | 81,343 | 0.10% | 56.8 | 1,467,000 | 1,433 |
| All | 944 | 130,747,057 | 374,397 | 0.29% | 190.6 | 686,000 | 1,964 |
By trade direction
One-sided templates are rejected by early filters more often (for example, no trades), so generation runs faster and the share of accepted strategies is lower.
| Direction | Builder runs | Strategies generated | Accepted | Accepted % | Build hours | Strategies per hour | Accepted per hour |
|---|---|---|---|---|---|---|---|
| Long and short | 299 | 8,238,008 | 117,600 | 1.43% | 30.3 | 272,000 | 3,884 |
| Long only | 277 | 34,357,650 | 110,401 | 0.32% | 52.2 | 658,000 | 2,113 |
| Short only | 368 | 88,151,399 | 146,396 | 0.17% | 108.1 | 815,000 | 1,354 |
By exit type
| Exit type | Builder runs | Strategies generated | Accepted | Accepted % | Build hours | Strategies per hour | Accepted per hour |
|---|---|---|---|---|---|---|---|
| Exit at end of day | 317 | 18,071,982 | 126,000 | 0.70% | 76.0 | 238,000 | 1,658 |
| Exit at end of Friday | 627 | 112,675,075 | 248,397 | 0.22% | 114.6 | 983,000 | 2,167 |
By project template
All 15 templates, speed calculated as strategies generated divided by Build hours.
| Template | Builder runs | Strategies generated | Accepted | Accepted % | Build hours | Strategies per hour | Accepted per hour |
|---|---|---|---|---|---|---|---|
| M30, long and short, end of day exit | 43 | 1,865,980 | 17,200 | 0.92% | 13.8 | 135,000 | 1,243 |
| M30, long only, end of day exit | 38 | 3,121,928 | 15,200 | 0.49% | 17.3 | 181,000 | 879 |
| M30, short only, end of day exit | 54 | 3,340,508 | 21,600 | 0.65% | 17.3 | 194,000 | 1,252 |
| H1, long and short, end of day exit | 60 | 2,117,405 | 23,200 | 1.10% | 8.0 | 264,000 | 2,890 |
| H1, long and short, end of Friday exit | 65 | 1,147,039 | 26,000 | 2.27% | 4.2 | 276,000 | 6,246 |
| H1, long only, end of day exit | 51 | 2,726,234 | 20,400 | 0.75% | 6.9 | 393,000 | 2,943 |
| H1, long only, end of Friday exit | 71 | 2,958,985 | 28,000 | 0.95% | 7.8 | 377,000 | 3,570 |
| H1, short only, end of day exit | 71 | 4,899,927 | 28,400 | 0.58% | 12.6 | 388,000 | 2,247 |
| H1, short only, end of Friday exit | 69 | 6,425,661 | 27,600 | 0.43% | 18.0 | 357,000 | 1,533 |
| H2, long and short, end of Friday exit | 65 | 1,366,187 | 26,000 | 1.90% | 2.6 | 527,000 | 10,035 |
| H2, long only, end of Friday exit | 63 | 4,504,659 | 25,200 | 0.56% | 6.0 | 745,000 | 4,169 |
| H2, short only, end of Friday exit | 86 | 12,967,267 | 34,254 | 0.26% | 19.2 | 674,000 | 1,781 |
| H4, long and short, end of Friday exit | 66 | 1,741,397 | 25,200 | 1.45% | 1.7 | 1,051,000 | 15,216 |
| H4, long only, end of Friday exit | 54 | 21,045,844 | 21,601 | 0.10% | 14.1 | 1,489,000 | 1,529 |
| H4, short only, end of Friday exit | 88 | 60,518,036 | 34,542 | 0.06% | 41.0 | 1,477,000 | 843 |
How long one Builder run lasted
| Shortest | 25% of runs shorter than | Median | 75% of runs shorter than | Longest |
|---|---|---|---|---|
| under 1 min | 1.8 min | 4.2 min | 10.6 min | 4 h 23 min |
Robustness checks
Time per strategy in each check
In practice most of the time goes to heavy checks, not to generation. Generation took 190.6 hours, the checks 501.8 hours. Randomize Data and Walk Forward alone used about 410 hours. If you run many checks like these, compare servers by them first. See how long Walk Forward and Monte Carlo take.
| Check | Strategies checked | Passed | Passed % | Hours | Seconds per strategy | Strategies per hour |
|---|---|---|---|---|---|---|
| OOS (out of sample) | 66,691 | 51,061 | 76.6% | 3.2 | 0.2 | 20,800 |
| Trade Manipulation | 7,270 | 6,186 | 85.1% | 0.6 | 0.3 | 12,100 |
| Tick Retest | 5,729 | 4,589 | 80.1% | 12.6 | 7.9 | 456 |
| SPP (System Parameter Permutation) | 6,936 | 5,699 | 82.2% | 75.7 | 39 | 92 |
| Walk Forward | 5,506 | 4,594 | 83.4% | 149.6 | 98 | 37 |
| Randomize Data (Monte Carlo) | 6,900 | 5,574 | 80.8% | 260.2 | 136 | 27 |
The number of checked strategies is counted as passed plus failed, because the Total field in the SQX retest logs is not filled correctly.
Your turn
How to compare with your own PC or server
- Open the folder
user\projects\<your project>\log\in your StrategyQuant X installation. - Open the newest
global_log_*.logand find the line written after a Builder task:
Strategies generated: 18524, Time per strategy: 3 ms., ... Strategies per hour: 1008109, Accepted strategies per hour: 21769- Compare Strategies per hour with the timeframe table above. Use a project with a similar data period (about 7 years) and similar backtest precision.
- For checks, find the line with
Time per strategyafter anAutomaticRetesttask and compare it with the checks table.
What changes the number
Timeframe and history length
M30 was 8 to 9 times slower than H4 here. Twice the history means roughly twice the bars to test.
Backtest precision
The Builder here used M1 base precision. Tick precision in the Builder would be much slower; in this pipeline the tick check ran as a separate stage.
Template complexity
More entry and exit conditions, more indicators and multi-timeframe logic all add work per strategy.
How early filters reject
The earlier a strategy is thrown away, the higher the strategies per hour. A high number can mean fast hardware or strict early filters.
Compare like with like. Strategies per hour is not a pure hardware number. On this one server templates ranged from about 135,000 to 1,489,000 strategies per hour. Seconds per strategy in Walk Forward and Randomize Data is the steadier figure for comparing machines.
Reference data
Load by week, rejection reasons and an earlier run
Load by week
In weeks with few Build hours the server was busy with retests: Randomize Data, Walk Forward and SPP.
| Week starting (2026) | Builder runs | Strategies generated | Accepted | Build hours |
|---|---|---|---|---|
| August 3 | 237 | 12,432,087 | 92,543 | 14.2 |
| August 10 | 75 | 31,983,679 | 30,000 | 39.8 |
| August 17 | 288 | 41,159,870 | 115,054 | 54.5 |
| August 24 | 106 | 4,911,777 | 42,400 | 8.7 |
| August 31 | 204 | 38,351,927 | 81,600 | 65.7 |
| September 7 | 34 | 1,907,717 | 12,800 | 7.8 |
Why the Builder rejected strategies
With this set of filters you can estimate how many strategies you need to generate to get a given number of candidates. In this run 0.29% of generated strategies were accepted. The full path from 130.7 million generated to 669 final strategies is in How many strategies do you need to generate?
| Rejection reason in the Builder | Share |
|---|---|
| Net profit (OOS) is zero or negative | 20.2% |
| No trades | 18.9% |
| Did not get into the initial population | 16.9% |
| Too few trades | 13.2% |
| Trading months below 85% of the period | 11.7% |
| Open drawdown above 150% of in-sample | 10.4% |
| Many trades close on the same bar | 4.9% |
| Net profit (IS) is zero or negative | 2.4% |
| Yearly return at or below 3% | 1.1% |
| Other | under 0.2% |
Earlier run on the same server
June 2026, StrategyQuant X build 143, different templates. The spread shows how much the template matters on identical hardware.
| Template (SQX build 143) | Builder runs | Strategies generated | Accepted | Build hours | Strategies per hour |
|---|---|---|---|---|---|
| Trend template | 8 | 3,681,644 | 5,887 | 13.6 | 270,000 |
| Mean reversion template | 8 | 3,329,852 | 4,445 | 1.2 | 2,680,000 |
What it means
Choosing a server for StrategyQuant X
Generation speed sells a server, but checks decide how long your research takes. In this run one strategy needed about 98 seconds of Walk Forward and 136 seconds of Randomize Data, so a few thousand candidates already mean days of work. Size the server for the checks you run, not only for the Builder.
For the difference between a VPS and a dedicated server, recommended CPU, RAM and NVMe ranges and current prices, see StrategyQuant X Server: VPS or Dedicated? and StrategyQuant X system requirements. License and hardware budget: how much StrategyQuant X costs. Servers and prices are on the dedicated servers page. If your bottleneck is the MT5 Strategy Tester rather than SQX, look at the MT5 optimization server and the MT5 backtest farm.
FAQ
Frequently asked questions
How many strategies per hour is normal in StrategyQuant X?
There is no single normal number. On the same server with 7 years of data we measured about 172,000 strategies per hour on M30, 352,000 on H1, 676,000 on H2 and 1,467,000 on H4. Template complexity, history length, backtest precision and filters change the result several times over, so compare only similar projects.
Where do I find Strategies per hour in StrategyQuant X?
Open the folder user\projects\your project\log\ in the StrategyQuant X installation and open the newest global_log file. After each Builder task SQX writes a line with Strategies generated, Time per strategy, Strategies per hour and Accepted strategies per hour.
Why is M30 so much slower than H4?
A lower timeframe has more bars for the same period, so every backtest does more work. In this run M30 was about 8.5 times slower than H4 on the same 7 years of data.
Which StrategyQuant X check takes the longest?
In this workload Randomize Data took about 136 seconds per strategy and Walk Forward about 98 seconds. Together they used about 410 of 692 working hours, more than strategy generation itself.
Is this a synthetic benchmark?
No. The numbers come from the project logs of a real five-week research run: 944 Builder runs and the full chain of robustness checks on one dedicated server. Nothing was tuned for the test.
Can I get a server like this for StrategyQuant X?
Yes. Winservers.NET rents dedicated servers with Windows Server and Remote Desktop for StrategyQuant X research. Tell us your timeframe, data period and the checks you run, and we will suggest a configuration.
Want these numbers for your own project?
Send us your timeframe, data period and the checks you run. We will suggest a dedicated server for StrategyQuant X and tell you what to expect from it.