Reference / variance
NLHE vs PLO variance, simulated
Take the same 5bb/100 winrate into both games. The NLHE 6-max player's median worst downswing over 100,000 hands is 22.1 buy-ins; the PLO player's is 34.4. For the same 5% risk of ruin, NLHE needs 29.1 buy-ins and PLO needs 57.5, a 1.98x gap. None of these are estimates or folklore: every number comes from 50,000 simulated runs of the same engine that powers the variance calculator, at the tool's own presets (NLHE SD 100, PLO SD 140 per 100 hands).
Why PLO swings harder
Four cards make hands run closer together. PLO stacks go in with 60/40 and 55/45 edges that would be 80/20 in hold'em, so more of your results ride on coinflips and the per-100-hand standard deviation lands around 140bb against hold'em's 100bb. Bankroll math punishes that quadratically: required bankroll scales with SD squared, and (140/100)^2 is 1.96. The simulated bankroll ratios below run 1.85x to 1.98x against that square law's 1.96, which is what you want to see when a simulator and the theory it implements agree.
The practical reading is not "play NLHE". It is that a PLO grinder needs roughly double the cushion for the same psychological experience, waits about twice as long for statistical proof the winrate is real, and should expect deeper troughs at every horizon. The PLO downswing tables go further down that road; this page is the side-by-side.
Worst downswing over 100,000 hands
Peak-to-trough, in 100bb buy-ins, across 50,000 simulated 100k-hand stretches per cell. "Still losing" is the share of runs that end below zero despite the positive winrate.
| Winrate | NLHE median | PLO median | NLHE p95 | PLO p95 | Still losing: NLHE | PLO |
|---|---|---|---|---|---|---|
| 2bb/100 | 28.4 BI | 41.9 BI | 56.2 BI | 83.5 BI | 26.3% | 32.5% |
| 5bb/100 | 22.1 BI | 34.4 BI | 41.5 BI | 66.8 BI | 5.6% | 12.8% |
| 8bb/100 | 18.2 BI | 29.2 BI | 32.2 BI | 54.5 BI | 0.6% | 3.5% |
Bankroll for the same risk of ruin
Buy-ins (100bb) so that the running minimum over a 1,000,000-hand horizon stays above ruin with the stated probability. Sim-derived quantiles, not the infinite-horizon formula (which is kept as a cross-check and sits just above these).
| Winrate | NLHE, 5% risk | PLO, 5% risk | NLHE, 1% risk | PLO, 1% risk | PLO/NLHE |
|---|---|---|---|---|---|
| 2bb/100 | 73.2 BI | 135.2 BI | 112.1 BI | 202.5 BI | 1.85x |
| 5bb/100 | 29.1 BI | 57.5 BI | 45.6 BI | 89.4 BI | 1.98x |
| 8bb/100 | 18.1 BI | 35.7 BI | 28.3 BI | 56.2 BI | 1.97x |
Hands until the winrate is proven
Closed-form: the sample where a 95% confidence interval on the observed winrate first excludes zero, at the true winrate and preset SD. Until then, your graph is compatible with being a losing player.
| Winrate | NLHE (SD 100) | PLO (SD 140) |
|---|---|---|
| 2bb/100 | 960k hands | 1.88M hands |
| 5bb/100 | 154k hands | 301k hands |
| 8bb/100 | 60k hands | 118k hands |
Methodology
Identical simulation code to the variance calculator: per-100-hand increments drawn from a Student-t distribution (fat tails; df 10 for the NLHE preset, df 5 for PLO), 50,000 trials per downswing cell over 100k hands, and 40,000 trials per bankroll cell over a 1.00M-hand horizon with risk read as the quantile of each trial's running minimum. Seed and trial counts match the published PLO variance tables, and the generator asserts that every overlapping cell reproduces those published numbers exactly before it will emit this page's data. A second-seed re-run of the NLHE bankroll anchor lands within 0.26 buy-ins, and the closed-form bound sits above the finite-horizon sim, as it should.
Download the data (CC BY 4.0, attribution: solvepoker.app)
SolvePoker, "NLHE vs PLO Variance", solvepoker.app/nlhe-vs-plo-variance, updated July 2026. Questions, answered straight
Is PLO really twice the variance of NLHE?
At these presets, close to exactly that. Variance scales with the square of the per-100-hand standard deviation, and (140/100)^2 = 1.96. The simulated bankroll requirement lands almost on top of it: at 5bb/100, PLO 6-max needs 57.5 buy-ins to the NLHE player's 29.1 for the same 5% risk, a 1.98x ratio.
Is comparing the same winrate in both games fair?
It isolates the variance difference, which is the question this page answers. It does not claim your winrate transfers: bb/100 distributions differ between the games and between pools, and a good PLO player might sustain a higher winrate than the same effort earns in NLHE, which buys back some of the extra bankroll. Measure your own winrate and SD in Hand Tracker, then put your real numbers into the variance calculator.
Where do SD 100 and SD 140 come from?
They are the variance calculator's own 6-max presets for typical online cash games, the same defaults the tool ships with. Full-ring runs lower, heads-up higher (the PLO HU preset is 160). If you have 50,000+ tracked hands, your measured SD is better than any preset; Hand Tracker reports it.
Are downswings measured in buy-ins or big blinds?
Buy-ins of 100bb. A "median worst downswing over 100,000 hands" of 34.4 buy-ins means half of the simulated 100k-hand stretches at that winrate contained a peak-to-trough drop deeper than 3,441bb somewhere inside them.
Why simulate instead of using a formula?
The closed-form bankroll answer assumes normal increments and an infinite horizon. Real per-100 results are fat-tailed, so the engine draws Student-t increments (df 10 for NLHE, df 5 for PLO, heavier tails) over a finite 1,000,000-hand horizon, and reports the quantile of the running minimum across 40,000 trials. The closed form is kept as a cross-check: it sits just above the finite-horizon number, exactly as theory says it should (30.0 vs 29.1 buy-ins for NLHE at 5bb/100).
Run it with your own numbers
The variance calculator takes your winrate, SD, and bankroll and simulates your actual situation: downswing odds, risk of ruin, staking splits. Free, no account.
Open the variance calculator