Skip to content

Standard Performance

Single asset, averaged across 4 option sets. Ratios show how many times slower the competitor is relative to Rust — higher means Rust wins by more.

Competitors: Tulip Indicators (C) and TA-Lib (C)

Indicator Rust (ns) C Tulip (ns) TA-Lib (ns) C / Rust TA-Lib / Rust
ad 4,746 5,086 5,076 1.07 1.07
adosc 6,668 9,418 8,851 1.41 1.33
adx 10,746 13,494 38,473 1.26 3.58
adxr 14,041 22,981 38,823 1.64 2.77
ao 5,643 11,723 N/A 2.08 —
apo 4,709 8,727 11,231 1.85 2.39
aroon 18,187 38,601 74,113 2.12 4.08
aroonosc 21,287 37,003 70,349 1.74 3.30
atr 4,714 10,825 28,077 2.30 5.96
avgprice 1,408 1,994 3,467 1.42 2.46
bbands 7,532 13,517 23,121 1.79 3.07
bop 2,363 2,794 5,027 1.18 2.13
cci 57,429 74,740 122,246 1.30 2.13
chaikinmf 7,073 N/A N/A — —
chandelierexit 18,979 N/A N/A — —
cmo 6,155 22,382 N/A 3.64 —
cvi 6,098 14,333 N/A 2.35 —
cybercycle 16,376 N/A N/A — —
dema 6,157 6,483 22,935 1.05 3.73
di 14,071 10,300 56,209 0.73 3.99
dm 9,233 6,753 N/A 0.73 —
donchianchannel 12,954 33,080 N/A 2.55 —
dpo 2,607 2,718 N/A 1.04 —
dx 9,184 6,567 N/A 0.72 —
ef 4,715 N/A N/A — —
elderray 6,281 15,121 N/A 2.41 —
ema 4,690 10,859 10,866 2.32 2.32
emv 2,392 5,034 N/A 2.10 —
fisher 50,358 81,210 N/A 1.61 —
fosc 7,790 9,487 N/A 1.22 —
highpass 4,679 N/A N/A — —
hilberttransform 16,848 N/A 201,032 — 11.93
hma 8,312 9,266 N/A 1.11 —
homodynediscriminator 218,285 N/A 200,951 — 0.92
ichimoku 66,476 N/A N/A — —
instantaneoustrendline 224,992 N/A 234,784 — 1.04
kama 7,041 8,568 10,847 1.22 1.54
keltnerchannel 6,513 N/A N/A — —
kvo 9,320 10,773 N/A 1.16 —
linreg 6,379 8,466 N/A 1.33 —
macd 6,303 10,572 36,780 1.68 5.84
mama 224,720 N/A 215,222 — 0.96
marketfi 2,369 2,713 N/A 1.15 —
mass 5,421 12,007 N/A 2.21 —
max 5,175 14,218 18,110 2.75 3.50
md 12,578 14,796 N/A 1.18 —
medprice 1,073 1,676 2,279 1.56 2.12
mfi 7,610 16,570 19,462 2.18 2.56
min 7,717 27,341 31,539 3.54 4.09
mom 770 1,351 1,898 1.75 2.46
msw 121,197 548,438 N/A 4.53 —
natr 4,875 10,814 28,327 2.22 5.81
nvi 2,433 3,479 N/A 1.43 —
obv 3,445 3,342 3,239 0.97 0.94
ppo 4,963 9,457 14,398 1.91 2.90
psar 10,378 12,247 7,898 1.18 0.76
pvi 2,456 3,448 N/A 1.40 —
qstick 2,809 3,055 N/A 1.09 —
roc 2,365 2,653 4,940 1.12 2.09
rocr 2,370 2,704 4,970 1.14 2.10
roofingfilter 10,548 N/A N/A — —
rsi 4,859 9,434 25,737 1.94 5.30
sma 2,475 2,656 4,895 1.07 1.98
smaenvelope 7,287 N/A N/A — —
stddev 3,645 10,530 N/A 2.89 —
stoch 20,573 48,855 51,371 2.37 2.50
stochrsi 19,548 43,887 N/A 2.25 —
supersmoother 10,352 N/A N/A — —
supertrend 14,633 N/A N/A — —
tema 6,911 6,867 32,521 0.99 4.71
tr 1,521 2,005 4,078 1.32 2.68
trendmode 220,885 N/A N/A — —
trima 5,493 7,209 7,183 1.31 1.31
trix 6,634 10,664 N/A 1.61 —
trvi 6,100 N/A N/A — —
tsf 6,345 8,408 N/A 1.33 —
typprice 1,036 1,802 N/A 1.74 —
ultosc 16,273 17,670 N/A 1.09 —
vhf 16,235 39,222 N/A 2.42 —
vidya 11,976 18,644 N/A 1.56 —
volatility 8,823 17,466 N/A 1.98 —
vortex 8,142 N/A N/A — —
vosc 3,866 4,986 N/A 1.29 —
vwap 3,451 N/A N/A — —
vwma 3,384 4,957 N/A 1.46 —
wad 3,796 4,989 N/A 1.31 —
wcprice 1,062 1,786 N/A 1.68 —
wilders 4,669 10,660 N/A 2.28 —
willr 15,760 37,673 39,831 2.39 2.53
wma 6,251 8,506 4,968 1.36 0.79
zlema 5,790 8,319 N/A 1.44 —
Notable results

Rust beats C Tulip on all but 4 indicators: di (0.73×), dm (0.73×), dx (0.72×), obv (0.97×). msw is now 4.53× faster than C Tulip (previously 0.85× — Rust was slower before the SDFT implementation).

Category Indicator C / Rust TA-Lib / Rust
Largest wins vs C Tulip msw 4.53× —
cmo 3.64× —
min 3.54× 4.09×
stddev 2.89× —
max 2.75× 3.50×
Largest wins vs TA-Lib hilberttransform — 11.93×
atr 2.30× 5.96×
macd 1.68× 5.84×
natr 2.22× 5.81×
rsi 1.94× 5.30×
Rust slower than C Tulip dx 0.72× —
di 0.73× 3.99×
dm 0.73× —
obv 0.97× 0.94×
Rust slower than TA-Lib psar 1.18× 0.76×
wma 1.36× 0.79×
homodynediscriminator — 0.92×
obv 0.97× 0.94×
mama — 0.96×

Competitor: RustTa — 20 indicators benchmarked.

Indicator Rust (ns) RustTa (ns) RustTa / Rust
atr 4,714 9,591 2.03
bbands 7,532 8,312 1.10
cci 57,429 67,261 1.17
chandelierexit 18,979 43,002 2.27
ef 4,715 23,624 5.01
ema 4,690 8,135 1.73
keltnerchannel 6,513 13,026 2.00
macd 6,303 8,131 1.29
max 5,175 16,567 3.20
md 12,578 27,524 2.19
mfi 7,610 24,943 3.28
min 7,717 23,701 3.07
obv 3,445 5,411 1.57
ppo 4,963 8,154 1.64
roc 2,365 2,959 1.25
rsi 4,859 8,158 1.68
sma 2,475 4,609 1.86
stddev 3,645 8,459 2.32
stoch 20,573 47,860 2.33
tr 1,521 7,708 5.07
Notable results

Rust beats RustTa on all 20 compared indicators.

Category Indicator RustTa / Rust
Largest wins tr 5.07×
ef 5.01×
mfi 3.28×
max 3.20×
min 3.07×
Closest bbands 1.10×
cci 1.17×
roc 1.25×

Competitor: kand (v0.2) — pure Rust, TA-Lib–inspired. Default features: 64-bit precision + basic check validation. Same validation level as tulip_rs.

Note: kand uses NaN-padded full-length outputs (processing all n bars), while tulip_rs outputs only the valid computed bars. Both perform comparable per-call validation.

Indicator Rust (ns) Kand (ns) Kand / Rust
ad 4,746 5,093 1.07
adosc 6,668 24,503 3.67
adx 10,746 73,406 6.83
adxr 14,041 75,707 5.39
aroon 18,187 369,326 20.31
aroonosc 21,287 370,515 17.41
atr 4,714 22,382 4.75
bbands 7,532 22,266 2.96
bop 2,363 2,779 1.18
cci 57,429 85,887 1.50
dema 6,157 23,001 3.74
di 14,071 24,367 1.73
dm 9,233 22,344 2.42
dx 9,184 51,084 5.56
ema 4,690 8,423 1.80
macd 6,303 28,800 4.57
medprice 1,073 1,619 1.51
mfi 7,610 327,569 43.04
mom 770 1,388 1.80
natr 4,875 25,498 5.23
obv 3,445 3,679 1.07
roc 2,365 2,701 1.14
rocr 2,370 2,697 1.14
rsi 4,859 23,546 4.85
sma 2,475 4,894 1.98
stoch 20,573 699,807 34.02
supertrend 14,633 47,267 3.23
tema 6,911 27,135 3.93
tr 1,521 2,191 1.44
trima 5,493 10,111 1.84
trix 6,634 28,462 4.29
typprice 1,036 2,667 2.57
vwap 3,451 9,608 2.78
wcprice 1,062 1,649 1.55
willr 15,760 355,999 22.59
wma 6,251 42,627 6.82
Notable results

tulip_rs beats kand on all 36 compared indicators.

Sliding-window indicators show the largest gaps — kand uses O(n×period) scanning while tulip_rs uses O(n) amortised algorithms:

Indicator Kand / Rust Notes
mfi 43×
stoch 34× Ratio scales with period (k=28: 16×, k=100: 70×)
willr 23×
aroon 20×
aroonosc 17×

Recursive-formula indicators (EMA, SMA, MOM etc.) show smaller but real gaps (~1.8–2×), likely due to kand's per-bar NaN-branch for warmup output and missing mul_add FMA.

Closest results: ad (1.07×), obv (1.07×), bop (1.18×), roc (1.14×)

Competitor: ta (bukosabino/ta, pandas-based), called via the tulip_rs_python PyO3 binding. Rust native times are shown for reference — they reflect the underlying computation cost before PyO3 overhead is added. See Python Binding for setup and how to run.

Indicator Rust native (ns) tulip_rs_python (ns) ta (ns) ta / Python
ad 4,746 10,957 142,420 13×
adx 10,746 70,807 20,992,998 296×
ao 5,643 6,179 163,708 26×
aroon 18,187 21,181 9,554,491 451×
atr 4,714 10,854 11,182,703 1,030×
bbands 7,532 42,489 209,385 5×
cci 57,429 57,754 24,544,724 425×
chaikinmf 7,073 10,035 250,266 25×
dema 6,157 31,458 120,997 4×
donchianchannel 12,954 16,094 246,303 15×
dpo 2,607 2,697 111,249 41×
ema 4,690 10,476 51,133 5×
emv 2,392 3,005 145,097 48×
hma 8,312 9,690 9,515,258 982×
kama 7,041 25,185 4,189,509 166×
keltnerchannel 6,513 22,281 368,636 17×
macd 6,303 32,308 176,798 5×
mass 5,421 23,336 195,055 8×
mfi 7,610 17,889 29,322,188 1,639×
mom 770 1,156 13,631 12×
nvi 2,433 3,028 62,984,210 20,802×
obv 3,445 4,475 111,026 25×
ppo 4,963 22,245 213,941 10×
psar 10,378 25,239 172,831,901 6,848×
roc 2,365 2,642 84,236 32×
rsi 4,859 23,091 425,801 18×
sma 2,475 2,718 60,603 22×
stoch 20,573 21,957 261,123 12×
stochrsi 19,548 48,370 814,170 17×
tema 6,911 63,917 195,147 3×
trix 6,634 62,343 280,398 4×
ultosc 16,273 16,923 1,513,116 89×
vortex 8,142 9,232 829,711 90×
willr 15,760 18,076 256,528 14×
wma 6,251 4,928 3,094,723 628×
Notable results

tulip_rs_python beats ta on all 35 compared indicators. Median speedup: ~22×.

The size of the win depends entirely on how ta implements each indicator:

ta falls back to pure-Python loops — tulip_rs_python wins by the largest margin:

Indicator ta / Python ta implementation
nvi 20,802× Pure-Python loop
psar 6,848× Pure-Python loop
mfi 1,639× Pure-Python loop
atr 1,030× Pure-Python loop
hma 982× Pure-Python loop
wma 628× Pure-Python loop
aroon 451× Pure-Python loop
cci 425× Pure-Python loop
adx 296× Pure-Python loop
kama 166× Pure-Python loop

ta uses pandas/numpy C paths — the gap narrows because both sides use compiled code; PyO3 call overhead (~5–25 µs) is visible here:

Indicator Rust native (ns) tulip_rs_python (ns) ta (ns) ta / Python
tema 6,911 63,917 195,147 3×
dema 6,157 31,458 120,997 4×
trix 6,634 62,343 280,398 4×
ema 4,690 10,476 51,133 5×
bbands 7,532 42,489 209,385 5×
macd 6,303 32,308 176,798 5×

Competitors: technicalindicators (anandanand84) and indicatorts (Onur Cinar), called via the tulip_rs_node napi-rs binding. Rust native times are shown for reference — they reflect the underlying computation cost before nAPI overhead is added. See Node Binding for setup and how to run.

Only indicators where at least one reference library ran are shown; the remaining 40 indicators have tulip_rs_node timings in the database but no JS competitor to compare against.

Indicator Rust native (ns) tulip_rs_node (ns) technicalindicators (ns) indicatorts (ns) TI / Node indicatorts / Node
ad 4,779 15,289 362,618 56,345 23.72 3.69
adx 9,999 72,195 1,208,832 — 16.74 —
ao 5,393 9,476 1,179,923 51,252 124.51 5.41
apo 4,595 24,242 — 47,272 — 1.95
aroon 17,988 29,235 — 755,916 — 25.86
atr 4,559 12,759 487,010 412,299 38.17 32.32
bbands 7,114 48,775 5,712,514 287,762 117.12 5.90
bop 2,312 6,102 — 20,898 — 3.43
cci 55,726 61,800 4,545,921 143,000 73.56 2.31
chaikinmf 6,833 12,364 — 69,273 — 5.60
chandelierexit 17,252 65,897 — 1,198,695 — 18.19
dema 5,993 33,911 — 57,471 — 1.69
donchianchannel 15,182 21,793 — 822,234 — 37.73
ema 4,547 12,942 144,942 19,981 11.20 1.54
emv 2,295 6,368 — 72,371 — 11.37
keltnerchannel 6,182 27,767 — 514,102 — 18.52
macd 7,045 37,723 601,426 66,440 15.94 1.76
max 5,360 9,523 — 541,056 — 56.82
mfi 7,513 18,663 1,727,668 440,229 92.57 23.59
min 7,808 15,168 — 489,133 — 32.25
nvi 2,395 6,126 — 29,651 — 4.84
obv 3,325 6,646 313,177 20,825 47.12 3.13
ppo 4,823 24,468 — 91,518 — 3.74
psar 10,055 27,320 290,166 48,245 10.62 1.77
qstick 2,731 6,459 — 19,149 — 2.96
roc 2,294 6,158 568,223 17,056 92.27 2.77
rsi 4,719 25,181 791,156 112,723 31.42 4.48
sma 2,451 6,456 534,399 17,495 82.77 2.71
stoch 20,391 28,822 1,329,714 580,434 46.14 20.14
stochrsi 19,106 51,162 3,466,890 — 67.76 —
tema 6,734 63,641 — 95,084 — 1.49
tr 1,545 6,313 — 347,741 — 55.09
trima 5,408 7,520 — 26,088 — 3.47
trix 6,457 65,811 636,172 83,927 9.67 1.28
typprice 1,128 6,234 — 20,974 — 3.36
vortex 7,504 14,169 — 534,445 — 37.72
vwma 3,310 7,549 — 39,233 — 5.20
wilders 4,536 13,540 157,851 42,339 11.66 3.13
willr 17,218 21,420 1,400,240 829,212 65.37 38.71
wma 6,155 8,296 3,344,025 — 403.07 —
candlestick 515,546 2,914,485 107,339,622 — 36.83 —
Notable results

tulip_rs_node beats every JS competitor on all compared indicators. Median speedup: ~46× vs technicalindicators, ~4.5× vs indicatorts.

Largest wins vs technicalindicators (pure JS, loop-heavy implementations):

Indicator TI / Node Notes
wma 403× technicalindicators iterates per-bar in JS
ao 125×
bbands 117×
roc 92×
mfi 93×
sma 83×
cci 74×
stochrsi 68×
willr 65×
obv 47×

Largest wins vs indicatorts (pure TypeScript):

Indicator indicatorts / Node
max 57×
tr 55×
willr 39×
donchianchannel 38×
vortex 38×
min 32×
atr 32×
aroon 26×
mfi 24×
stoch 20×

nAPI boundary overhead — the gap between Rust native and tulip_rs_node columns reflects the fixed per-call cost of the nAPI boundary (argument marshalling, Float64Array handoff), roughly 4–25 µs depending on the number of input/output series:

Indicator Rust native (ns) tulip_rs_node (ns) Overhead
obv 3,325 6,646 ~2×
sma 2,451 6,456 ~2.6×
ema 4,547 12,942 ~2.8×
rsi 4,719 25,181 ~5.3×
bbands 7,114 48,775 ~6.9×
trix 6,457 65,811 ~10×

Candlestick scanner — scans 81 patterns in a single pass:

Time
Rust native 515,546 ns
tulip_rs_node 2,914,485 ns
technicalindicators 107,339,622 ns
TI / Node ratio 36.83×