GPU bench

Puzzle 71 is 2⁷⁰ keys and the public key is unknown, so Kangaroo/BSGS does not apply. Sequential CUDA (BitCrack-class) is what these numbers mean. A range finishes in milliseconds. The puzzle does not.

btcpuzzle.info prints 8.06 Bkey/s on a 5090 and a column of speeds ending in .44 (982.44, 187.44…). That tail is not a measurement. They also list a GTX 980 slower than a GTX 970. We use conservative pool/author figures for compressed keys, one card. Optimized kernels can run ~2× this. Treat anything with fake decimals as marketing.

Time for 65,536 keys (our machine chunk) vs expected wait for the whole 71 space. Expected wait = space / rate. Median = ln(2) × that. Chance per day = 1 − e^(−keys_day / 2⁷⁰).

Filter cardsYour rate · Mkey/s

Card	Rate	1 range	Expected 71	Chance / day	W
RTX 5090 CUDA pool avg	4.30 Gkey/s7.48 Mkey/J	15 µs	8.70 thousand years	3.15e-7	575
RTX 4090 CUDA pool avg	2.65 Gkey/s5.89 Mkey/J	25 µs	14.12 thousand years	1.94e-7	450
RTX 5080 CUDA pool avg	2.49 Gkey/s6.92 Mkey/J	26 µs	15.02 thousand years	1.82e-7	360
RTX 4090 D CUDA pool avg	2.43 Gkey/s5.72 Mkey/J	27 µs	15.40 thousand years	1.78e-7	425
RTX 4080 Super scaled vs 4090	2.10 Gkey/s6.56 Mkey/J	31 µs	17.81 thousand years	1.54e-7	320
RTX 4080 scaled vs 4090	1.90 Gkey/s5.94 Mkey/J	34 µs	19.69 thousand years	1.39e-7	320
RTX 3090 Ti scaled vs 3090	1.75 Gkey/s3.89 Mkey/J	37 µs	21.38 thousand years	1.28e-7	450
RTX 4070 Ti Super scaled vs 4090	1.70 Gkey/s5.96 Mkey/J	39 µs	22.01 thousand years	1.24e-7	285
RTX 3090 CUDA pool avg	1.62 Gkey/s4.63 Mkey/J	40 µs	23.09 thousand years	1.19e-7	350
RTX 4070 Ti scaled vs 4090	1.50 Gkey/s5.26 Mkey/J	44 µs	24.94 thousand years	1.10e-7	285
RTX 3080 Ti scaled vs 3090	1.40 Gkey/s4.00 Mkey/J	47 µs	26.72 thousand years	1.02e-7	350
RTX 3080 scaled vs 3090	1.20 Gkey/s3.75 Mkey/J	55 µs	31.18 thousand years	8.78e-8	320
RTX 4070 scaled vs 4090	1.10 Gkey/s5.50 Mkey/J	60 µs	34.01 thousand years	8.05e-8	200
RTX 3070 Ti scaled vs 3090	900 Mkey/s3.10 Mkey/J	73 µs	41.57 thousand years	6.59e-8	290
RTX 3070 scaled vs 3090	800 Mkey/s3.64 Mkey/J	82 µs	46.76 thousand years	5.85e-8	220
RX 7900 XTX OpenCL; CUDA N/A	800 Mkey/s2.25 Mkey/J	82 µs	46.76 thousand years	5.85e-8	355
RTX 2080 Ti scaled vs 1080	700 Mkey/s2.80 Mkey/J	94 µs	53.44 thousand years	5.12e-8	250
RTX 4060 Ti scaled vs 4090	680 Mkey/s4.25 Mkey/J	96 µs	55.02 thousand years	4.98e-8	160
RTX 3060 Ti scaled vs 3090	650 Mkey/s3.25 Mkey/J	101 µs	57.55 thousand years	4.76e-8	200
RTX 4060 scaled vs 4090	600 Mkey/s5.22 Mkey/J	109 µs	62.35 thousand years	4.39e-8	115
GTX 1080 Ti scaled vs 1080	480 Mkey/s1.92 Mkey/J	137 µs	77.94 thousand years	3.51e-8	250
RTX 3060 scaled vs 3090	450 Mkey/s2.65 Mkey/J	146 µs	83.13 thousand years	3.29e-8	170
GTX 1080 BitCrack author	340 Mkey/s1.89 Mkey/J	193 µs	110.03 thousand years	2.49e-8	180
RTX 3050 measured BitCrack	220 Mkey/s1.69 Mkey/J	298 µs	170.05 thousand years	1.61e-8	130
GTX 1070 BitCrack reports	220 Mkey/s1.47 Mkey/J	298 µs	170.05 thousand years	1.61e-8	150
GTX 980 must beat 970	200 Mkey/s1.21 Mkey/J	328 µs	187.05 thousand years	1.46e-8	165
GTX 970 scaled vs 1080	160 Mkey/s1.10 Mkey/J	410 µs	233.82 thousand years	1.17e-8	145
GTX 1060 6GB scaled vs 1080	120 Mkey/s1.00 Mkey/J	546 µs	311.76 thousand years	8.78e-9	120
Ryzen 9 7950X KeyHunt CPU ballpark	8.0 Mkey/s0.05 Mkey/J	8.19 ms	4.68 million years	5.85e-10	170
6 BTC Node miner noble JS, ~1 thread	30 kkey/s0.00 Mkey/J	2.18 s	1.25 billion years	2.20e-12	65
6 BTC browser tab Web Worker	8 kkey/s0.00 Mkey/J	8.19 s	4.68 billion years	5.9e-13	15

TDP is the sticker, not the puzzle load. Mkey/J is keys per joule at that sticker. Browser and Node rows are this site — useful to start, not a farm. For rent vs prize see cost. For the 2⁷⁰ math see odds.