Bitcoin III is new enough that nobody publishes hashrate tables for it. These are mine, measured at the wall on cards I own. Three of them are pool runs with shares accepted. Three are benchmarks that have never had a share taken, and they are marked so you can tell the difference.
It is a Bitcoin fork. Same consensus, except the block hash is SHA3-256t.
That one change means no Bitcoin ASIC can mine it. A SHA-256 machine cannot compute SHA3-256t, so today a GPU is competitive in a way it has not been on Bitcoin itself for over a decade.
BE WARNED: that is not the same as being ASIC-resistant by design. Nothing about a triple SHA3 raises an ASIC's cost above a GPU's. What protects Bitcoin III right now is that the whole network is worth roughly 530 GPUs, so taping out a chip is not worth the money. That is economic protection rather than design protection: it holds while the chain is small, and it would weaken if the chain grew. It changes with price rather than with time.
Ergo is the useful contrast. Autolykos2 is protected by design, through a memory-bandwidth floor, so its protection does not depend on the chain's size. Both are real protection. They just fail under different conditions, and it is worth knowing which one you are relying on.
It is still testing. Treat these numbers as a starting point, not a promise.
Every row here mined to a real pool and had its shares credited. None were rejected.
| GPU | Backend | MH/s | Wall watts | MH/W | Shares |
|---|---|---|---|---|---|
| RTX 4090 | CUDA | 1543 | 434 | 3.555 | 468 accepted, 0 rejected |
| RTX 5080 | CUDA | 940 | 248 | 3.790 | 258 accepted, 0 rejected |
| RTX 4060 Ti | CUDA | 415 | 137 | 3.029 | 157 accepted, 0 rejected |
The 4090 is fastest. The 5080 is the most efficient of the three, and by a wider margin than the raw speed suggests.
Watts are measured at the wall, not read off the driver. Efficiency is MH/s divided by those watts.
These ran the algorithm correctly and produced the right digests. No pool has taken a share from the Vulkan build yet.
| GPU | Backend | MH/s | Wall watts | MH/W |
|---|---|---|---|---|
| RTX 4090 | Vulkan | 1086 | 396 | 2.742 |
| RX 7900 XT | Vulkan | 538 | 289 | 1.862 |
| RX 6700 XT | Vulkan | 213 | 171 | 1.246 |
BE WARNED: a benchmark is not a pool run. The Vulkan build is checked digest for digest against the same host reference the CUDA build is, so the maths is right. That is not the same as a pool crediting your work, and until it does I am not going to call it one.
These are speed measurements. Nobody has measured what Bitcoin III actually pays, so this page does not tell you whether mining it earns anything.
BE WARNED: on this miner, every coin whose economics have been measured comes out net negative at 11.4 cents a kilowatt hour. A 4090 mining Ergo earns less per day than it costs to run. Pearl is worse. Bitcoin III might be better or it might not, and until somebody measures it the honest answer is that nobody knows.
So read the tables above as what the hardware does, not as what it pays. If you want the second question answered, put your own power price into a per-coin calculator and check the pool for what it actually credits you.
The same 4090 does 1543 MH/s on CUDA and 1086 on Vulkan. That is about 40% faster for changing one setting.
Efficiency moves the same way, 3.555 against 2.742.
The miner picks CUDA for you on an NVIDIA card. That choice is made per algorithm, not per vendor, because Ergo goes the other way and prefers Vulkan on Blackwell.
If you are on AMD, Vulkan is the only option and it works.
That matters most on older AMD cards. Pearl cannot run on RDNA2 at all, because its matrix multiply needs int8 cooperative matrix support the hardware does not have. Bitcoin III has no such requirement, so an RX 6700 XT that is refused for Pearl mines this perfectly well. The 6700 XT row above is that card doing exactly that.
These are single runs on specific cards with specific clocks and specific cooling.
A screenshot elsewhere on this site shows a 5080 at 899.97 MH/s rather than the 940 here. Same card, different run. That is normal and it is why one number is not a spec.
Expect to land within a few percent. If you are 30% off, something is wrong and the troubleshooting guide is the place to start.
Run the miner on Bitcoin III and let it settle for a minute before reading anything.
Take the hashrate from the readout, not from the startup line. The startup figure skips job handling and runs high.
Take watts at the wall with a plug meter. Driver-reported power leaves out board losses and will flatter your efficiency.
BE WARNED: a hashrate on screen is not proof you are getting paid. Check your worker on the pool site before you believe any of this, including mine.