Technologies
Made for GPUs and CPUs. Made hard for ASICs.
TenQ is the proof of work behind Tenrai, designed and built from scratch, and it opens mining to graphics cards and CPUs alike. Instead of one fixed formula, every attempt runs small programs over a 2 GiB dataset: general-purpose work that every GPU and CPU can do, and that special chips find costly to copy. At its core sits SHAKE256, ready for the quantum era.
GPU + CPU
Both can mine
Graphics cards mine fastest, and every CPU can join in. No special hardware needed.
2 GiB
Dataset
Read at unpredictable places and renewed about once a day. Memory, not raw hashing speed, sets the pace.
16
Kinds of integer instructions
Add, multiply, rotate, shift, compare, byte swap and more: a chip must build them all, not one hash circuit.
SHAKE256
Quantum-ready core
From the SHA-3 standards, FIPS 202 and SP 800-185, at the start and the end of every attempt.
Mine on GPU or CPU
TenQ asks for what general-purpose processors already bring: memory, integer math and the ability to run programs. That is why graphics cards and CPUs can both mine it. A graphics card works through many attempts at once and mines fastest. A CPU mines too, just slower, and every thread adds to your hashrate.
Graphics cards mine fastest
NVIDIA and AMD cards mine through CUDA or OpenCL. The miner checks every card against the CPU and tunes its batch size on its own.
Every CPU can mine
Any processor mines too, just slower than a graphics card. You choose how many threads it uses, from one to all.
Both at once
One miner runs your graphics cards and your CPU threads side by side. A single switch keeps it to the CPU or the GPU alone.
Right in the browser
The browser miner uses the graphics card through WebGPU or the CPU threads, with nothing to install.
Not just another hash
Classic proof-of-work hashes such as SHA-256 are one fixed formula. A chip that does nothing else runs it thousands of times more efficiently than any graphics card, and mining turns into a race of special hardware. TenQ turns every attempt into small programs instead: 128 rounds across 32 lanes, each round reading 1 KiB of the 2 GiB dataset at a place the round before chose, running one of the day's 16 programs that the data picks, writing private memory and mixing 16 kinds of integer instructions.
Designed from scratch, not a fork
TenQ is Tenrai's own design, not a fork of an existing mining algorithm. The dataset, the programs, the instruction mix and the miners for CPUs, CUDA, OpenCL and WebGPU were designed and developed for Tenrai from the ground up.
Hard to copy in silicon
Any public algorithm can be built as a chip. What counts is how much the chip gains. TenQ spreads its work over everything a modern GPU is good at, so a special chip has to rebuild the parts of a graphics card that matter: fast memory, private memory for every attempt in flight, a full integer unit and program control. That keeps the edge of ASICs and FPGAs small, and mining open to every graphics card and CPU.
Memory sets the pace
Every round reads the 2 GiB dataset at a place no one can predict. Speed comes from memory, and memory is the same commodity for everyone.
The data picks the program
Every round the data picks one of the day's 16 programs, so a chip has to carry all of them and switch between them the way a processor does, round after round.
No easy days
The programs change every day, but every day's 16 programs hold exactly the same count of each instruction. There are no easy days to wait for.
No memory shortcut
A chip can leave out the dataset, but then it must rebuild each entry it needs from the 256 MiB cache, with 352 reads and mixing steps for every one. That path is many times slower, so the 2 GiB stay the cheapest way to mine.
Fair, not exclusive
TenQ does not lock special hardware out. Its aim is to take away the shortcut: an ASIC or FPGA may run TenQ like any graphics card, it just should not get there much faster or cheaper. In theory even quantum computers could take part one day, as fair participants and not as a shortcut.
That is on purpose. With Proof of Useful Work, which we are researching, every kind of computing hardware could earn with the verified jobs it does best. And because block rewards shrink by 0.3 % every day, verified work is meant to become the main way miners earn over time.
Our research into Proof of Useful WorkReady for the quantum era
Grover's algorithm lets a quantum computer search for a valid block in fewer steps, the square root of a classic search. TenQ makes each of those steps enormously expensive: every one has to run the complete work function reversibly, with lookups into a 2 GiB dataset, 128 dependent rounds and private writes. And all of it has to finish in about four minutes, because a block counts only within 96 cuts of the cut it builds on. At the core sits SHAKE256 from the SHA-3 standard.
Why we kept difficulty and nonce
We could have taken another route. Other mining formats are built so that a quantum search has nothing to speed up, because they drop the search for a nonce below a difficulty target. For Tenrai, that was never an option.
We kept the classic format on purpose. Difficulty and nonce are the proven heart of proof of work: open to anyone, quick and exact to check, with a difficulty that follows whatever hardware joins. Rather than steer around quantum computers, TenQ meets them head on and makes every quantum step expensive.
The classic format also keeps the way open for proof of useful work, a possible next step in which the same hardware produces blocks and does useful computing.
Reliable on every machine
Every node can check
Full mode keeps the 2 GiB dataset, light mode needs only the 256 MiB cache. Both give exactly the same result.
No pause at the daily change
The dataset changes every 34,560 cuts, about once a day. Nodes, miners and pools build the next one ahead, so mining goes on through the change.
Same answer everywhere
Only integer arithmetic, no floating point: every CPU and GPU computes the same value, bit for bit.
At a glance
- Hardware
- graphics cards through CUDA, OpenCL or WebGPU, and any CPU
- Dataset
- 2 GiB, renewed every 34,560 cuts
- Cache
- 256 MiB, rebuilds any dataset entry with 352 reads
- Private memory
- 4 KiB scratchpad per attempt
- Programs
- 16 per day of 12 instructions each, one picked by the data every round
- Rounds
- 128 across 32 lanes of 16 registers
- Cryptographic core
- SHAKE256 and cSHAKE256, FIPS 202 and SP 800-185
- Result
- a 256-bit value, valid at or below the difficulty target
Put your hardware to work
Mine TRAI with the graphics card or CPU you already have. Half of all new TRAI goes to the miners.