The TonHunter

Have you ever wanted to create a custom address for your TON wallet? It’s incredibly easy to do with vanitygen-plusplus. But what if you’re feeling lazy and just want a cool-looking address complete with statistics? Let me introduce you to TonHunter!

Usage

I don't have a powerful Nvidia graphics card, I'll be demonstrating everything using the Vast.ai website. I recommend it to you as well.

Step 0

• Top up your Vast.ai account with $5, that will be more than enough.
• Generate an SSH key on your computer and add it to your Vast.ai account (Account → SSH Keys).
• Copy the TonHunter.sh from my repository and put it in ~/Downloads.

Step 1 — Template

Vast.ai → Templates → Create/Edit:

• Image Path:Tag: nvidia/cuda
• Version Tag: 12.6.2-devel-ubuntu22.04
• Docker Options: -e NVIDIA_DRIVER_CAPABILITIES=all
• Launch Mode: Interactive shell server, SSH (+ check "direct SSH")
• Disk: 20 GB
• On-start Script (at the bottom):

mkdir -p /etc/OpenCL/vendors && echo "libnvidia-opencl.so.1" > /etc/OpenCL/vendors/nvidia.icd
echo 'export LANG=C.UTF-8 LC_ALL=C.UTF-8' >> /root/.bashrc
apt update && apt install -y build-essential git libssl-dev libpcre3-dev libcurl4-openssl-dev opencl-headers ocl-icd-opencl-dev clinfo gawk
cd /root && git clone https://github.com/10gic/vanitygen-plusplus && cd vanitygen-plusplus
sed -i 's|"-I %s", path_dirname(kernel_path)|"-cl-std=CL1.2 -I %s", path_dirname(kernel_path)|' ocled25519engine.c
make oclvanitygen++

This bakes in everything you'd otherwise hit one by one: the NVIDIA OpenCL ICD registration, UTF-8 locale, gawk for stats parsing, and the -cl-std=CL1.2 patch that fixes kernel compilation on recent NVIDIA drivers.

Step 2 — Picking a machine

Counter-intuitive but true: the CPU matters more than the GPU here. The ed25519 engine of the tool is CPU-fed, so a monster GPU next to an ancient Xeon will idle at 20–30% load.

• CPU: modern, many allocated cores. Best: EPYC 9xx4/9xx5, Ryzen 7950X/9950X. Good: EPYC 7xx2/7xx3 with 32+ cores. Avoid: Xeon E5 v3/v4.
• Cores: the listing shows a fraction like 32.0/128 CPU — the first number is what you get. Aim for 16+, ideally 32+.
• GPU: a single RTX 4090 / 4080 / A6000 is plenty. Multiple GPUs won't help — the CPU is the bottleneck.
• Other: verified host, Reliability 99%+, Max CUDA 12.x+. Rough total speed you can expect: weak old CPU ≈ 2–3 Mkey/s, EPYC 7742 with 32 cores ≈ 5–6 Mkey/s, fresh EPYC with 48–64 cores ≈ 10–20 Mkey/s.

Step 3 — Rent and verify

1. Rent the machine using your template.
2. Wait 3–5 minutes — the on-start script is installing packages and building the tool.
3. Connect: ssh -p <PORT> root@<IP> (Vast.ai shows the exact command under Connect).
4. Verify everything is ready:

cd /root/vanitygen-plusplus
clinfo | grep "Device Name"          # must show your NVIDIA GPU
ls oclvanitygen++                     # binary exists
grep -c "cl-std" ocled25519engine.c  # must print 1 (patch applied)

If any check fails, the on-start script is probably still running — wait a couple of minutes and retry. Still failing after 10 minutes? The machine is bad: Destroy it and rent another one.

Step 4 — Upload the script (from YOUR computer)

scp -P <PORT> ~/Downloads/TonHunter.sh root@<IP>:/root/vanitygen-plusplus/

Then on the server:

chmod +x /root/vanitygen-plusplus/TonHunter.sh

Step 5 — Hunt

cd /root/vanitygen-plusplus
./TonHunter.sh <Name> [processes] [-i]

Examples:

./TonHunter.sh Jasur                # name at the END of the address, 32 processes
./TonHunter.sh Jasur 48             # 48 processes (if you have 32+ cores)
./TonHunter.sh Jasur 48 -i          # case-insensitive (much faster)
./TonHunter.sh -m any Jasur 48      # name ANYWHERE in the address
./TonHunter.sh -m start ABcd 48     # at the start, right after UQ (note: the 3rd
                                    # address char is constrained, usually A/B/C/D)
./TonHunter.sh -k Jasur 48          # keep going after a match, collect several

How many processes: roughly equal to your allocated CPU cores (32 cores → 32–48 processes). If uptime shows load average way above your core count, reduce it.

Time to 50% probability at ~5.5 Mkey/s total:

Pattern length	Case-sensitive	With -i	With -m any -i
4 chars	seconds	instant	instant
5 chars	~2 min	seconds	instant
6 chars	~2.5 h	~5 min	seconds
7 chars	~6 days (don't)	~1 h	~2 min

Every 10 seconds the script prints: live process count, total Mkey/s, keys tried, cumulative probability, and ETA to 50%. If probability passes 90% with no hit — that's just bad luck, not a bug; keep waiting, it keeps climbing.

Ctrl+C stops only the monitor; the search keeps running in the background. To kill everything manually: pkill -9 -f oclvanitygen.

Step 6 — The catch

On a hit the script automatically:

• prints the address + PrivKey,
• stops all workers,
• writes a report hunt_stats_<Name>_<date>.txt (machine, GPU, CPU, timing, speed, keys tried, luck factor, and the result with the key),
• prints a ready-to-run scp command to download the report. Run that command on YOUR computer; it looks like:

scp -P <PORT> root@<IP>:/root/vanitygen-plusplus/hunt_stats_<Name>_<date>.txt ~/Downloads/

Step 7 — Verify and clean up (MANDATORY)

1. Import the PrivKey into a wallet (Tonkeeper: Add wallet → Import → by private key).
2. Select contract version W5 (v5r1) — otherwise you'll get a different address!
3. Check the address in the wallet matches the found one.
4. Only after that: Vast.ai → Instances → Destroy (Stop is not enough — disk still bills).

Security notes

• The key is generated on a rented machine — the host can theoretically see it. Don't keep serious money on such addresses.
• Store the PrivKey in a password manager / offline. Never share it.
• Destroy wipes the instance, but as a habit don't leave found/stats files lying around on servers.

Troubleshooting

Symptom	Cause / fix
clinfo shows nothing	ICD not registered: mkdir -p /etc/OpenCL/vendors && echo "libnvidia-opencl.so.1" > /etc/OpenCL/vendors/nvidia.icd. If libnvidia-opencl.so doesn't exist at all — recreate the instance with -e NVIDIA_DRIVER_CAPABILITIES=all
clinfo shows only CPU (pocl)	Same as above: NVIDIA ICD missing
changes address space of pointer kernel errors	Patch not applied: run the sed line from Step 1, then make oclvanitygen++
awk: syntax error	apt install -y gawk
Garbled text / underscores	export LANG=C.UTF-8 LC_ALL=C.UTF-8
SSH keeps dropping	Workers survive (nohup). Use one-shot commands: ssh -p <PORT> root@<IP> "command"
Vast.ai dashboard shows 0% / 100% load	The dashboard lies in both directions. Trust nvidia-smi and uptime on the machine
tmux frozen, login kicks you out	ssh -p <PORT> root@<IP> "tmux kill-server"

Credits

• Key search engine: vanitygen-plusplus by 10gic
• Wallet V5 contract: ton-blockchain/wallet-contract-v5