NVIDIA GeForce GTX 1080 Ti Hashcat Benchmarks (152 Hash Modes)

These are the hashcat benchmark numbers I captured on a single GeForce GTX 1080 Ti Founders Edition card across all 152 hash modes hashcat supported at the time. The card was top-tier when this benchmark was taken (January 2018), held its own through 2020, and now in 2026 it sits squarely in the "budget password-cracking rig" tier where used 1080 Ti cards trade for under $250 each. Below: the original full benchmark table, a comparison with the current generation cards (RTX 3090, 4090, 5090), the cases where a 1080 Ti is still the right call in 2026, and the hashcat version / driver caveats that matter when running this card on a modern Linux box.

How fast is a GTX 1080 Ti for hashcat?

A single NVIDIA GeForce GTX 1080 Ti achieves roughly 34 GH/s on raw MD5, 57 GH/s on NTLM, *20 kH/s on bcrypt $2$ (5 cost)**, 577 kH/s on scrypt, and 533 kH/s on WPA/WPA2 under hashcat 4.0.1 in 2018 benchmark mode. Performance on hashcat 6.x and 7.x is broadly similar on the same hardware (driver and OpenCL stack matter more than hashcat version). Compared to a current-generation RTX 4090, the 1080 Ti is roughly 4-5x slower on most fast hashes (MD5, NTLM, SHA-1) and 8-10x slower on the modern slow hashes (bcrypt, argon2, scrypt) due to the 4090's larger memory bandwidth and tensor-core acceleration. The 1080 Ti remains useful in 2026 for low-budget multi-card rigs, learning, and forensic recovery on hashes that aren't deliberately slow.

Jump to:

• Test setup
• How the benchmark was run
• Full benchmark table (152 hash modes)
• GTX 1080 Ti vs RTX 3090, 4090, 5090
• When the 1080 Ti is still the right call in 2026
• Running this benchmark on hashcat 7.x today
• Driver and compatibility notes for 2026
• Legal and ethical context
• FAQ

Test setup

The Founders Edition (reference) card was deliberate: the blower-style cooler exhausts hot air out the back of the case, which matters when you stack multiple GPUs in an enclosed chassis. Aftermarket "open-air" cards dump heat back into the case and need either an open mining-rig frame or strong intake/exhaust fans to keep stable at sustained 100% utilization.

For comparison, modern rigs in 2026 usually run open frames or dedicated thermal chambers for similar reasons.

How the benchmark was run

The full benchmark mode in hashcat:

hashcat -b

Output preamble:

hashcat (v4.0.1) starting in benchmark mode...

OpenCL Platform #1: NVIDIA Corporation
======================================
* Device #1: GeForce GTX 1080 Ti, 2816/11264 MB allocatable, 28MCU

Started: Thu Jan 25 18:43:07 2018
Stopped: Thu Jan 25 19:08:29 2018

The numbers below are the optimal-case rates hashcat reports — single-hash, no salt variance, no rules. Real-world cracking speeds drop as the candidate space and salt count grow (especially for salted hashes where every new salt is effectively a separate keyspace).

Full benchmark table (152 hash modes)

GTX 1080 Ti vs RTX 3090, 4090, 5090

Approximate single-card rates from public benchmarks (hashcat.net forums, github, hashes.com community benchmarks). Real numbers vary by driver, OpenCL/CUDA stack, hashcat version, ambient temperature, and power-limit configuration — treat these as order-of-magnitude:

Pattern: the 1080 Ti is ~5x slower than a 4090 on fast hashes and 8-10x slower on slow hashes. For a heavy multi-GPU rig the comparison is more useful per-dollar than per-card — a used 1080 Ti at $200 vs a new 4090 at ~$1,800 is closer to 1.2x more H/$ on the 1080 Ti for MD5-class fast hashes, but the 4090 wins decisively on slow hashes per dollar due to its memory-bandwidth advantage.

Power draw matters at scale: 1080 Ti is rated 250W TDP, the 4090 is 450W, the 5090 is 575W. For a 24/7 cracking rig in a region where electricity costs $0.30/kWh, the per-month power cost of one card running flat-out is roughly $54 (1080 Ti), $97 (4090), or $124 (5090). The hash-per-watt math usually still favors the newer cards on slow hashes; the older cards win on the cheapest fast hashes.

When the 1080 Ti is still the right call in 2026

• You bought one used for under $250. The card was discontinued in 2019, NVIDIA stopped publishing new game-ready drivers in 2022. Used inventory is plentiful and prices are at the floor.
• You're learning hashcat / building a pentesting lab. The 1080 Ti runs every hashcat mode that the latest cards do, just slower. Same skills transfer; no need to spend $2k on a learning rig.
• You're targeting fast hashes that aren't deliberately slow. MD5 / SHA-1 / NTLM in 2026 are usually only seen in legacy systems and CTF challenges. For these, the 1080 Ti is fine — keyspaces small enough to crack on this hardware finish quickly enough that the modern-card speed advantage doesn't matter.
• You need a multi-card rig with cheap parts. Stacking 4-8 used 1080 Ti cards on a mining-frame motherboard gives ~136-272 GH/s on MD5 for the price of a single 4090. The downside is the rack power draw (1-2 kW continuous) and cooling.
• You're recovering your own forgotten password from a personal file (1Password, KeePass, encrypted PDF, etc.) and you have the card already. No need to upgrade hardware for a single recovery job.

Running this benchmark on hashcat 7.x today

The benchmark in this article was captured on hashcat 4.0.1. Hashcat 7.x (current as of 2026) added many new hash modes — the table above represents the modes that existed in 2018, not the full current set of 400+ modes.

To re-run the full current benchmark on your own 1080 Ti:

# Install or upgrade hashcat
brew install hashcat        # macOS
sudo apt install hashcat    # Debian/Ubuntu

# Run full benchmark, output to file
hashcat -b -O > my-1080-ti-benchmark.txt 2>&1

The -O flag enables "optimized kernels" (a smaller candidate space per kernel launch, faster) and is the default for hashcat -b since v6.x. Drop -O for slower-but-more-thorough numbers if you want apples-to-apples comparison with rigorous published benchmarks.

For specific modes only:

hashcat -b -m 0       # MD5
hashcat -b -m 1000    # NTLM
hashcat -b -m 22000   # WPA-PBKDF2 (modern)
hashcat -b -m 3200    # bcrypt

Mode numbers stayed stable across hashcat 4.x → 7.x; the 152 modes in the table above all still exist in hashcat 7, plus ~250 new ones for newer applications (Bitwarden, BitLocker recent variants, RAR5, Argon2, etc.).

Driver and compatibility notes for 2026

• Last NVIDIA Game Ready Driver for Pascal cards (1080 Ti generation) was 472.12 (December 2021). Pascal moved to the "legacy" Production Branch — security and bug fixes until October 2024, then no new updates.
• For Linux, the NVIDIA 535.xx legacy driver branch supports Pascal through 2026. Use the open kernel module or the proprietary one; both work for hashcat.
• On Windows 10/11 in 2026, you may need to manually pin the 472.12 or 528.xx driver if newer auto-updates roll back card compatibility. The "Studio Driver" channel is more stable than "Game Ready" for hashcat workloads.
• macOS dropped Pascal support in 10.14; the 1080 Ti has not been usable on Mac since Mojave EOL. Pure-CPU hashcat on macOS works but is dramatically slower.
• CUDA 12.x still supports compute capability 6.1 (Pascal). CUDA 13.x (if and when released) is expected to drop Pascal — at which point hashcat OpenCL is the only path forward.

Legal and ethical context

Password cracking with hashcat is legitimate for:

• Penetration testing with explicit written authorization (scope-of-work signed).
• Forensic recovery of your own data (encrypted personal files, forgotten passwords on hardware you own).
• Security research in CTF / lab environments.
• Auditing your organization's password policy against your own user database.

It is NOT legitimate for cracking hashes that belong to anyone else without authorization. The Computer Fraud and Abuse Act (US), the Computer Misuse Act 1990 (UK), the EU NIS2 Directive, and equivalent legislation in most jurisdictions treat unauthorized cracking as a crime regardless of how the hashes were obtained.

For the WordPress-specific password-hash work many readers come here for, the legitimate flow is:

• How to Change a WordPress Password — recover your own admin password via WP-CLI or wp_users direct edit, no cracking required.
• How to Store MD5 Hashes in a MySQL Database — why MD5 is broken and what to use instead (bcrypt, argon2id).
• WPScan Usage Cheat Sheet — authorized WordPress security audits.

What to do next

For the broader security cluster:

• WPScan Usage Cheat Sheet — the WordPress-specific equivalent for site audits.
• How to Store MD5 Hashes in MySQL — why MD5 is unsafe for passwords (cracked at 34 GH/s on this card) and the bcrypt / argon2id migration.
• Tor Country Codes for ExitNodes and torify — privacy/security adjacent reference.
• AWS IAM Policy Examples — IAM least-privilege patterns that make the hashes hashcat would target less valuable in the first place.

For the broader DevOps cluster:

• How to SSH into an EC2 Instance — for hosting a cracking rig in the cloud (note: AWS, GCP, and Azure all prohibit password-cracking workloads on shared infrastructure; this is for self-hosted or pentesting-licensed compute only).

External references:

• hashcat downloads and documentation
• hashcat hash mode reference (canonical mode → example hash mapping)
• Community benchmarks for GPU vs hash-mode performance comparison

FAQ