Chick3nman

NVIDIA Driver Version: 456.38       CUDA Version: 11.1

Credit: blazer
For benchmarking the card and allowing me to release the benchmarks here 

There are some warnings about the CUDA Toolkit/Driver version but they dont appear to affect functionality or speed. 
It appears to be some sort of version missmatch issue during detection and is seemingly just cosmetic. Noted for further investigation.


hashcat (v6.1.1-83-g90fb4aad) starting in benchmark mode...

Chick3nman

NVIDIA Driver Version: 522.25        CUDA Version: 11.8

Credit: blazer
For benchmarking the card and allowing me to release the benchmarks here 

The hashcat installation used includes a change to the tuning ALIAS.hctune file to include the RTX 4090 as "ALIAS_nv_sm50_or_higher".
The "Kernel exec timeout" warning is cosmetic and does not affect the speed of any of the benchmarked modes.
Benchmark was run at stock clocks on an Asus Strix 4090.

Chick3nman

This benchmark was run on an Amazon EC2 p3.8xLarge instance with 4x Nvidia Tesla Volta V100 GPUs.

Nvidia Driver Version: 390.30 
Hashcat Release: 4.1.0



hashcat (v4.1.0) starting in benchmark mode...

OpenCL Platform #1: NVIDIA Corporation

Chick3nman

NVIDIA Driver Version: 450.80.02       CUDA Version: 11.0

This is the PCIE variant of the Nvidia Tesla A100 GPU
The PCIE variant is limited to 250W and this limit is visible in the faster algorithms benchmarked
Clock speed is also limited on this card, which explains some of the numbers being lower than expected
When comparing to the RTX 3090, consider the difference in Power and Clock speed limitation, as well as the use of HBM2e memory instead of GDDR6x
A benchmark of the full SMX4 A100 variant will follow as I get access


hashcat (v6.1.1) starting in benchmark mode...

Chick3nman

Apple M1 Ultra, 20 Core CPU, 48 Core GPU, 64GB of RAM, 1TB SSD

Thanks to @fhlipZero(https://twitter.com/fhlipZero) for running the benchmark on his hardware and allowing me to publish it.
A copy of both a short benchmark and the following full run can be found at https://gist.github.com/fhlip0



hashcat (v6.2.5-340-g98b89e43d) starting in benchmark mode

Benchmarking uses hand-optimized kernel code by default.

Chick3nman

Apple M3 Pro, 11-Core CPU, 14-Core GPU, 18GB Unified Memory, 512GB SSD Storage

Thanks to disanxian for running the benchmark on his hardware and allowing me to publish it.

This benchmark uses the OpenCL Backend Runtime. A counterpart benchmark that uses the Metal runtime is also available.


hashcat (v6.2.6-827-g46ce637d3) starting in benchmark mode

Benchmarking uses hand-optimized kernel code by default.

Chick3nman

Apple M3 Pro, 11-Core CPU, 14-Core GPU, 18GB Unified Memory, 512GB SSD Storage

Thanks to disanxian for running the benchmark on his hardware and allowing me to publish it.

This benchmark uses the Metal Backend Runtime. A counterpart benchmark that uses the OpenCL runtime is also available.


hashcat (v6.2.6-827-g46ce637d3) starting in benchmark mode

Benchmarking uses hand-optimized kernel code by default.

Chick3nman

Nvidia Driver Version: 525.85.12    CUDA Version: 12.0  

This is the PCIe variant of the H100, which is limited to 350W TDP and has a more limited clock speed than the SXM5 H100.

Hashcat was built from the github master branch at the time of running. 1 mode failed to benchmark due to a thread count issue.
The H100 PCIe was added to the tuning Alias file for this run.

hashcat (v6.2.6-514-gbf1d7d078) starting in benchmark mode

Benchmarking uses hand-optimized kernel code by default.

Chick3nman

CUDA: 10.0

First generation Nintendo Switch featuring the NVIDIA Tegra X1 running a compatible version of L4T(Linux 4 Tegra).

Credit: Allan from the Hashcat Discord - https://discord.gg/HFS523HGBT

Command: `hashcat -b -O -w 4 --benchmark-all`


hashcat (v6.2.6) starting in benchmark mode

Chick3nman

NVIDIA Driver Version: 455.23.05       CUDA Version: 11.1

Credit: r4d1x 
For benchmarking the card and allowing me to release the benchmarks here 

There are a handful of algorithms failing, mostly appears related to SCRYPT and
is liking a tuning issue or small driver issue that we will need to take a look at.
Otherwise, seems fairly stable.