Compute energy surrounding datacenter AI GPUs retains rising at a rare tempo with every new GPU era. This led a analysis group at Specops to check whether or not some standard AI GPUs might additionally carry out properly at password hacking, on the belief that these GPUs will want a second job as soon as the AI bubble lastly bursts. The outlet put Nvidia’s H200, AMD’s MI300X, and Nvidia’s RTX 5090 to see if actually costly $30,000 AI GPUs can outperform shopper graphics playing cards at password cracking.
The analysis group benchmarked 5 standard hashing algorithms, MD5, NTLM, bcrypt, SHA-256, and SHA-512, with the three aforementioned GPUs utilizing Hashcat (a preferred password restoration software). Hashcat is designed to revive passwords from password hashes saved in a file as a place to begin. Unsurprisingly, this software can also be used illegally by hackers to automate password cracking.
Go deeper with TH Premium: GPUs
| Row 0 – Cell 0 |
H200 | MI300X | RTX 5090 |
| MD5 | 124.4 GH/s | 164.1 GH/s | 219.5 GH/s |
| MTLM | 218.2 GH/s | 268.5 GH/s | 340.1 GH/s |
| bcrypt | 275.3 kH/s | 142.3 kH/s | 304.8 kH/s |
| SHA-256 | 15092.3 MH/s | 24673.6 MH/s | 27681.6 MH/s |
| SHA-512 | 5173.6 MH/s | 8771.4 MH/s | 10014.2 MH/s |
Testing reveals the H200 and MI300X falling properly behind the RTX 5090, regardless of each GPUs being considerably dearer. On common, the RTX 5090 was 20% sooner than the MI300X and a whopping 63.7% sooner than the H200. At most, the RTX 5090 was 33.7% sooner than the Mi300X in MD5 and 93.5% sooner than the H200 in SHA-512.
Article continues beneath
The downside with these AI GPUs is how Hashcat is processed; password cracking depends on 32-bit integer operations and is extraordinarily compute-intensive. It’s the precise reverse of machine studying workloads that leverage instruction sorts corresponding to FP4, BF16, FP8, and INT8.
As a outcome, Datacenter AI GPUs prioritize these directions over different directions. For occasion, the H200 has solely half as many INT32 cores as FP32 and considerably fewer cores than the RTX 5090, as a result of a lot of the work it is designed to do is dealt with by the Tensor cores. Ironically, the MI300X has a lot better INT32 efficiency than the RTX 5090, however nonetheless loses as a consequence of Nvidia optimizations baked into Hashcat code.
Specops’s testing demonstrates how streamlined trendy AI GPUs are at their job; there’s not a lot these GPUs can do past their supposed position. For now, shopper desktop GPUs will stay the quickest for cracking passwords.
Follow Tom’s Hardware on Google News, or add us as a preferred source, to get our newest information, evaluation, & critiques in your feeds.