The demand for AI hardware is soaring, making NVIDIA GPUs highly sought after. However, production capacity struggles to keep up with the acquisition needs of companies, leading to weeks or months of waiting for NVIDIA’s AI graphics cards, which come with a hefty price tag. For instance, an NVIDIA H200 GPU costs around $30,000. While it’s a powerhouse for AI, it performs poorly when it comes to cracking password algorithms. Surprisingly, a RTX 5090, which costs significantly less, is faster in these tasks.
NVIDIA has prioritized its AI graphics cards over gaming ones due to the substantial revenue generated by the AI sector. This strategic decision has enabled the company to achieve record profits, and this trend is expected to continue as long as interest in AI remains high. There are no indications of a decline, unless the speculated AI bubble bursts, but nothing is confirmed.
RTX 5090 vs. NVIDIA H200 vs. AMD MI300X: The Gaming GPU Reigns Supreme in Password Cracking Tests
Investment in AI is escalating, with major corporations setting new records exceeding $600 billion this year. NVIDIA remains the top choice for performance if substantial hardware investment is available. However, for other tasks, even gaming GPUs prove to be superior. One such task is password cracking, a function often associated with hackers. Last year, the RTX 5090 demonstrated impressive speed, capable of cracking an 8-digit password in just 3 hours. Recent comparisons show the RTX 5090 outperforming both the $30,000 NVIDIA H200 and the AMD Instinct MI300X in password cracking, with both AI-focused cards proving slower.
The table (referencing an image not included in the text) illustrates that the gaming GPU is superior across various popular hashing algorithms. The RTX 5090 shines brightest with SHA-512, achieving 10,014 MH/s, which is 93.5% faster than the H200’s 5,173.6 MH/s. It also surpasses the MI300X by 33.7% in MD5, reaching 219.5 GH/s compared to the MI300X’s 164.1 GH/s. In MD5 specifically, the RTX 5090’s 219.5 GH/s is 76% faster than the H200’s 124.4 GH/s, despite costing 8 to 10 times less.
MI300X Possesses More INT32 Power but Lacks NVIDIA’s Optimizations
When it comes to the MTML encryption algorithm, the performance gap narrows slightly, but the RTX 5090 still maintains a clear advantage over the much more expensive AI cards. In bcrypt, the AMD MI300X falls significantly behind the others, a performance not seen in other algorithms where it generally outperforms the NVIDIA H200. Notably, the RTX 5090 is more than twice as fast as the MI300X in bcrypt.
In SHA-256, AMD’s AI accelerator initially appears to outperform NVIDIA, with only a 12% difference compared to the RTX 5090. The same trend repeats with SHA-512, where AMD is considerably faster. The reason these expensive AI cards are outperformed by a gaming GPU costing €3,000-€4,000 is their differing design philosophies. The H200 has half the number of INT32 cores compared to FP32 cores, and they are less powerful. The MI300X, while offering better INT32 performance, lacks the crucial NVIDIA optimizations found in tools like Hashcat, resulting in lower overall performance in these specific tasks.
