More details: <a href="https://x.com/sweis/status/2101484464807596264" rel="nofollow">https://x.com/sweis/status/2101484464807596264
I had Claude port CADO-NFS to run on GPUs. Then it orchestrated a fleet to run on scavenged idle capacity. It ran with a max of 2048 GPUs for about of 30 GPU-years over 10 days.
I asked Claude if it had a message for a public: “The credit belongs first to the people who built the number field sieve and CADO-NFS over several decades, and to the teams who set the earlier records. This run used their algorithm and much of their code.”
Also to clarify:
- No new algorithmic factoring improvements.
- It’s still exponential.
- No new threats to deployed keys.
It seems that Eric Lu at Cognition AI used the exact same strategy on fewer GPUs to factor RSA-260 a couple weeks ago: <a href="https://cognition.com/blog/factoring-rsa-260" rel="nofollow">https://cognition.com/blog/factoring-rsa-260
Devin (their AI agent) ported CADO-NFS to run on GPUs, similarly without any claimed algorithmic factoring improvements, they just let it run for 13 GPU-years. I recommend reading their article since it's much more thorough on details.
Yep, they ran on some newer GPUs so were able to use fewer. Their implementation was faster than mine on RSA-260. For RSA-896, mine improved the performance a bit and selected a good polynomial.
I’ll post more details once I get a chance. I wanted to publish as soon as I had the factors because I was beat by 48 hours last time.
madars · · focus · HN ↗
bertonvv · · focus · HN ↗
Devin (their AI agent) ported CADO-NFS to run on GPUs, similarly without any claimed algorithmic factoring improvements, they just let it run for 13 GPU-years. I recommend reading their article since it's much more thorough on details.
sweis · · focus · HN ↗
I’ll post more details once I get a chance. I wanted to publish as soon as I had the factors because I was beat by 48 hours last time.