Hacker News Reader

HNHacker News

Top New Best Ask Show Jobs

Comment by layla5alive | Hacker News Reader

Parent Full thread

layla5alive·

Some 32-bit counter somewhere used when in NVLINK overflows?

View on HN

66 days + 12 hours are 5,745,600,000,000,000 ns. The log2 of this is 52.351...

Javascript and some other languages only have integer precision up to 52 bits then they switch to floating point.

Curious.

It's 32 bits of milliseconds, right? Hm, no, that would overflow much sooner (49.7 days).

It's a uint32_t of 750 Hz "jiffies", which does overflow at ~66 days.

While that seems like a convincing explanation, 750Hz is a rather odd value to use for a timer, and more importantly the overflow would be at 66d6h43m43s instead of the reported ~66d12h.

66 days 12 hours would put it at 747.5 Hz. A different report had 66 days 10 hours 16 minutes which works out to 748 Hz.

Maybe the clock was just feeling a little sluggish? /s

It is indeed the explanation: https://github.com/NVIDIA/open-gpu-kernel-modules/pull/1014

Wild.

Bingo! Someone decided to store timestamps in float64 which has 52 bit mantissa, and the time functions break when losing precision.

Isn't 32bit counter 49 days? Assuming that one was counting milliseconds, at least.

Only remember that because that's the limit for Windows 95…

100ns intervals. My favorite part of that story is how long after Windows 95 was released before anybody discovered the bug.

That's because people actually powered off their computer after work/leisure sessions. Someone on an unlimited night dial-up could had discovered it well before "anybody" but it's not like there was a built-in function to actually send a crash report to Redmond.

https://i.sstatic.net/p9hUgGfg.png

Reply on news.ycombinator.com