RDRAND on AMD Ryzen 9 5900X is flakey

Re:RDRAND on AMD Ryzen 9 5900X is flakey by mzs on Monday January 11, 2021 @03:53PM (#25736679)

please don't over complicate this:

>you're right, after I've generated many randoms, I was able to get collisions.

RDRAND() = 0x0081da17 RDRAND() = 0x0081da17 RDRAND() = 0x0178d2ea RDRAND() = 0x0178d2ea RDRAND() = 0x02a91db5 RDRAND() = 0x02a91db5 RDRAND() = 0x06c4385b RDRAND() = 0x06c4385b RDRAND() = 0x095d1bf8 RDRAND() = 0x095d1bf8 RDRAND() = 0x0990b335 RDRAND() = 0x0990b335 RDRAND() = 0x0ab033e4 RDRAND() = 0x0ab033e4 RDRAND() = 0x0ac21fae RDRAND() = 0x0ac21fae RDRAND() = 0x0d39390b RDRAND() = 0x0d39390b RDRAND() = 0x0df2f5ce RDRAND() = 0x0df2f5ce RDRAND() = 0x109e5c8a RDRAND() = 0x109e5c8a

https://github.com/systemd/systemd/issues/18184#issuecomment...

[ Parent | on HN ]

Re:RDRAND on AMD Ryzen 9 5900X is flakey by floatboth on Monday January 11, 2021 @05:15PM (#25738193)

It's literally just the birthday paradox. You get collisions very quickly for 32-bit numbers. This test finds collisions on Intel too.

[ Parent | on HN ]

Re:RDRAND on AMD Ryzen 9 5900X is flakey by edwintorok on Monday January 11, 2021 @06:23PM (#25739526)

A better test is to use dieharder, see here: https://github.com/systemd/systemd/issues/18184#issuecomment...

[ Parent | on HN ]

Re:RDRAND on AMD Ryzen 9 5900X is flakey by rurban on Tuesday January 12, 2021 @03:02PM (#25752645)

I tested it with dieharder here. It's atrocious without the BiOS patch.

https://rurban.github.io/dieharder/QUALITY.html

[ Parent | on HN ]

Re:RDRAND on AMD Ryzen 9 5900X is flakey by mzs on Friday January 15, 2021 @04:00PM (#25796393)

yep was a FW HW issue: https://github.com/systemd/systemd/issues/18184#event-420318...

[ Parent | on HN ]

Re:RDRAND on AMD Ryzen 9 5900X is flakey by mzs on Tuesday January 12, 2021 @12:14AM (#25742595)

That was before sort was used, but then the later tests were no longer sound.

[ Parent | on HN ]

Re:RDRAND on AMD Ryzen 9 5900X is flakey by holtalanm on Monday January 11, 2021 @05:16PM (#25738225)

I read through the comments on the issue -- correct me if I'm wrong, but with the program they were running (generating 2000 random numbers) isn't there a 5% chance of generating duplicates, even with RDRAND working as intended?

If that is case, maybe RDRAND shouldn't be used for uniqueness.

[ Parent | on HN ]

Re:RDRAND on AMD Ryzen 9 5900X is flakey by dragontamer on Monday January 11, 2021 @05:17PM (#25738240)

RDRAND can output a 64-bit number.

The intrinsic used is the 32-bit version of RDRAND, which is... suboptimal... for a test like this. As you've noted.

Even then, RDRAND has no entropy guarantees (again: see my top comment about RDSEED vs RDRAND).

[ Parent | on HN ]

Re:RDRAND on AMD Ryzen 9 5900X is flakey by dragontamer on Monday January 11, 2021 @03:58PM (#25736790)

>> https://github.com/systemd/systemd/issues/18184#issuecomment...

>> If you can, please try it without using the C intrinsic. I have seen multiple compiler bug reports--https://patchwork.ozlabs.org/project/gcc/patch/2012052017042... is the easiest to find--that indicate that the intrinsics don't always return the right value.

Hypothetically, the sequence discussed there could be a compiler-bug, unrolling the loop and failing to treat the results of RDRAND as volatile.

So... yeah. Lets not overcomplicate things, but lets not undercomplicate things either. Without an actual disassembly of the code posted, there's no proof yet.

[ Parent | on HN ]

Re:RDRAND on AMD Ryzen 9 5900X is flakey by rcxdude on Monday January 11, 2021 @04:42PM (#25737684)

I doubt it, a compiler bug like failing to consider rdrand volatile would not occasionally fail, but would fail consistently (perhaps every nth iteration, but it would have a pattern). I struggle to see how a compiler could miscompile it to make it inconsistent.

[ Parent | on HN ]

Re:RDRAND on AMD Ryzen 9 5900X is flakey by dragontamer on Monday January 11, 2021 @05:16PM (#25738214)

32-bit is also a pretty bad RNG to test in this manner.

Birthday paradox says that you'll only need 2^16 (65,536) runs before you get a collision. 64-bit would have a collision in one in 2^32 (4-billion). With 2000 x 10 tests (== 20,000 tests per run), there's a pretty high chance of a collision actually.

There's a lot of issues complicating the testing of something like this. I'm sure the core fact (that RDRAND might be doing something wrong) is there. But to properly vet all issues and actually get to the bottom of this bug will take some effort.

[ Parent | on HN ]