Conversation

Notices

1. Embed this notice
   Wolf480pl (wolf480pl@mstdn.io)'s status on Sunday, 20-Oct-2024 21:56:43 JST Wolf480pl

   I was wondering why the colder a CPU is, the further you can overclock it...

   And by colder I don't mean more watts of heat taken away from it at the same die temperature, I mean lower temperature at the same heat flow.

   Well, one of the limiting factors of overclocking is square signal becoming insufficiently square. That's because of stray capacitance. But a capacitor alone can't round the edges of a square signal - it needs some resistance on where the signal is coming from.

   1/

   • Embed this notice
     Wolf480pl (wolf480pl@mstdn.io)'s status on Sunday, 20-Oct-2024 21:56:38 JST Wolf480pl

     in reply to

     • [SUYA] ひみつ :blabcatverified:

     @ivesen but yeah I guess it can be confusing.

     I just opened GamersNexus's last livestream where the AMD guys visited him to do some XOC, skipped to the part when they started running some benchmarks, and they were saying their target with that particular CPU is -120°C, because at -130°C it won't boot, and at -125°C it only sometimes works.

     And it wasn't clear if they meant Tdie or Tpot

     Haelwenn /элвэн/ :triskell: likes this.
   • Embed this notice
     Wolf480pl (wolf480pl@mstdn.io)'s status on Sunday, 20-Oct-2024 21:56:39 JST Wolf480pl

     in reply to

     • [SUYA] ひみつ :blabcatverified:

     @ivesen I mean T_die.

     The aggregate of the readings of the dozens of temperature sensors built into the CPU silicon.

   • Embed this notice
     Wolf480pl (wolf480pl@mstdn.io)'s status on Sunday, 20-Oct-2024 21:56:40 JST Wolf480pl

     in reply to

     • [SUYA] ひみつ :blabcatverified:

     @ivesen yeah, the cold bug

     > When you're throwing a workload at it, you can go to, say, -160C, but when you reboot, there's no more workload, all of a sudden you can't boot anymore. Then you have to increase to -140C to boot."

     https://www.pcgamer.com/overclocking-a-cpu-to-7-ghz-with-the-science-of-liquid-nitrogen/

     I'm guessing at that point there isn't enough free charges in the silicon for it to (semi)conduct.

     But IIRC there is a sweet spot and it's below zero... lemme do some seraching

   • Embed this notice
     [SUYA] ひみつ :blabcatverified: (ivesen@miniwa.moe)'s status on Sunday, 20-Oct-2024 21:56:40 JST [SUYA] ひみつ :blabcatverified:

     in reply to
     @wolf480pl that sweet spot would vary wildly based on where you're measuring it though? When working with cpu's you can get some pretty huge temperature deltas between the logic and heat spreader
   • Embed this notice
     Wolf480pl (wolf480pl@mstdn.io)'s status on Sunday, 20-Oct-2024 21:56:41 JST Wolf480pl

     in reply to

     So my guess is that the reason overclocking benefits from sub-ambient temperatures is that the metal wires in the CPU have less resistance when they're cold.

   • Embed this notice
     [SUYA] ひみつ :blabcatverified: (ivesen@miniwa.moe)'s status on Sunday, 20-Oct-2024 21:56:41 JST [SUYA] ひみつ :blabcatverified:

     in reply to
     @wolf480pl pretty sure sub-ambient temperatures actually distort the signal more than anything, you actually have to heat up the cpu somewhat before applying ln2 as hitting too low temperatures makes the cpu unstable
   • Embed this notice
     Wolf480pl (wolf480pl@mstdn.io)'s status on Sunday, 20-Oct-2024 21:56:42 JST Wolf480pl

     in reply to

     Idk how stray capacitance reacts to temperature. But if it's anything like ceramic capacitors, it should have more capacitance when it's cold, right? So that can't be it.

     And what about resistance?

     Well, semiconductors have higher resistance when they're cold, and lower when they're hot, right? So that shouldn't help either.

     But! the silicon is only a small part of the chip. The rest is wires. Layers upon layers of tiny metal wires. And in metals, resistance increases with temperature.

     2/

   • Embed this notice
     翠星石 (suiseiseki@freesoftwareextremist.com)'s status on Sunday, 20-Oct-2024 22:34:34 JST 翠星石

     in reply to
     @wolf480pl There is no such thing as a "square wave", as that would require infinite bandwidth, there are only approximations.
     
     The colder something is, the less resistance it has and therefore hardware becomes much better at signalling and you can push it much harder before it loses "clock lock".
     
     This isn't really related, but some GSM towers used LN₂ cooling (not sure if just test towers), as that would allow the weakest, most feeble signals to be received from mobile devices.
   • Embed this notice
     Wolf480pl (wolf480pl@mstdn.io)'s status on Sunday, 20-Oct-2024 22:34:35 JST Wolf480pl

     in reply to

     (please someone tell me I'm wrong, or that there are other reasons I didn't take into account)

   • Embed this notice
     翠星石 (suiseiseki@freesoftwareextremist.com)'s status on Sunday, 20-Oct-2024 22:35:32 JST 翠星石

     in reply to

     • [SUYA] ひみつ :blabcatverified:
     • Nadia 🏳‍⚧
     @ivesen @wolf480pl @nadia That would likely destroy the CPU via too excessive thermal contraction before you could get to the overclocking part.
   • Embed this notice
     Wolf480pl (wolf480pl@mstdn.io)'s status on Sunday, 20-Oct-2024 22:35:33 JST Wolf480pl

     in reply to

     • [SUYA] ひみつ :blabcatverified:
     • Nadia 🏳‍⚧

     @ivesen @nadia
     there's also this:
     https://www.youtube.com/watch?v=IawfrWLDN4U

     which is sub-ambient but above-freezing and yes, the IHS is a problem...

   • Embed this notice
     [SUYA] ひみつ :blabcatverified: (ivesen@miniwa.moe)'s status on Sunday, 20-Oct-2024 22:35:33 JST [SUYA] ひみつ :blabcatverified:

     in reply to

     • Nadia 🏳‍⚧
     @wolf480pl @nadia could surpass the IHS by applying a vapor chamber directly to the die I suppose
   • Embed this notice
     [SUYA] ひみつ :blabcatverified: (ivesen@miniwa.moe)'s status on Sunday, 20-Oct-2024 22:35:34 JST [SUYA] ひみつ :blabcatverified:

     in reply to

     • Nadia 🏳‍⚧
     @nadia @wolf480pl anyway, rest of discussion aside, here's a video where LTT cools a multi hundred watts cpu with a big water loop, hitting about 40C on 400W I believe. Granted the CPU itself is significantly bigger than the ones you'd find in a laptop or whatever
     https://www.youtube.com/watch?v=Bu6PxzNR3dY
   • Embed this notice
     Nadia 🏳‍⚧ (nadia@raru.re)'s status on Sunday, 20-Oct-2024 22:35:36 JST Nadia 🏳‍⚧

     in reply to

     @wolf480pl
     > by colder I don't mean more watts
     Not saying this is wrong, but I wonder: Do we have cooling systems that can keep the CPU below 70C as effective as a pot of LN2?