Nvidia Said This Temperature Didn’t Matter

[00:00:00]Hotspot temperature sensor on Nvidia's Blackwell GPUs is still accessible if you have access to Nvidia's internal mods tool.

[00:00:07]>> Okay, since that headline was written, uh, this has gotten a little bit more complicated.

[00:00:14]>> Uh, let me do the read for this one just in case they missed something. When the RTX50 series came out, reviewers noticed that hotspot temperatures were missing from tools like Hardware Info and MSI Afterburner. And while it looked like Nvidia had gotten rid of the feature, like it was no longer reported by the die, turns out the sensor was still there, it just wasn't accessible through the usual software.

[00:00:40]Well, Paulo Gomez, I assume a Brazilian repair technician and YouTuber, investigated a customer's Gigabyte RTX 570 Ti that was mysteriously throttling despite showing normal GPU temperatures in Windows. In Linux, Gomez was able to use Nvidia's internal mods, so modular diagnostic software, to discover the hidden hotspot sensor was hitting 107° C, even though the standard tools reported only around 68° C average GPU temperatures. And this was exactly why enthusiasts cared so much about being able to access hotspot temperatures because it would reveal if a bad mount or a bad pace job was making it so that overall the die was cool enough but the hottest zones the the most compute inensive parts of the chip were overheating. He disassembled the card found poor thermal paste coverage. Okay.

[00:01:40]Yeah, there you go. So that's exactly what I was talking about. He reapplied the paste and confirmed that the hotspot temperature dropped and the throttling disappeared. Uh Lionus comment. Oh, okay. So, they did put that in here. Uh, this seems to have been starting to make its way into regular tools. So, I think hardware info 50 series hotspot. Hold on. I'm just going to Google this live.

[00:02:04]I think yes, hardware info. This is two days ago. according to video cards adds RTX50 hotspot temperature reporting, RTSS can display it now, too. [gasps] So, I'm a little frustrated because while I'm not sure that Nvidia ever said formally that it was removed, they certainly didn't do anything to correct the speculation that it had been removed. Um, and it's a really useful tool >> for for diagnosing, especially if you're if you're doing aftermarket cooling, like mounting something like a full coverage water block can can be a little bit hazardous because there's so many surface level components like it'll it'll directly it'll directly cover uh, you know, with thermal pads. Although again, that's a great example of why this matters cuz if you have too thick of a thermal pad, right, it's making contact with all these other components on the board like your memory modules, uh, with your, uh, VRM, and if you screw up something or you put on too thick of a thermal pad or you don't tighten something down enough, it can contact the die enough. Remember, this is a water block, so it's moving heat away really fast. So it can contact the dye enough to carry heat away, but it can not be in good contact with one small area that could be a very crucial spot that you know uh like Paul like Paulo saw that could cause the car to thermal throttle. And if we don't have access to that temperature sensor, then we have no way if that knowing if that's happening.

[00:03:44]So taking away useful tools that help us diagnose hardware is just like not cool. Apparently roughly twoish years ago, um Roman from Derbau asked Nvidia about it not being there anymore and got a response of saying it was no longer accurate and no longer relevant.

[00:04:12]He obviously, you know, had thoughts about that, [laughter] but that was what they said.

[00:04:21]>> Oh man.

[00:04:24]Yeah, there's some comments saying effectively exactly this, but I had the thought of like, man, telling telling Derbower that some thermal measurement doesn't matter.

[00:04:34]I wonder if they knew who he was.

[00:04:35][laughter] >> I mean, if it's not accurate, then that's one thing.

[00:04:41]But this this proves though that they're taking action on the behavior of the GPU based on this number.

[00:04:50]>> Yeah.

[00:04:52]>> Well, okay. Which one is it, Nvidia? Is it inaccurate and irrelevant or is it actionable? Because it cannot be both.

[00:05:01]>> There's a lot of speculation.

[00:05:02]>> Yeah.

[00:05:04]>> But it makes sense to me. Uh but it is still uh obviously speculation saying that maybe you know the the modern cards run really hot and maybe it would have been bad for um perceptions or sales or or marketing or whatever to have these hotspot sensors blasting out crazy high numbers.

[00:05:23]>> Bad for marketing is the only real reason that I can think of for them to >> no longer relevant to the interest of selling more graphics cards.

[00:05:35]maybe not no longer relevant to the user.

[00:05:38]>> I see.

[00:05:39]>> But yeah, I this is just one thing I found. I don't know if they've had a more recent statement on that.

[00:05:44]>> I think everyone just kind of accepted it and moved on.

[00:05:46]>> Moved on. Yeah.

[00:05:48]>> Well, either way, I'm glad to see that this is out in the open now because it's a very useful tool.