How does the back side of the card behind the gpu feel?

the3dfxdude wrote on 2024-03-03, 13:15:

How does the back side of the card behind the gpu feel?

There is a backplate at the back and it is a little warmer but not hot

I've never done packaging simulations so I don't really know if those kinds of thermals would be of any concern to the packaging. But when it comes to semiconductor design, the die isn't going to output heat in a uniform manner. It could be the core temperature sensor is next to a really hot portion of the chip. But it looks like they accounted for that in the monitoring program here. You didn't put in that 134 C slowdown threshold? If those are stock numbers and the monitoring program and sensor readings are accurate, you are only sitting at half of your thermal budget. The junction temperature for the design could be 150 C. So like I said, if the packaging is fine for this, or the cooling system just happens to be really good for this design, or there are other factors like good microcode clock controls from nvidia or the board manufacturer, then maybe there is nothing to worry about. Maybe someone knows this exact variant or what nvidia is doing here, but like I said, even 150 C doesn't actually cause worry for me seeing designs being done there, although it is interesting to see this in the commercial space, as that considered pretty high end design, but that is only if the packaging allows for it. The thing to do would be to monitor the temperature at load and see if it changes.

the3dfxdude wrote on 2024-03-03, 15:25:

I've never done packaging simulations so I don't really know if those kinds of thermals would be of any concern to the packaging. But when it comes to semiconductor design, the die isn't going to output heat in a uniform manner. It could be the core temperature sensor is next to a really hot portion of the chip. But it looks like the accounted for that in the monitoring program here. You didn't put in that 134 C slowdown threshold? If those are stock numbers and the monitoring program and sensor readings are accurate, you are only sitting at half of your thermal budget. The junction temperature for the design could be 150 C. So like I said, if the packaging is fine for this, or the cooling system just happens to be really good for this design, or there are other factors like good microcode clock controls from nvidia or the board manufacturer, then maybe there is nothing to worry about. Maybe someone knows this exact variant or what nvidia is doing here, but like I said, even 150 C doesn't actually cause worry for me seeing designs being done there, although it is interesting to see this in the commercial space, as that considered pretty high end design, but that is only if the packaging allows for it. The thing to do would be to monitor the temperature at load and see if it changes.

Yea, 134 is set by default, I can't change the temperature threshold, but 150°C sounds crazy. I've attached a photo. Still maybe the issue is with the bad contact? Because there's barely any thermal paste on heatsink..

You said you changed the thermal paste right? What was it doing before you changed it? Just a little more info on before and after, and where that picture comes in.

the3dfxdude wrote on 2024-03-03, 21:39:

You said you changed the thermal paste right? What was it doing before you changed it? Just a little more info on before and after, and where that picture comes in.

To be honest, I have no idea, because I replaced the paste after I bought it, that was a couple of months ago. I took this photo a few hours ago.

nuvyi wrote on 2024-03-03, 21:23:

Yea, 134 is set by default, I can't change the temperature threshold, but 150°C sounds crazy. I've attached a photo. Still maybe the issue is with the bad contact? Because there's barely any thermal paste on heatsink..

If that's how little contact there is you should add more paste. Air is far worse than lots of paste.

Did you get paste on the components around the die? There's usually a thermal sense diode. If you get thermal paste that is conductive or even capacitive on it, you will get crazy temp readings.

I've had to fix this many times on repaste jobs people did.

One reason I like MX-4, you can smear that all over exposed components and nothing will happen, except for making a mess 🤣

>134 C slowdown threshold

yeah that doesnt look right, makes no sense even in fahrenheit . I think driver is broken/glitching

rasz_pl wrote on 2024-03-04, 03:11:

>134 C slowdown threshold

yeah that doesnt look right, makes no sense even in fahrenheit . I think driver is broken/glitching

I can almost guarantee it's a sloppy repaste job. I've seen this on many nvidia cards over the years. Some will even report negative temps.

but the slowdown threshold is not reported by the card, its something driver decided on. In early 2000 nothing ran above 90C.

rasz_pl wrote on 2024-03-04, 04:09:

but the slowdown threshold is not reported by the card, its something driver decided on. In early 2000 nothing ran above 90C.

READ. It's in my post.

It's not thermal. The paste gets on the thermal diodes around the die and causes capacitive coupling which causes bad readings.

kingcake wrote on 2024-03-04, 04:27:

rasz_pl wrote on 2024-03-04, 04:09:

but the slowdown threshold is not reported by the card, its something driver decided on. In early 2000 nothing ran above 90C.

READ. It's in my post.

It's not thermal. The paste gets on the thermal diodes around the die and causes capacitive coupling which causes bad readings.

"slowdown threshold" is a number provided by the driver, not read from the card. Think "P4 thermal throttling" setting in bioses.
But this is moot as even in 2003 this setting was showing stupid numbers https://www.neoseeker.com/Articles/Hardware/R … 900ultra/9.html

kingcake wrote on 2024-03-04, 01:26:

Did you get paste on the components around the die? There's usually a thermal sense diode. If you get thermal paste that is conductive or even capacitive on it, you will get crazy temp readings.

I've had to fix this many times on repaste jobs people did.

One reason I like MX-4, you can smear that all over exposed components and nothing will happen, except for making a mess 🤣

And where's that thermal sense diode? I took a photo, did I get it right? By the way, I used exactly MX4.

I would do some copper shimming and see if the temps improve, those temps are definitely not normal.

nuvyi wrote on 2024-03-04, 09:11:

kingcake wrote on 2024-03-04, 01:26:

Did you get paste on the components around the die? There's usually a thermal sense diode. If you get thermal paste that is conductive or even capacitive on it, you will get crazy temp readings.

I've had to fix this many times on repaste jobs people did.

One reason I like MX-4, you can smear that all over exposed components and nothing will happen, except for making a mess 🤣

And where's that thermal sense diode? I took a photo, did I get it right? By the way, I used exactly MX4.

Well darn, I stand corrected.

I thought sure that was the problem.

kingcake wrote on 2024-03-04, 10:08:

nuvyi wrote on 2024-03-04, 09:11:

kingcake wrote on 2024-03-04, 01:26:

Did you get paste on the components around the die? There's usually a thermal sense diode. If you get thermal paste that is conductive or even capacitive on it, you will get crazy temp readings.

I've had to fix this many times on repaste jobs people did.

One reason I like MX-4, you can smear that all over exposed components and nothing will happen, except for making a mess 🤣

And where's that thermal sense diode? I took a photo, did I get it right? By the way, I used exactly MX4.

Well darn, I stand corrected.

I thought sure that was the problem.

👏👏👏
Elaborate? Where's this diode exactly?

Does the GPU core temp move when you stress the card? Does it get to 80?

There's no way it's hot if the heatsink isn't hot. Paste looks fine

I guess at this point, other people are saying that these are just stupid numbers being reported by the driver. I think at this point, the task is to just monitor the temperature both in program and directly yourself. If it doesn't get so hot to touch on the outside with heatsink and fan (what's that, like, 50C max?) then I think I'd move on.