Question Does Intel actually physically show where the cores are located on their CPUs ?
- Thread starter ooey
- Start date
KingLoki
Upstanding
- Jul 10, 2024
- 457
- 73
- 270
These will provide some info you need.
https://www.cpuid.com/softwares/hwmonitor.html
https://www.intel.com/content/www/us/en/download/17881/intel-extreme-tuning-utility-intel-xtu.html
https://www.cpuid.com/softwares/hwmonitor.html
https://www.intel.com/content/www/us/en/download/17881/intel-extreme-tuning-utility-intel-xtu.html
uWebb429
Admirable
- May 22, 2020
- 1,091
- 391
- 6,640
@ooey
Most Intel temperature sensors are only accurate to +/- 5°C across the range of temperatures they report. Everyone assumes that these sensors are space shuttle worthy. Definitely not. As long as they can trigger thermal throttling at 100°C, more or less, they are good enough.
They were never designed or intended to be 100% accurate temperature monitoring devices from 30°C to 100°C. A difference of 10°C is within the range of normal variation. Two cores sitting side by side might be physically at the exact same temperature. One might report a temperature higher than the actual temperature and the other one might report a temperature lower. This is not anything to be concerned about. Trying to micro manage the thermal paste application will not make any difference to the temperature spread if the real issue is less than perfect temperature sensors.
Most Intel temperature sensors are only accurate to +/- 5°C across the range of temperatures they report. Everyone assumes that these sensors are space shuttle worthy. Definitely not. As long as they can trigger thermal throttling at 100°C, more or less, they are good enough.
They were never designed or intended to be 100% accurate temperature monitoring devices from 30°C to 100°C. A difference of 10°C is within the range of normal variation. Two cores sitting side by side might be physically at the exact same temperature. One might report a temperature higher than the actual temperature and the other one might report a temperature lower. This is not anything to be concerned about. Trying to micro manage the thermal paste application will not make any difference to the temperature spread if the real issue is less than perfect temperature sensors.
- Mar 16, 2013
- 175,381
- 19,866
- 184,590
"more thermal paste" is rarely the solution.
geofelt
Titan
- Oct 9, 2006
- 55,151
- 4,473
- 174,240
Agreed.
In fact less paste is better.
Paste is an insulator.
What you want is as close to metal to metal contact as possible.
Some great logical answers here. Thank you for your input!
With regards to thermal paste, I always thought that there was no point to it. Copper is one of the best heat transferrers for its price and it has also been proven that rough surfaces are better than smooth for conducting heat. I suppose the argument is that if you have a paste there are less contact gaps between the surfaces as the paste is semi-liquid, and fills gaps better (more contact = more even heat transference).
The best thermal pastes are copper-based, I would assume.
The thermal sensors though... It does beg the question - why put four thermal sensors in such close proximity to each other on a CPU if there is quite a large tolerance allowed (i.e. 1 per core)? Why not just have one? This is probably what lulls people into the thinking that these sensors are so accurate that they must be space shuttle technology. In case one fails perhaps? Never known this to be the case for a CPU!
I guess the only answer to heat dissipation is to use as large a heatsink as you can fit in your case and make sure you use the best conductive material you can get your hands on.
Does it make sense to have the CPU heatsink in contact with the case in as large an area as possible or would that cause potential electrical vulnerabilities for the CPU? I've never liked the concept of water-cooling!
With regards to thermal paste, I always thought that there was no point to it. Copper is one of the best heat transferrers for its price and it has also been proven that rough surfaces are better than smooth for conducting heat. I suppose the argument is that if you have a paste there are less contact gaps between the surfaces as the paste is semi-liquid, and fills gaps better (more contact = more even heat transference).
The best thermal pastes are copper-based, I would assume.
The thermal sensors though... It does beg the question - why put four thermal sensors in such close proximity to each other on a CPU if there is quite a large tolerance allowed (i.e. 1 per core)? Why not just have one? This is probably what lulls people into the thinking that these sensors are so accurate that they must be space shuttle technology. In case one fails perhaps? Never known this to be the case for a CPU!
I guess the only answer to heat dissipation is to use as large a heatsink as you can fit in your case and make sure you use the best conductive material you can get your hands on.
Does it make sense to have the CPU heatsink in contact with the case in as large an area as possible or would that cause potential electrical vulnerabilities for the CPU? I've never liked the concept of water-cooling!
Last edited:
CountMike
Titan
- Oct 31, 2015
- 34,768
- 3,509
- 147,990
A lot of wrong assumptions.
Paste or pad is needed to fill up microscopic imperfections on the surfaces which don't allow full contact. If you look at best polished mirror like surface under microscope, you would see mountains and valleys which disallow full contact. Paste may be an insulator but it's lousy at the job and transfer more heat than air trapped in between would,
Copper is not best heat transfer material, it's Silver but more expensive although there were some pastes with it.
Paste is constituted from some kind of grease (now mostly special high temp silicone) but what really transfers heat is microscopic particles of metal or in latest pastes special ceramic particles which transfer heat even better than metal and can also be smaller which allows for more of them and also to fill up smaller pits.
Temperature sensors are many, more than can be seen, in modern CPUs even for every core. They are used internally in CPU microcode to regulate core's behavior according to it's temperatures. Chipset, VRM ,GPUs have own sensors for same reasons.
Yes - I knew that about silver! Copper is the most cost effective conductor for heat dissipation as you say. Best to use as little paste as possible though. Some slap it on thick, which I don't think does any good at all. It is only really helpful to fill those micro-gaps in solid conductors.
Last edited:
uWebb429
Admirable
- May 22, 2020
- 1,091
- 391
- 6,640
@ooey
Intel used the CPU socket temperature to control thermal throttling back in the Pentium 4 days. The socket temperature was far removed from the peak core temperature. Sensor accuracy from one motherboard to the next was an even bigger issue. Intel had to build in a lot more temperature headroom to try and guarantee that no part of the core would ever get too hot. This could compromise long term reliability.
Individual core temperature sensors positioned on the hottest spots on the core were a big improvement. This allowed Intel to avoid premature thermal throttling which ultimately increased performance. Reduced fan noise was another benefit.
It is not critical to use temperature sensors that are 100% accurate. Thermal shutdown which protects the CPU from any damage is not supposed to happen until a peak core temperature somewhere in the 125°C to 130°C range. Triggering thermal throttling in the 100°C to 105°C range is good enough to protect the CPU against any temperature related damage. That is the only purpose of Intel core temperature sensors.
Most Intel Core i CPUs are set to begin thermal throttling at 100°C. For the 3rd Gen 3770K, Intel increased the thermal throttling temperature to 105°C. For the 4th Gen and beyond, Intel went back to 100°C. For the most recent 285K that was just released last week, Intel has once again gone back to 105°C.
https://ark.intel.com/content/www/u...-processor-285k-36m-cache-up-to-5-70-ghz.html
If Intel was having any temperature related reliability issues, they would have lowered the thermal throttling temperature many years ago.
You definitely need to use good quality thermal paste. This helps maximize heat transfer from the cores to the cooler so the heat can be dissipated. I too once thought, who needs this stuff. Big mistake!
Intel used the CPU socket temperature to control thermal throttling back in the Pentium 4 days. The socket temperature was far removed from the peak core temperature. Sensor accuracy from one motherboard to the next was an even bigger issue. Intel had to build in a lot more temperature headroom to try and guarantee that no part of the core would ever get too hot. This could compromise long term reliability.
Individual core temperature sensors positioned on the hottest spots on the core were a big improvement. This allowed Intel to avoid premature thermal throttling which ultimately increased performance. Reduced fan noise was another benefit.
It is not critical to use temperature sensors that are 100% accurate. Thermal shutdown which protects the CPU from any damage is not supposed to happen until a peak core temperature somewhere in the 125°C to 130°C range. Triggering thermal throttling in the 100°C to 105°C range is good enough to protect the CPU against any temperature related damage. That is the only purpose of Intel core temperature sensors.
Most Intel Core i CPUs are set to begin thermal throttling at 100°C. For the 3rd Gen 3770K, Intel increased the thermal throttling temperature to 105°C. For the 4th Gen and beyond, Intel went back to 100°C. For the most recent 285K that was just released last week, Intel has once again gone back to 105°C.
https://ark.intel.com/content/www/u...-processor-285k-36m-cache-up-to-5-70-ghz.html
If Intel was having any temperature related reliability issues, they would have lowered the thermal throttling temperature many years ago.
You definitely need to use good quality thermal paste. This helps maximize heat transfer from the cores to the cooler so the heat can be dissipated. I too once thought, who needs this stuff. Big mistake!
Very interesting info there, thanks for that. So the heat sensors are actually built into the cores? Looks like they've built in a good bit of thermal leeway there too. If this is the case is there any real danger of a nuclear meltdown or is it just a case that you will get an unstable o/c until you revert backwards?
My idle loads are all under 30 degrees currently, which is a good start. I'm going to de-lid the CPU and use quality silver paste on it as many have had great success with this method with regards heat reduction (one guy on youtube got a 40 degree reduction at 100% load from 98 degrees on 1 of his cores!).
It's worth doing because for an old CPU the 3770k seems to be a great over-clocker. I'd be happy with 4.5GHz but I know some have pushed things up to 4.7 stable and beyond. Today I also learnt just why the I5-2500k was such a good CPU to o/c whilst the 3770k can be a bit hit and miss - intel resorted to quite cheap paste in the latter (the whole Ivy Bridge range I think) whilst for the former they used solder. It seems that you need to put a little more work into the 3770 to make it a superstar.
Would I be right in saying that if you are going to O/C the CPU, you might as well do the same with the memory and GPU too - as you'd start to get bottlenecks if you didn't? I find that anything done by ASUS s/w tends to be very aggressive and leaves the pc unstable. The MSI afterburner utility seems to be much better though.
My idle loads are all under 30 degrees currently, which is a good start. I'm going to de-lid the CPU and use quality silver paste on it as many have had great success with this method with regards heat reduction (one guy on youtube got a 40 degree reduction at 100% load from 98 degrees on 1 of his cores!).
It's worth doing because for an old CPU the 3770k seems to be a great over-clocker. I'd be happy with 4.5GHz but I know some have pushed things up to 4.7 stable and beyond. Today I also learnt just why the I5-2500k was such a good CPU to o/c whilst the 3770k can be a bit hit and miss - intel resorted to quite cheap paste in the latter (the whole Ivy Bridge range I think) whilst for the former they used solder. It seems that you need to put a little more work into the 3770 to make it a superstar.
Would I be right in saying that if you are going to O/C the CPU, you might as well do the same with the memory and GPU too - as you'd start to get bottlenecks if you didn't? I find that anything done by ASUS s/w tends to be very aggressive and leaves the pc unstable. The MSI afterburner utility seems to be much better though.
Last edited:
BFG-9000
Splendid
- Sep 17, 2016
- 3,759
- 592
- 29,590
All you have to do is google for the die shot. For example this is 32nm Sandy Bridge like your 2500k:
which is 35% larger than 22nm Ivy Bridge like your 3770k :
Even though Ivy Bridge is some 15-20w less at the same clock, the reduction of heat transfer area (160 mm² vs 216 mm² die size) and use of thermal paste makes it run much hotter. While the cores aren't numbered, usually the middle two are the hottest. Unless you delid yours, you'd normally google for pics of delidded CPUs to for example not orient your heatsink so all the cores are under one heatpipe.
which is 35% larger than 22nm Ivy Bridge like your 3770k :
Even though Ivy Bridge is some 15-20w less at the same clock, the reduction of heat transfer area (160 mm² vs 216 mm² die size) and use of thermal paste makes it run much hotter. While the cores aren't numbered, usually the middle two are the hottest. Unless you delid yours, you'd normally google for pics of delidded CPUs to for example not orient your heatsink so all the cores are under one heatpipe.
- Mar 16, 2013
- 175,381
- 19,866
- 184,590
still - it can't hurt to give everything a lenient o/c, can it? There's always tolerance built in to gfx cards and memory. I let the ASUS s/w give the GPU an auto o/c and it boosted everything on the benchmarks from around 25% to 35% (but became unstable).
There must be a sweet spot for all the components. Finding it is probably time consuming though.
There must be a sweet spot for all the components. Finding it is probably time consuming though.
Last edited:
The actual core area is small, so it can be quite easily be covered with silver thermal paste when de-lided. Just a little worried about doing it because although silver liquid it has high heat transference it is electrically conductive, so you have to be careful what you are doing. It seems to be worth it for the results you get though.
Thanks for your input there
Last edited:
Facebook
Twitter
Instagram