Discussion Raptor Lake Build Thread

Page 9 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Carfax83

Diamond Member
Nov 1, 2010
6,805
1,494
136
niiiiiiiiiiiiiiiiiiiice, temps? and cooler
Spoke too soon. I had two crashes with the offset voltage set to -0.165. And my PC wasn't doing anything strenuous at all. First crash was watching Amazon Prime using the Windows app, and the second was right clicking on the Windows start button. See, that is the danger of using offset voltages because it takes away voltage from all phases of the CPU, including idle voltage as well. So I think my idle voltage was a bit too low and that caused the crash. I have now set it to -0.150 and it seems more stable but I will monitor.

As for temps, I ran Cinebench R23 for about 5 minutes and the CPU package temp was 77c, which is significantly less than the 84c I got the last time. The cooler I'm using is a Noctua NH-U12A with a thermalright contact frame in a Fractal Design Torrent case.
 

Carfax83

Diamond Member
Nov 1, 2010
6,805
1,494
136
BTW, does anyone know at what temp HWinfo will report "thermal throttling" in red font.

Seem like it's something over 80C but I'm not sure exactly?
I think thermal throttling happens at 100c if I'm not mistaken but don't quote me on that.
 

DrMrLordX

Lifer
Apr 27, 2000
20,518
9,605
136
It's been a few years since Heat Transfer in college... but here goes, let's discuss.
Look, you can go that route, but you really need to look at what AMD users (in particular Radeon VII users) have had to deal with when operating these products. If you know your heat transfer, you'll know that a surface with variable thermal flux is really challenging to cool, especially if your best material making contact anywhere on the surface is copper @ ~400 w/mK. If all your heat is coming from a relatively small area, and if the heat can't spread well through the silicon immediately adjacent to that hotspot (which it generally can't), then the effectiveness of your cooling solution is significantly diminished vs. trying to cool a hot plate which has the same thermal flux spread out among resistors evenly spaced across the plate. When designing a cooling solution, you can't just move the source of heat flux - that's a given. The only thing you can do is increase the w/mK of your cooling solution. Most of us are stuck with copper for immediate contact with the source of heat flux, not counting the TIM interface.

But to drive home what I am saying, let's consider my Radeon VII for which I have custom water cooling, thanks to a Bykski waterblock, two D5 pumps in series, and a MO-RA3 420 radiator. Complete overkill, right? Okay, great. It's a pretty big die, too.

Using good old Unigine Superposition in default 1080p medium to stress the chip (with a mild undervolt, because Radeon VII is overvolted out of the box), I get the following relevant figures:

Max GPU power draw: 228W
Max GPU temp (bulk): 35C
Max GPU temp (hotspot): 55C

Note that GPU-z is nice enough to track hotspot temperature. Not sure if RDNA2 tracks that, but if it doesn't then color me disappointed. In any case, looking at the GPU bulk temp (idle temp is 27.5C), it would seem the cooling really did a great job. A delta T of only ~8C on a ~230W heat load? Amazing! Ship it. But the hotspot temp tells a different story: AMD's extensive sensor network picked up temps as high as 55C somewhere on the die. Woops. Still okay for a dGPU, but it tells a completely different story from the bulk die temperature. Back when I tried overclocking Radeon VII, I found that even slight increases in voltage caused major increases in hotspot temperature. And the TIM I used (Kyronaut) is reported to fail from repeated exposure to temps higher than 80C, depending on whom you ask. It's also got a reputation for pump-out due to differences in temperature across the IHS/die (in the case of Radeon VII, there is no IHS, but for Raptor Lake there there would be). I try to keep my temps low enough to minimize pump-out and eliminate any risk that the TIM would dry out. This particular installation is over three years old now, so I'm guessing whatever I did, worked.

Now imagine that same phenomenon, but spread out across multiple dice and with an IHS in the way. That's pretty much my 3900X, and I can barely overclock that thing without temps getting out of control. Matisse is kind enough to just report hotspot temperatures (or at least, most monitoring software records just that and reports it as the primary temperature of concern) so that I do not concern myself with anything else. It also rises in temp rapidly beyond a certain point, and that isn't just because of the process (per se), it's also because of the challenges of cooling hotspots. And that CPU never really got over 160W no matter what I did to it.

That's why I was asking if Raptor Lake has hotspots or not, because it is highly-relevant. Currently Raptor Lake is the hungriest CPU you can buy, and it does push a lot of cooling solutions to their limits; at the same time, I'm reasonably certain that my current cooling setup would be more-effective in helping Raptor Lake achieve higher clocks than on my old 3900X due to the centralized source of heat flux (only one die) and (maybe) the absence of prominent hotspots. That being said, it's even easier for my cooling setup to handle Radeon VII due to the larger die size, even if the heat flux is about the same. But the hotspots really prevented my Radeon VII from gaining serious clocks/voltage at higher volts, which (combined with driver problems wrt overclocking) was always a disappointment.
 

phillyman36

Golden Member
Jun 28, 2004
1,755
160
106
If I use an Asus z90 Hero with a 13900k can I set the bios to run Intel forced limits or do I have to manually set pl1 and pl2 limits? I dont want to mess with setting voltages myself as that above my paygrade. I want to go air cooling so what pl limits need to be set to tame it a little.(with a Fractal Torrent and either a Noctua d15 or U12a)
 

Carfax83

Diamond Member
Nov 1, 2010
6,805
1,494
136
If I use an Asus z90 Hero with a 13900k can I set the bios to run Intel forced limits or do I have to manually set pl1 and pl2 limits? I dont want to mess with setting voltages myself as that above my paygrade. I want to go air cooling so what pl limits need to be set to tame it a little.(with a Fractal Torrent and either a Noctua d15 or U12a)
The Asus Z790 Hero is a great motherboard with tons of options so you will definitely be able to run Intel forced limits and anything else you'd like to change. I don't know why'd you buy such an elite motherboard though if you don't plan on doing any kind of extensive tweaking.

If you're going the air cooling route, the Noctua NH-D15 would probably be slightly better than the U12A for handling the default power settings as it's a much bigger heatsink. The U12A can still do it though, especially when paired with the Fractal Torrent. The main advantage of using the U12A is that it has exceptional RAM clearance and while being smaller, it has an extra heatpipe so the cooling ability is still top notch.

I would definitely recommend getting a Thermalright contact frame as well. As for what PL limits, it depends on what clockspeed you are gunning for. If you want to retain stock clockspeeds, then you will probably need at least 230w in my opinion or just stick with the default 253w. But if you underclock like what I've done with mine, you can go much lower. Mine is maxed at 5.2ghz as I don't need anything more and it's blazing fast at just 1.066v 215w with low temps.

 
Last edited:

Hulk

Diamond Member
Oct 9, 1999
3,695
1,299
136
Look, you can go that route, but you really need to look at what AMD users (in particular Radeon VII users) have had to deal with when operating these products. If you know your heat transfer, you'll know that a surface with variable thermal flux is really challenging to cool, especially if your best material making contact anywhere on the surface is copper @ ~400 w/mK. If all your heat is coming from a relatively small area, and if the heat can't spread well through the silicon immediately adjacent to that hotspot (which it generally can't), then the effectiveness of your cooling solution is significantly diminished vs. trying to cool a hot plate which has the same thermal flux spread out among resistors evenly spaced across the plate. When designing a cooling solution, you can't just move the source of heat flux - that's a given. The only thing you can do is increase the w/mK of your cooling solution. Most of us are stuck with copper for immediate contact with the source of heat flux, not counting the TIM interface.

But to drive home what I am saying, let's consider my Radeon VII for which I have custom water cooling, thanks to a Bykski waterblock, two D5 pumps in series, and a MO-RA3 420 radiator. Complete overkill, right? Okay, great. It's a pretty big die, too.

Using good old Unigine Superposition in default 1080p medium to stress the chip (with a mild undervolt, because Radeon VII is overvolted out of the box), I get the following relevant figures:

Max GPU power draw: 228W
Max GPU temp (bulk): 35C
Max GPU temp (hotspot): 55C

Note that GPU-z is nice enough to track hotspot temperature. Not sure if RDNA2 tracks that, but if it doesn't then color me disappointed. In any case, looking at the GPU bulk temp (idle temp is 27.5C), it would seem the cooling really did a great job. A delta T of only ~8C on a ~230W heat load? Amazing! Ship it. But the hotspot temp tells a different story: AMD's extensive sensor network picked up temps as high as 55C somewhere on the die. Woops. Still okay for a dGPU, but it tells a completely different story from the bulk die temperature. Back when I tried overclocking Radeon VII, I found that even slight increases in voltage caused major increases in hotspot temperature. And the TIM I used (Kyronaut) is reported to fail from repeated exposure to temps higher than 80C, depending on whom you ask. It's also got a reputation for pump-out due to differences in temperature across the IHS/die (in the case of Radeon VII, there is no IHS, but for Raptor Lake there there would be). I try to keep my temps low enough to minimize pump-out and eliminate any risk that the TIM would dry out. This particular installation is over three years old now, so I'm guessing whatever I did, worked.

Now imagine that same phenomenon, but spread out across multiple dice and with an IHS in the way. That's pretty much my 3900X, and I can barely overclock that thing without temps getting out of control. Matisse is kind enough to just report hotspot temperatures (or at least, most monitoring software records just that and reports it as the primary temperature of concern) so that I do not concern myself with anything else. It also rises in temp rapidly beyond a certain point, and that isn't just because of the process (per se), it's also because of the challenges of cooling hotspots. And that CPU never really got over 160W no matter what I did to it.

That's why I was asking if Raptor Lake has hotspots or not, because it is highly-relevant. Currently Raptor Lake is the hungriest CPU you can buy, and it does push a lot of cooling solutions to their limits; at the same time, I'm reasonably certain that my current cooling setup would be more-effective in helping Raptor Lake achieve higher clocks than on my old 3900X due to the centralized source of heat flux (only one die) and (maybe) the absence of prominent hotspots. That being said, it's even easier for my cooling setup to handle Radeon VII due to the larger die size, even if the heat flux is about the same. But the hotspots really prevented my Radeon VII from gaining serious clocks/voltage at higher volts, which (combined with driver problems wrt overclocking) was always a disappointment.
Good analysis. Can we agree that hotspots or not, the goal for the cooling solution is to maximize the delta T between the cooler and the CPU, right? If we increase delta T we increase heat transfer.

If the cooler can move a lot of thermal energy then it will also maximize delta T and thus heat transfer, which ultimately lowers temps.

A cooler really can't be designed to accommodate hotspots. It can only be designed to move as much heat away from the contact area as possible. That is in a nutshell why I think the 13900K is the best CPU to test thermal solutions. Again, this assumes the heat transfer coefficient from CPU to cooler is about the same for high end coolers. Once I know that I just want to know how much energy the cooler can dissipate into the environment.
 

DrMrLordX

Lifer
Apr 27, 2000
20,518
9,605
136
Good analysis. Can we agree that hotspots or not, the goal for the cooling solution is to maximize the delta T between the cooler and the CPU, right? If we increase delta T we increase heat transfer.
Given a static ambient temperature, the only way to accomplish that feat is to let the CPU get hotter or to use active refrigeration of some kind. Both options have distinct negatives. Otherwise your goal is to try to hold the CPU as close to ambient as possible, and that requires reducing thermal resistance.
 

Hulk

Diamond Member
Oct 9, 1999
3,695
1,299
136
Given a static ambient temperature, the only way to accomplish that feat is to let the CPU get hotter or to use active refrigeration of some kind. Both options have distinct negatives. Otherwise your goal is to try to hold the CPU as close to ambient as possible, and that requires reducing thermal resistance.
Agreed.
 

DrMrLordX

Lifer
Apr 27, 2000
20,518
9,605
136
One point I think needs to be emphasized for Raptor Lake (and Alder Lake) builders is to think carefully about TIM choice. Specifically, Kryonaut. I love the TIM's rated performance, but above ~80C interface temperature, it reportedly can dry out and become cracked/crumbly which is not going to be good for your mount. Raptor Lake has high reported die temperatures, but I have no idea how that translates into interface temps between the IHS and base of your HSF/waterblock. Also while Intel 7 CPUs may not have designs that lend themselves to hotspotting local to a particular area of logic (maybe? still looking for thermal maps of active Alder Lake or Raptor Lake dice), it should be noted that there will be hotter regions on a Raptor Lake die due to the P-cores and e-cores having different power draw and (presumably) heat densities. That may encourage pump-out. Some of my favorite TIM like Conductonaut and Kryonaut are subject to pump-out, especially if there is a large delta T between different areas of the thermal interface.

What kind of TIM are people here using for their Raptor Lake builds? I haven't seen a whole lot of commentary here. Pretty sure the ubiquitous MX-4 also suffers from pump-out. Supposedly MX-5 is less prone to pump-out and may be a better product for modern CPUs with highly-variable heat regions on the IHS.
 

DooKey

Golden Member
Nov 9, 2005
1,782
440
136
One point I think needs to be emphasized for Raptor Lake (and Alder Lake) builders is to think carefully about TIM choice. Specifically, Kryonaut. I love the TIM's rated performance, but above ~80C interface temperature, it reportedly can dry out and become cracked/crumbly which is not going to be good for your mount. Raptor Lake has high reported die temperatures, but I have no idea how that translates into interface temps between the IHS and base of your HSF/waterblock. Also while Intel 7 CPUs may not have designs that lend themselves to hotspotting local to a particular area of logic (maybe? still looking for thermal maps of active Alder Lake or Raptor Lake dice), it should be noted that there will be hotter regions on a Raptor Lake die due to the P-cores and e-cores having different power draw and (presumably) heat densities. That may encourage pump-out. Some of my favorite TIM like Conductonaut and Kryonaut are subject to pump-out, especially if there is a large delta T between different areas of the thermal interface.

What kind of TIM are people here using for their Raptor Lake builds? I haven't seen a whole lot of commentary here. Pretty sure the ubiquitous MX-4 also suffers from pump-out. Supposedly MX-5 is less prone to pump-out and may be a better product for modern CPUs with highly-variable heat regions on the IHS.
Prolimatech PK-3 is my TIM of choice. I've been using it for years now.
 
  • Like
Reactions: igor_kavinski

Carfax83

Diamond Member
Nov 1, 2010
6,805
1,494
136
What kind of TIM are people here using for their Raptor Lake builds? I haven't seen a whole lot of commentary here. Pretty sure the ubiquitous MX-4 also suffers from pump-out. Supposedly MX-5 is less prone to pump-out and may be a better product for modern CPUs with highly-variable heat regions on the IHS.
I just use the good old NT-H1. No TIM crushes temps more than undervolting though :D
 

Carfax83

Diamond Member
Nov 1, 2010
6,805
1,494
136
What's the highest temp you have seen on your 13700KF?
Since I've undervolted my CPU down to -150mv, my temps in CBR23 and handbrake maxes out at 77c, and in games like Spider-Man Remastered which is very CPU intensive, I'm in the mid 40s to low 50s.

Undervolting crushes temps like nothing else for 13th gen because they are so overvolted at stock settings.
 
  • Like
Reactions: igor_kavinski

JTsyo

Lifer
Nov 18, 2007
11,392
566
126
Ordered the last of my parts over Black Friday.

PCPartPicker Part List: https://pcpartpicker.com/list/f6Gbfv

CPU: Intel Core i5-13600K 3.5 GHz 14-Core Processor
CPU Cooler: Thermalright Peerless Assassin 66.17 CFM CPU Cooler
Motherboard: ASRock Z690 Steel Legend ATX LGA1700 Motherboard
Memory: G.Skill Ripjaws V 32 GB (2 x 16 GB) DDR4-3600 CL16 Memory
Storage: Crucial P5 Plus 1 TB M.2-2280 PCIe 4.0 X4 NVME Solid State Drive
Case: Fractal Design Focus 2 ATX Mid Tower Case
Power Supply: Thermaltake Toughpower GF1 PE 650 W 80+ Gold Certified Fully Modular ATX Power Supply

GPU is 3060 that I had in my old computer.
 

Hulk

Diamond Member
Oct 9, 1999
3,695
1,299
136
Anyone able to get their 13900K to run at 5.8GHz during a CB R23 ST run? Should result in a score of 2250. I can't get better than 2150 for some reason, which means I'm at 5.5GHz?
 

Kocicak

Senior member
Jan 17, 2019
763
796
136
It helps to utilise the boost frequency better, while you allow only two particular best cores to run at 5.8, to allow total of three cores run at that frequency. Allowing just two limits the work the two boosting cores get.
 

Hulk

Diamond Member
Oct 9, 1999
3,695
1,299
136
It helps to utilise the boost frequency better, while you allow only two particular best cores to run at 5.8, to allow total of three cores run at that frequency. Allowing just two limits the work the two boosting cores get.
I'm sorry I'm not following?
 

Kocicak

Senior member
Jan 17, 2019
763
796
136
Ahh, lets say you have best cores 4 and 5, your per core limits look like: 55, 55, 55, 58, 58, 55, 55, 55

The core usage frequencies limits are 58, 58, 55, 55, 55, 55, 55, 55, but you set it 58, 58, 58, 55, 55, 55, 55, 55 instead, allowing 3 cores to run at that frequency, while only 2 will be able to run because of the first limitation.

It helps mainly with running 2 thread load, but it could have some effect on 1 thread load as well.
 

Hulk

Diamond Member
Oct 9, 1999
3,695
1,299
136
Ahh, lets say you have best cores 4 and 5, your per core limits look like: 55, 55, 55, 58, 58, 55, 55, 55

The core usage frequencies limits are 58, 58, 55, 55, 55, 55, 55, 55, but you set it 58, 58, 58, 55, 55, 55, 55, 55 instead, allowing 3 cores to run at that frequency, while only 2 will be able to run because of the first limitation.

It helps mainly with running 2 thread load, but it could have some effect on 1 thread load as well.
I see. But since I don't have any core limits set, it's on auto, shouldn't it just ramp up to 5.8 during the CB ST run?
 

Kocicak

Senior member
Jan 17, 2019
763
796
136
Yes, it should. It is probably some peculiarity in the bios, I have Gigabyte Z690 MB and it works as intended, except it lets the CPU run full blast (350W) without intervention. I dont know how Asus bios looks like/works. The Gigabyte bios unfortunatelly does not have the silicone quality rating in it.
 

Hulk

Diamond Member
Oct 9, 1999
3,695
1,299
136
I think a 2150 score is 5.8GHz. 2250 scores must be overclocks to 6GHz. I just locked the CPU at 4GHz and got a 1480 score. Assuming linear scaling, 2150 is correct for 5.8GHz.

Also when running ST I can see one of the cores is always (or almost) at 5.8GHz during the run.

My system is behaving as expected. I think the reviewers who pulled 2250 ST scores got some cherry chips and were using good WC, which allowed the cpu's to ramp up to 6GHz.
 
Last edited:

Carfax83

Diamond Member
Nov 1, 2010
6,805
1,494
136
I think 5.8ghz is achievable only when TVB is active, and TVB is active under certain conditions provide you have the cooling and power for it. I doubt you can hit 5.8ghz on air cooling at all though.

I just ran CBR23 single core and the score I got was 2043 at 5.2ghz. Hardly worth it considering how much more power I'd have to burn to get into the 2100+ range.
 

ASK THE COMMUNITY