Intel 13th Gen CPU's

[demetri]

Below are the power levels for the CPU and GPU. I used HWINFO64 to get this data. I did not collect the data from the CPU cooler. It may be interesting to see what the cooler is doing.

For comparison, I've got a 12700K and a two-tower air cooler (Thermaltake Frost Spirit 140). Running CPU tests (Prime95) at ~215W constant load warms it up to ~85 degrees. But I do have a good airflow case with 2x140 intake Noctua fans and one 140 exhaust and I configured them to spin at max RPM (~1600) when the CPU or the GPU go over 75C.

 

[demetri]

Just a question.
I can see the CPU is watercooled.
And even that the max temp line reach 90 degr.
I wonder how high the temp reaches air cooled.

See above.
That 90 degrees could be a combination of a low-airflow case and a 450w RTX 4090 dumping all that heat into the case. FE version probably would've been better as it partially exhausts it outside

 

[Deleted member 197115]

I've had a couple of watercoolers for CPU's.
Both CPU's got cooler when I switched to a good air cooler.

 

[RCHeliguy]

I've had a couple of watercoolers for CPU's.
Both CPU's got cooler when I switched to a good air cooler.

If you look at many tests, a lot of the performance comes down to the combination of case and CPU cooling method.

My cavernous Torrent case has massive airflow and my Noctua CPU cooler has no issues with my i9-13900K what-so-ever even with a 4090 in the case with it.

Both stay pretty cool these days. I'll be curious to see how things look in the near future with a much harder load on them. I don't expect much will change.

 

[demetri]

If you look at many tests, a lot of the performance comes down to the combination of case and CPU cooling method.

My cavernous Torrent case has massive airflow and my Noctua CPU cooler has no issues with my i9-13900K what-so-ever even with a 4090 in the case with it.

Both stay pretty cool these days. I'll be curious to see how things look in the near future with a much harder load on them. I don't expect much will change.

You can run Furmark alongside Prime95 now and see how that future might look like

 

[R8pilot]

I presume you have all seen the 4 slot "Titan" photos?

Who will be the first on the block to own one?

(Oops, wrong forum, but RCHeliguy, I know you want to... )

 

[RCHeliguy]

I presume you have all seen the 4 slot "Titan" photos?

Who will be the first on the block to own one?

(Oops, wrong forum, but RCHeliguy, I know you want to... )

I haven't seen the pictures, sorry.

I just heard that ProSim is finally going to give us a time to ship the hydraulic lockout for which I've been on a waiting list for over 2 years.

 

[BillyBobSenna]

A little update on my computer temperatures. I ran Cinebench and my CPU temps jumped to 100°C immediately. What the heck! This surprised me and I thought something was definitely wrong. After a little Google searching, I learned I needed to change a bios setting. I needed to set “ASUS MultiCore Enhancement” to “Disable - Enforce All limits”. I also took this opportunity to install a CPU contact frame (see photo). In case you are not familiar with contact frames, the idea is to flatten the CPU and produce better interface with the cooler surface. As a result of the bios change and addition of a contact frame, my temperatures dropped ~10°C and the temperatures during Cinebench are much better.

It was interesting to see the GPU and CPU Package temps were almost exactly the same.

I also took a look at the air temperatures in my case. All the Corsairs fans in my case are set to "extreme". I have a probe to monitor the air temps coming into the case and into the CPU radiator. The air temps going into the radiator are about 10°C above inlet temps.

The CPU coolant temps stayed below 40°C however it looks like the coolant temps were continuing to rise even after 40 minutes of playing ACC.

At this point, I think I am happy with the thermals of my computer.

 

[Neilski]

After a little Google searching, I learned I needed to change a bios setting. I needed to set “ASUS MultiCore Enhancement” to “Disable - Enforce All limits”.

Did you originally change that setting yourself? I think the default would be off for the MCE.

This reminds me btw that some Asus BIOSes out there (including on my Z170-A board) are pretty damn stupid, and will happily jack up some voltages to insane and dangerous values if you switch them to XMP mode to get the best out of your RAM. Presumably it's limited to DIMMs that aren't explicitly recognised (though I'm not certain of this and the certified list is very short) but I am pretty certain that my own machine was damaged by the BIOS setting VCCIO and VCCSA much too high.
The galling thing is that if you attempt to raise those voltages manually the BIOS warns you by changing the colour of the text; I think it's yellow as you first get iffy, and then purple (?) when it's truly inadvisable, and IIRC both of my voltages were autoset into the purple zone ...

I have a vague memory that having the core CPU voltage too high (or at least needlessly high) is one of the side-effects of MCE, which might mean you can keep using it without getting too hot if you adjust the voltages.

 

[RasmusP]

Nice reports!

The air temps going into the radiator are about 10°C above inlet temps.

That does make a significant difference!
I'm yet to study some fluid dynamics etc, but I know that the temperature difference is a very important factor.

The radiator probably has a temperature of slightly below the package temp, so around 80°c.
The ambient air at the intake probably has around 20°c.
That's 60°c difference.

If the input air into the radiator is 10° higher, that delta shrinks to 50°.

That's 17% worse cooling.

Which is why it's "better" to place the CPU radiator into the front, if the cpu is the hotter part.

But that often raises the input temp at the GPU.
So if your graphics card is the hotter part, the radiator at the top is better.

To the mce:
It probably raised some voltages and took away some limits too, as Neil said.
Would be interesting how much performance you "lost" due to that?

13900k shooting into the temp limit with MCE without limits is "normal" btw.
Iirc Jay or Gamersnexus mentioned their 420 radiator on their open benchtable wasn't able to keep it from temp-throttling under full load on all cores!

 

[BillyBobSenna]

Nice reports!

That does make a significant difference!
I'm yet to study some fluid dynamics etc, but I know that the temperature difference is a very important factor.

The radiator probably has a temperature of slightly below the package temp, so around 80°c.
The ambient air at the intake probably has around 20°c.
That's 50°c difference.

If the input air into the radiator is 10° higher, that delta shrinks to 40°.

That's 20% worse cooling.

Which is why it's "better" to place the CPU radiator into the front, if the cpu is the hotter part.

But that often raises the input temp at the GPU.
So if your graphics card is the hotter part, the radiator at the top is better.

To the mce:
It probably raised some voltages and took away some limits too, as Neil said.
Would be interesting how much performance you "lost" due to that?

13900k shooting into the temp limit with MCE without limits is "normal" btw.
Iirc Jay or Gamersnexus mentioned their 420 radiator on their open benchtable wasn't able to keep it from temp-throttling under full load on all cores!

Thanks for the response. I'm not sure I understood all your comments. The highest air temperature in the case and going into at least the back part of the heat exchanger is ~30°C. This is about 10°C higher than would be the case if the radiator most mounted in the front.

 

[BillyBobSenna]

Did you originally change that setting yourself? I think the default would be off for the MCE.

By default, MCE was NOT off.

 

[BillyBobSenna]

Would be interesting how much performance you "lost" due to that?

I ran Cinebench with and without MCE disabled. With MCE disabled the Cinebench score was slightly lower. Only the order of %.

 

[RasmusP]

Thanks for the response. I'm not sure I understood all your comments. The highest air temperature in the case and going into at least the back part of the heat exchanger is ~30°C. This is about 10°C higher than would be the case if the radiator most mounted in the front.

Yep, exactly.
The cpu heats up the radiator (heat exchanger) but the radiator will always be a bit cooler than the cpu.

Cooling is based on the temperature difference and 10° can make a significant difference for PCs!

That's also why different air coolers get closer together, the higher the cpu temps become.
And why it's very difficult to cool CPUs a lot lower than the average of good air coolers achieve.

At 20 ambient and 50 cpu, the difference is 30.
Cooling the cpu down to 45 would make the difference become 25.

25 to 30 is 16.7% difference in cooling performance.

But having 20 ambient and 80 cpu, the difference is 60.
Cooling the cpu to 75 would mean a difference of 55.

55 instead of 60 is only 8.3% less difference.

I ran Cinebench with and without MCE disabled. With MCE disabled the Cinebench score was slightly lower. Only the order of %.

Yeah, I expected that.. Probably even less difference when gaming!

 

[Deleted member 197115]

By default, MCE was NOT off.

Default is Auto, which is ON as it gives 10-20% performance boost when multithreding.
And you cannot manually overclock if it's disabled.
But yes, not boosting other cores easier on thermals, but there is no free lunch, you might lose some performance.

 

[Sascha Reynders]

Replaced my mediocre bin (79 SP) 12900K with a 13900KS over the weekend, and so far I like what I'm seeing. Most BIOS settings are still on "Auto" (apart from LLC, SVID Behaviour and memory timings), and it's running stable on 100% load at 5.6/4.3 all-core on 1.296 vcore with LLC set to level 2. Max temp after 8 hours of OCCT CPU testing was 85c, averaging in the 65-70c range. In comparison to the 12900K, that's a 500mhz increase on the P-cores (400mhz E-cores) at lower vcore and with equal to better temps.
It also has a much stronger IMC: I can finally run my RAM at it's advertised speed of 4400CL17 in Gear 1 mode. Fastest the 12900K could manage in Gear 1 was 4000CL15.

Regarding MCE: Asus boards indeed default this to "Auto - Let's BIOS optimize", but I think on current platforms it mostly removes the power limits. I haven't seen much difference in voltages with it being on or off. What does give me better thermals though, is setting SVID Behaviour to "Typical Scenario" instead of "Auto", resulting in more reasonable voltages being requested by the CPU (about 0.05v lower on average), and therefore lower temps.

Now I've got some catching up to do on how to manually overclock Raptor Lake, to see if I can extract some more performance out of this beast

 

[RasmusP]

Nice write up!
About Asus boards:
When I searched for recommendations about overclocking/undervolting and memory tuning for my 7600X + B650 build, I've found a lot of reports about Asus boards pushing the SOC voltage (memory controller afaik) quite a bit too high on auto, especially if loading an Expo/xmp profile.

They fixed it with a Bios update within a few weeks/months. With 0804 or so.
On their site for the B650-A gaming wifi board, I can find 0613 from 2022/09/29 and 08009 from 2022/10/25.
These was, apparently, the crucial bios updating time and it were only 4 weeks for quite a few iterations.

Buildzoid however probably killed his 7950X due to this. He didn't check that voltage while pushing his RAM

 

[Sascha Reynders]

Nice write up!
About Asus boards:
When I searched for recommendations about overclocking/undervolting and memory tuning for my 7600X + B650 build, I've found a lot of reports about Asus boards pushing the SOC voltage (memory controller afaik) quite a bit too high on auto, especially if loading an Expo/xmp profile.

They fixed it with a Bios update within a few weeks/months. With 0804 or so.
On their site for the B650-A gaming wifi board, I can find 0613 from 2022/09/29 and 08009 from 2022/10/25.
These was, apparently, the crucial bios updating time and it were only 4 weeks for quite a few iterations.

Buildzoid however probably killed his 7950X due to this. He didn't check that voltage while pushing his RAM

Yeah, ASUS (but I guess other board makers as well) tend to go a little overboard sometimes with the default voltages they pump into the system when applying "overclocking" profiles such as MCE or XMP... That's why I always check all the important voltage readouts in HWInfo64 after booting a new build for the first time, to make sure there's no weirdness going on before I launch any kind of stress/stability test.

In my case, I was a bit worried about the VCCSA (system agent) and IVR Transmitter VDDQ (IMC, I believe) voltages that were applied at "Auto" settings: 1.328-1.341v and 1.350v respectively. On the 12900K I ran these at 1.250v (already in the yellow ) and 1.300v, but when I apply these voltages to the 13900KS it BSOD's the moment I hit the "start" button on the OCCT CPU test. According to Buildzoid, both should be "safe" up to 1.450v though, so I'm gonna trust him on that and leave them at "Auto". Looks like the new 13th Gen "Top Dog" likes some juice

EDIT: IVR VDDQ TX is Vdimm apparently, IMC voltage would be VDD2. Need to check that readout later on.
EDIT 2: nope, VDD2 seems to be a DDR5 specific voltage. According to Buildzoid -again- VDDQ/VDDQ TX actually is the memory controller voltage.

 

[RasmusP]

Yeah other brands do the same for sure!
It depends on every new chipset which manufacturers screw up and which don't...

I just wanted to pick a bit on Asus due to their mobos costing 30-50% more while often having less features...
You do get a nicer Bios in my experience with way better descriptions, overview etc., and with more expensive boards, you often get better OC stability too.

But their "budget" boards with some sweetspot OC'ing are too expensive imo, especially when they do screw up too from time to time

 

[Sascha Reynders]

I just wanted to pick a bit on Asus

I've been using Asus boards exclusively over the past 10 years, never had any complaints
I preferred the ROG Maximus boards (Code, Hero, etc) but when moving to Z690 I wanted to stay on DDR4, and the whole Maximus product line was DDR5 only. So I went with a lower-tier Strix Z690-A. Still a pretty good board. And I'm so used to Asus' BIOS, I feel no desire to start learning another manufacturer's one