Z790 Godlike - iGPU

[hasancankamilogl158d02e1]

Hello,

Could someone clarify whether using igpu from CPU is possible on z790 Godlike somehow. I am lost in specsheet and I really need to use my second monitor off the CPU.

 

[jeducb]

Hmm, that's odd. Can you make a screenshot of that BIOS page (either press F12, save to FAT32 drive and convert to PNG for lower size, or make a photo)?

Yes, that is odd. I've tried different ways, and played with other options to see if those two are enabled after change any others, but unsuccessfully.
Here you have the screenshot.

 

[citay]

So if you type "250" there on the numpad (with Numlock on), it won't take? How about when you press + or - there?

 

[jeducb]

I have a Godlike z790 and 13900ks, when you're saying you're struggling to keep the CPU cooled, what is the CPU temp you're having when running light to normal programs, and what's the temp when you run a load on all cores by using a 3d rendering app or some load test like Cinebench?

Well, when idle temps are around 38c~42c. When gaming I've noticed something like 65c~75c. When compiling C++ projects with Visual Studio temps go to 85c~92c (which seems too high for me just for a compiling task, but anyways). And running Cinebench temps reach 100c much before a single multithread test completes. Haven't tested it yet for Photoshop and Lightroom.

 

[jeducb]

So if you type "250" there on the numpad (with Numlock on), it won't take? How about when you press + or - there?

I feel myself so dumb now. I was expecting some dropdown with a list of pre-defined values, and when tried to type a number there didn't notice my Numlock was off by default ‍
It worked now. Will see what I get limiting those number.
Thanks for the help.

 

[h_abe159802e1]

the reading you gave should be ok. I run 3d program which build large world map terrain 180km/90km and it takes 12h to 20h to run each build for test using all cores and pushing my pc to Max more than what Cinebench can do.
My temp for 12 hours none stop goes between 80 to 102 , but 60% of the time its running on 99-100 (i seen few spikes going to 105 sometimes).
Same as you when i stop rendering and just browsing net, lots of tabs open + watching videos temp is 38-42, when doing light-medium 3d work between 50-80.
I contacted Intel and they advised to not underclock the CPU otherwise you can just go and buy a lower CPU for low temp and low performance, they said 13900ks is fine running on 100c when running heavy load as its build this way to reach high frequencies and nothing will happen to the CPU, as long as it doesn't run with high temp when idle or when doing light work you should be ok.

• Is It Bad If My Intel® Processor Frequently Approaches or Reaches Its Maximum Temperature?
• Information about Temperature for Intel® Processors

here are my temp during 3d workload with a 360 AIO liquid cooler

and here after stopping the workload

 

[citay]

Thanks for the help.

I contacted Intel and they advised to not underclock the CPU otherwise you can just go and buy a lower CPU for low temp and low performance

They completely misunderstand what we have done in this thread. We have not done underclocking, we have done a type of undervolting (lowering the V/F-curve to what the specific individual CPU actually needs). When you undervolt the right way, you can get even BETTER performance. Why? The CPU is constantly monitoring the frequency headroom for turbo boost and the power draw / temperature headroom. When it approaches a critical temperature (no temperature headroom left), it will initiate thermal throttling, as is evident on your screenshots. This is what lowers performance!

What happens when you undervolt the right way? All of a sudden, at the same frequencies, the temperatures are lower, because it uses less VCore to reach them. This means there is temperature headroom left and it doesn't need to clock itself down at all.

By properly undervolting and setting well-adjusted power limits in the BIOS, you can improve performance for long workloads. By staying away from approaching 100°C under fully multithreaded load, you also stay away from thermal throttling, which is the CPU trying to save itself from death. Yes, as Intel say, nothing will happen to the CPU (because the throttling is effective), but a lot can happen to your performance, depending on your cooling. So when i advise people here on the forum, i will always recommend to stay below 90°C if possible (to have some headroom for higher ambient temperatures), and avoid throttling territory at all cost.

With different CPU models and CPU cooling, lowering CPU Lite Load (stability-tested) can often already be enough to get everything under control. If it still reaches critical temperatures, i would advise to additionally set lower power limits, and monitor the temperature and performance impact with CB23, which is an example of fully multithreaded AVX load, the worst-case of normal workloads which aren't a full-blown artificial stress test like Prime95. Most of the time you will find that the performance hit from safe power limits is minimal, because unlike what Intel will have you believe, the CPU performance only scales quite minimally above a certain power draw, low single-digits. Just enough for Intel to have slightly longer benchmark bars than the rest in some launch reviews, in order to convince some potential buyers that this is "the best".

 

[h_abe159802e1]

What happens when you undervolt the right way? All of a sudden, at the same frequencies, the temperatures are lower, because it uses less VCore to reach them. This means there is temperature headroom left and it doesn't need to clock itself down at all.

Let me give this a try and see how it goes. so just need to change long duration power limit from Auto to 250?

Note I noticed the 5 cores that were flagged as thermal throttling were all running with constant boost 5600mhz at that time, usually when thermal throttling is flagged they should drop performance, but they were the one topping on the CPU boost, while other cores were not throttling there were running at 4300mhz during the workload

 

[jeducb]

View attachment 170788

392+ Watts... that's insane.
I've set:
CPU Lite Load to Mode 5
Long and Short Duration Power to 275

System is running very stable and no Temp Throttling, some PCores are still running at 6Ghz and GB23 is giving ~38.5k,which sounds good to me since I'm not trying to beat records but having system stable, efficient and as far as I can from temp throttling. I still have some room for Long and Short Duration Power, maybe 280 or 290. Will give them a try and see what happens.

By the way, as per MSI Support suggestion, the Enhanced Turbo is also disabled (see my BIOS screenshot in one of my previews post).

 

[h_abe159802e1]

System is running very stable and no Temp Throttling, some PCores are still running at 6Ghz and GB23 is giving ~38.5k,which sounds good to me since I'm not trying to beat records but having system stable, efficient and as far as I can from temp throttling. I still have some room for Long and Short Duration Power, maybe 280 or 290. Will give them a try and see what happens.

once you get a the perfect duration, let me know all setting you changed, then will run with it a 3d build and see how long it take to complete Vs old settings.

 

[jeducb]

once you get a the perfect duration, let me know all setting you changed, then will run with it a 3d build and see how long it take to complete Vs old settings.

Sure, I will.

 

[citay]

let me know all setting you changed

Note that the lowest stable "CPU Lite Load" mode is depending on the individual CPU, and thus has to be individually discovered via rigorous stress-testing.
I linked here how to do that. Basically, if it's Linpack- and Prime95-stable, there won't ever be a problem. For Prime95 i recommend "Small FFTs" torture test.

Similarly, the ideal power limits have to be tested individually, they depend on the actual resulting power draw and the cooling capabilities. I would aim to stay below 90°C, especially if you got very good CPU cooling with a large AIO, because it means that the power draw will already be very high (~250W). Above that the efficiency tends to drop pretty low anyway, because the power draw goes up way more steeply than the performance, so it becomes a losing game. If you don't want to break any performance records, then allowing close to 300W is usually not the thing to do.

By the way, for the sensors, i recommend HWinfo64 instead of HWmonitor.

 

[h_abe159802e1]

Note that the lowest stable "CPU Lite Load" mode is depending on the individual CPU, and thus has to be individually discovered via rigorous stress-testing.
I linked here how to do that. Basically, if it's Linpack- and Prime95-stable, there won't ever be a problem. For Prime95 i recommend "Small FFTs" torture test.

Similarly, the ideal power limits have to be tested individually, they depend on the actual resulting power draw and the cooling capabilities. I would aim to stay below 90°C, especially if you got very good CPU cooling with a large AIO, because it means that the power draw will already be very high (~250W). Above that the efficiency tends to drop pretty low anyway, because the power draw goes up way more steeply than the performance, so it becomes a losing game. If you don't want to break any performance records, then allowing close to 300W is usually not the thing to do.

On the Bios i only set XMP profile which correct the ram to the right value from 4800 to 5600
Everything else left to default

Now the changes i made is set
CPU Lite Load to Mode 5 (was set to auto before and looks like it was pointing to mode9)
Both Long and Short Duration Power to 275

but now when i run a small 3d project test build which push CPU to max more than Cinebench it take 13 min to build while previously was taking 12 min (means in large project taking like 15 hours there will be a 1 hour delay to complete the build) also cinebench score dropped by 1-2k.
The temp doenst go much higher than 88 and no throttling, sometimes some P cores can reach 6000mhz but for a very short period and they spend most of the time between 5400-5600 compared to my previous build while it was throttling but cores where constantly on 5600mhs 90% of the time.
What are the next power limit to attempt? something like both long and short to 290 and mode to 8 ?
also should CPU core voltage stay in auto or I limit it to 1.35?

this is results after power limit and mode changes

and this is result of the default auto settings which give more performance for now

 

[citay]

Please read carefully in this thread again, i already explained everything before. I will try to be a bit clearer about it all. "CPU Lite Load" and the power limits are two completely independent things. You have to find the ideal point for both, but finding them takes two completely different approaches.

"CPU Lite Load", in simple terms, lowers the VCore (it's a bit more complicated than that, but that's not important right now). You can imagine that when you lower VCore too much, there can be instability. So you want to lower it only to what your specific CPU needs to remain stable at all times. So how to find that ideal mode? See here. You run stability tests, i suggest Linpack Xtreme and also Prime95 Small FFTs. For a lot of CPUs, there will be a Lite Load mode that is too low, then you will get either errors in Linpack or Prime95, or both. Say Mode 3 has errors, then you test Mode 4. Say mode 4 has no errors in either stress test program. Then you would set Mode 5 (!). Why is that? We need to leave some headroom, so we go one mode higher than the first stable mode. We don't want to be on the edge of stability, we want to have a "stability buffer".

Now, you found that your performance was a bit worse than usual. This normally has nothing to do with "CPU Lite Load", so leave that at whatever final mode you have determined. This setting will not be changed after you have found the final mode. It's possible that your CPU is even stable in Mode 1, each CPU is completely individual, they come with different VIDs from factory (again, not important to understand). So test this, don't just take someone else's mode, they don't have your CPU.

The explanation lies in the power limits. It seems that for your specific workload, which seems to be unusually demanding, power limits of 275W are indeed limiting performance somewhat. But the good thing is, you have some cooling headroom left, however small, but it is there. And the most important thing, we need to keep this away from thermal throttling. Normal slight throttling because of power limits is always preferred over thermal throttling. So yes, raise the limits to 290W for testing (for now). You could potentially even go to 300-320W or so, let's see how your cooling performs. Leave CPU core voltage completely alone, you are already influencing that with CPU Lite Load.

 

[h_abe159802e1]

left mode 5 and changed Long and Short Duration Power to 290.
Small decrease by 10-15 seconds in my workflow and Icebench score went slightly up by 400. should I try 310 next?
no much changes in core temp, run most of the time on 85-86 and rarely reach 88 max

 

[citay]

Yeah go up to 310. Low 90s maximum temperature is also acceptable if performance for your workload is of paramount importance.

And try to find the ideal CPU Lite Load mode for your CPU.

 

[h_abe159802e1]

Thanks man i got now a stable load with same previous performance without any extreme heat or throttling issues.

I have noticed that increasing power limit does not influence the voltage used by the CPU, I have tried different power limits values and always same results, It seems CPU Lite Load govern power limit and not matter how much high you put it CPU will not reach it.
For example having CPU lite load on mode 5 with power limit 290 is same as power limit 390, CPU core will be limited between 1.22 and 1.27 volt no matter how much you increase the power limit to and max core temp will never be greater than 89C.
Since CPU lite load govern the power limit I left power limit to Auto, also tried different Modes and noticed the lowest the mode is the low voltage used and low CPU temp.
Mode 4 performance was lower than 5 and cpu voltage used was low, mode 5 still low performance to default setting. mode 6 better than 5, cpu voltage was slightly higher than 5 but temp still don't reach 90 and performance was nearly same as my default setting
Mode 7 turned to be the best for me giving slightly better performance than the default setting by few seconds during workload but good thing is no throttling at all and temp during workload are always between 80-93, rarely spike to 95-97 for a milliseconds then go back to 90.
Mode 9 was the one picked by default auto setting, mode 8 is nearly similar to 9, causing throttling and temp reaching 100.

only thing I need to find now is how to push those P cores to 5800MHz or 5900MHz during workload as it seems somewhere they are locked to 5600MHz and never goes beyond. remember seen someone here showing his cores reaching 5900MHz with stable heat

 

[citay]

only thing I need to find now is how to push those P cores to 5800MHz or 5900MHz during workload as it seems somewhere they are locked to 5600MHz and never goes beyond.

Fully multithreaded load is 5600 MHz as per Intel spec.

This is confirmed in practice:

So whatever they have done, it was probably manual OC, but you have no temperature headroom for that (it will enter thermal throttling again because of higher VCore required).

 

[hendershotspencer914f202d1]

Hello,

Could someone clarify whether using igpu from CPU is possible on z790 Godlike somehow. I am lost in specsheet and I really need to use my second monitor off the CPU.

Did this get answered am I still not getting it ? The motherboard has a display port 8k usb-c ? Can it use IGPU or not seems dumb that you wouldn’t be able to or be able to use intel quick sync ?

 

[citay]

See here and following. To my knowledge, the Z790 GODLIKE does not power the iGPU at all (no VRM phase for it), so nothing that depends on the iGPU will work.
So unless someone can prove otherwise, i would say, no, you cannot get an output from the iGPU, or even use QuickSync, which depends on the iGPU being active.