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.

 

[citay]

Holy crap, 366W in CB23. Your AIO is impressive, almost being able to deal with that. So, you're even beyond the Intel-sanctioned "Extreme Power Delivery Profile" of 320W which would still be within Intel's warrany terms, but even allowing 320W would be too much. It only gives Intel a slightly longer result bar than AMD in some theoretical benchmark, but the efficiency is just not there anymore.

Another approach for undervolting is to use the CPU Lite Load setting. It lowers the voltage-per-frequency curve (VF curve), which is another method of undervolting, but i like to look at it as an optimization to the quality of your specific individual CPU. I would test with that as well. Leave the power limits maxed out for now, because it's interesting to see how the CPU behaves with a lower CPU Lite Load mode instead of conventional VCore undervolting. Then we can set a good power limit later.

Another thing, the Samsung 980 PRO had a firmware update last year through Samsung Magician which is quite important, and the 990 PRO is apparently gonna get one this February, so keep an eye out for that.

 

[busines157402e0]

Ok let's see the results now.
I've reset the VCore back to Auto.
Started Lite from Mode 1 and had crashes up to Mode 5, which went without a hitch
Then, as per your suggestion, I set Mode 6 and ran another 10min of CB23 Throttle Test.
Interestingly, on both Mode 5 and Mode 6 the CB23 score is extremely similar to when I undervolted -0.100

What do you make of the screenshots now?

Mode 5 (first stable CB23 10min Throttle Test):

Mode 6 (went one up from the lowest stable, as per your suggestion)

 

[IA64]

I don't know about your OC settings but I get around 42-43K in CB R23 with minimal OC in the bios ( temp of p-core 5 ~ 95c ) ... You should be getting 42K easily

13900K
MSI Coreliquid S280
64GB DDR5 6400MHz
Z790 ACE

 

[busines157402e0]

I don't know about your OC settings but I get around 42-43K in CB R23 with minimal OC in the bios ( temp of p-core 5 ~ 95c ) ... You should be getting 42K easily

View attachment 168796


13900K
MSI Coreliquid S280
64GB DDR5 6400MHz
Z790 ACE

I did no OC-ing at all.

Could you share your BIOS tweaks?

 

[citay]

Your -0.1V undervolt seems to fall in between CPU Lite Load Mode 5 and Mode 6. While the -0.1V shows a higher peak "CPU package power" than even Mode 6, it might've really been just a short peak. Because when we look at what the PWM controller of the VRM is reporting (Renesas RAA229131, just below the board's Nuvoton in the second column) for Current OUT and Power OUT, which is the CPU-facing side of the VRM, it matches up better with the temperatures we're seeing. So i'm thinking that's a more correct measure of the averages.

Therefore, if you thoroughly test the stability, both methods are suitable. Just make sure to leave a similar headroom with normal undervolting. It wouldn't be a fair comparison if you went all the way to the edge on the undervolt vs. leaving more headroom on the Lite Load setting.

As for the power limits, it depends, if you want to have some efficiency left, then i would set the standard power limits of 253W for both. If you only want to prioritize performance, then whatever it takes to stay below 90°C no matter the workload, maybe like 280-300W.

I don't know about your OC settings but I get around 42-43K in CB R23 with minimal OC in the bios ... You should be getting 42K easily

Again, it comes down to the efficiency and the cooling capabilities. He has the -KS model, it already goes into throttling without any OC applied. Also, the 13900KS is less of a step up from the 13900K than the 12900KS was from the 12900K (clocks-wise). The 12900KS could eke out another 300 MHz over the 12900K in gaming and light loads / single-core. The 13900KS is only good for 66 MHz (Multi) to 200 MHz (Single) more, partly because it is also limited by the temperatures causing CPU throttling.

What's more, it depends on how you overclock. With traditional OC (focused on all-core OC), the scores in full-multithreaded-load benchmarks like CB23 Multi may look a bit better (at the expense of much higher power consumption and thus worse efficiency), but usually you can't gain much in real-world scenarios where the normal turbos already do a great job of boosting the clocks. Depending on the type of OC, it can mainly be good for raising numbers a little in benchmarks that use somewhat artificial all-core load, and in other scenarios it may actually underperform, when it has to settle for a lower all-core OC than the normal turbo boost would allow for up to four cores loaded for example.

So with any OC, you have to know what you're doing, and not just search a higher CB23 Multi score which might not be representative for your daily workloads.

 

[busines157402e0]

Your -0.1V undervolt seems to fall in between CPU Lite Load Mode 5 and Mode 6. While the -0.1V shows a higher peak "CPU package power" than even Mode 6, it might've really been just a short peak. Because when we look at what the PWM controller of the VRM is reporting (Renesas RAA229131, just below the board's Nuvoton in the second column) for Current OUT and Power OUT, which is the CPU-facing side of the VRM, it matches up better with the temperatures we're seeing. So i'm thinking that's a more correct measure of the averages.

Therefore, if you thoroughly test the stability, both methods are suitable. Just make sure to leave a similar headroom with normal undervolting. It wouldn't be a fair comparison if you went all the way to the edge on the undervolt vs. leaving more headroom on the Lite Load setting.

As for the power limits, it depends, if you want to have some efficiency left, then i would set the standard power limits of 253W for both. If you only want to prioritize performance, then whatever it takes to stay below 90°C no matter the workload, maybe like 280-300W.



Again, it comes down to the efficiency and the cooling capabilities. He has the -KS model, it already goes into throttling without any OC applied. Also, the 13900KS is less of a step up from the 13900K than the 12900KS was from the 12900K (clocks-wise). The 12900KS could eke out another 300 MHz over the 12900K in gaming and light loads / single-core. The 13900KS is only good for 66 MHz (Multi) to 200 MHz (Single) more, partly because it is also limited by the temperatures causing CPU throttling.

What's more, it depends on how you overclock. With traditional OC (focused on all-core OC), the scores in full-multithreaded-load benchmarks like CB23 Multi may look a bit better (at the expense of much higher power consumption and thus worse efficiency), but usually you can't gain much in real-world scenarios where the normal turbos already do a great job of boosting the clocks. Depending on the type of OC, it can mainly be good for raising numbers a little in benchmarks that use somewhat artificial all-core load, and in other scenarios it may actually underperform, when it has to settle for a lower all-core OC than the normal turbo boost would allow for up to four cores loaded for example.

So with any OC, you have to know what you're doing, and not just search a higher CB23 Multi score which might not be representative for your daily workloads.

Could you point me to the BIOS settings that I should be tweaking to lower the power limit? A screenshot would be much appreciated.

 

[citay]

They are listed here under OC - "Advanced CPU Configuration" in the BIOS, Long and Short duration power limit just above CPU Lite Load:

So you could set Long and Short Duration Power Limit to 253W there for a test, just to see if it impacts the score a lot, and of course how the temperatures are. I would also set "C1E Support" to Enabled.

We can assume it affects the CB23 score a little though, because that uses fully multithreaded AVX load which is known to be very demanding. That doesn't have to mean that it affects your daily tasks though, if they make the CPU work below or around that level of power consumption anyway.

 

[busines157402e0]

They are listed here under OC - "Advanced CPU Configuration" in the BIOS, Long and Short duration power limit just above CPU Lite Load:

So you could set Long and Short Duration Power Limit to 253W there for a test, just to see if it impacts the score a lot, and of course how the temperatures are. I would also set "C1E Support" to Enabled.

We can assume it affects the CB23 score a little though, because that uses fully multithreaded AVX load which is known to be very demanding. That doesn't have to mean that it affects your daily tasks though, if they make the CPU work below or around that level of power consumption anyway.

And I should revert the CPU Lite Load to "Auto", yes?

 

[citay]

No. The power limits are independent of all that. They only cap off the allowed power consumption at the Wattage you set, they don't lower VCore or anything by themselves.
So you still have to keep the CPU Lite Load going on Mode 6 for example. If you were to set it back to Auto, it would use Mode 12 again or whatever the default is. Meaning a considerably higher VCore again, which would make the CPU throttle harder when it reaches the power limits, just because it now uses a higher voltage for a given frequency. So the less CPU voltage you can get away with, the less throttling you will see.

Note that you will see throttling when setting lower power limits, that's normal. But it is then no longer temperature-induced throttling, but simply power-draw-induced, as we want to limit that to a sensible level.

 

[IA64]

Again, it comes down to the efficiency and the cooling capabilities. He has the -KS model, it already goes into throttling without any OC applied. Also, the 13900KS is less of a step up from the 13900K than the 12900KS was from the 12900K (clocks-wise). The 12900KS could eke out another 300 MHz over the 12900K in gaming and light loads / single-core. The 13900KS is only good for 66 MHz (Multi) to 200 MHz (Single) more, partly because it is also limited by the temperatures causing CPU throttling.

What's more, it depends on how you overclock. With traditional OC (focused on all-core OC), the scores in full-multithreaded-load benchmarks like CB23 Multi may look a bit better (at the expense of much higher power consumption and thus worse efficiency), but usually you can't gain much in real-world scenarios where the normal turbos already do a great job of boosting the clocks. Depending on the type of OC, it can mainly be good for raising numbers a little in benchmarks that use somewhat artificial all-core load, and in other scenarios it may actually underperform, when it has to settle for a lower all-core OC than the normal turbo boost would allow for up to four cores loaded for example.

So with any OC, you have to know what you're doing, and not just search a higher CB23 Multi score which might not be representative for your daily workloads.

For some reason I thought he had the KF KS is still not available in my region hence the confusion. Sorry !

My OC is very basic ( Turbo offset + 2 and E-core 4.5 / 4.5 / 4.6 / 4.6 )

Temps under control and 2% difference in single load / 6-7% in multicore

 

[citay]

2% single-core makes sense for 200 MHz more, this is about the same as the -KS gets over the -K (at 20-25% higher price). CB23 really seems to benefit from your higher E-core clocks, which are of course still nailed to 4.3 GHz on the K and KS. By how much did your power consumption go up in CB23 Multi?

 

[IA64]

2% single-core makes sense for 200 MHz more, this is about the same as the -KS gets over the -K (at 20-25% higher price). CB23 really seems to benefit from your higher E-core clocks, which are of course still nailed to 4.3 GHz on the K and KS. By how much did your power consumption go up in CB23 Multi?

Ahem... 362 Watts I guess

 

[citay]

Man, those power consumption numbers you guys come up with... crazy. Ok, yeah, so you're trading any kind of efficiency for higher performance. That's a valid approach, if you don't care about energy spent for completion of a workload. I have a bit of an opposite approach, i primarily want to maximize the efficiency while only keeping stock performance (which i know is already plenty enough).

 

[IA64]

Man, those power consumption numbers you guys come up with... crazy. Ok, yeah, so you're trading any kind of efficiency for higher performance. That's a valid approach, if you don't care about energy spent for completion of a workload. I have a bit of an opposite approach, i primarily want to maximize the efficiency while only keeping stock performance (which i know is already plenty enough).

You are absolutely right; I do care about power efficiency hence the " balanced " power mode in windows and tbh I rarely put the system under full load.

What I care about the most is single thread performance but going above 6GHz is very difficult for what I can get in return.

Actually I'm not into the whole overlocking thing... never had any interest in it, never spent time on it so my experience is very limited in this area.

 

[busines157402e0]

Man, those power consumption numbers you guys come up with... crazy. Ok, yeah, so you're trading any kind of efficiency for higher performance. That's a valid approach, if you don't care about energy spent for completion of a workload. I have a bit of an opposite approach, i primarily want to maximize the efficiency while only keeping stock performance (which i know is already plenty enough).

I agree, my concern is to keep the CPU cool and consistent, as I can't afford disruptions in my workflow. I have zero overclocking experience and was shocked at my KS drawing 366W peak.

The CPU Lite Modes are, for some reason, finicky for my CPU/MB, the temps just get too high for the performance lost - maybe it's the "XMP Profile 1" pushing the 128GB RAM to 5200mhz(?) At any rate, I reverted back to Offset VCore -0.100 and, as per your suggestions, I limited power draw, although to 270W. Temps are nice and cool, hitting 90C max in only a few instances and never going above that during the CB23's 10min Throttle Test, which now sits at 40098 pts. Maybe I'll do tests at 280W to 290W, but overall I am happy with the 270W-limit results; pretty significant power draw reduction, in my opinion, and still getting nice juice out of the 13900KS; must say I'm warming up to the KS (no pun intended), which I force-bought in the first place, as the K wasn't available where I am, but the LF II 420mm has certainly played its part too.

Cheers for all your help Citay, not only was your advice insightful but, most importantly, you saved me a lot of time! <3

I'll circle back with any updates/questions! Rock on!

 

[citay]

You're welcome, always glad to help. Don't put too much emphasis on Cinebench results, mainly look at your cooling. I always try to keep people below 90°C so there are reserves in the tank for higher ambient temperatures in the summer etc. Plus it's only meant as a top-end limiter, many other workloads won't stress the CPU to that extent.

Let's crunch some numbers just for fun. You are getting 40100 points in CB23 multi at 270W (power limited). Efficiency = 148 points per Watt.
IA64 is getting 42400 points with OC, at 360W unlimited. That's a 5.7% higher score at 33% higher power draw. Efficiency = 118 points per Watt.

Efficiency would continue to improve with lower power limits, showing there are diminishing returns for allowing this high of a power draw, and proving Intel have only done it to reach a certain performance target (efficiency be damned). But AMD are also guilty of this on their top-end CPU models. The same happens on the GPU market with NVIDIA vs. AMD. It seems that most buyers are fixated on a tiny lead in the launch day reviews to make up their mind about a product, hopefully we'll get away from that a bit. There's more to a GPU/CPU than just the raw performance. If we keep prioritizing performance over everything, we're gonna keep having hot-headed, inefficient, borderline-uncoolable products as a result.

 

[busines157402e0]

You're welcome, always glad to help. Don't put too much emphasis on Cinebench results, mainly look at your cooling. I always try to keep people below 90°C so there are reserves in the tank for higher ambient temperatures in the summer etc. Plus it's only meant as a top-end limiter, many other workloads won't stress the CPU to that extent.

Ah yes, I've accounted for this (I think?). My apartment is very warm, with wild central heating (for context, even during the December/January period I have to strip to shorts and t-shirt unless I air the place a few times per day), and the new 13th gen PC is temporarily right next to the central-heating wall radiator in my office (I deliberately did this).

During initial tests I aired the room to lower ambient temperature, but after I gained stability I kept the balcony door closed and, predictably, the minimum CPU temps went up 5-7 degrees, which in my case is literally equal to increases I get during mid-summer.

Still remains to be seen, but I think the temps I got with your help will hold up during the hottest days.

Let's crunch some numbers just for fun. You are getting 40100 points in CB23 multi at 270W (power limited). Efficiency = 148 points per Watt.
IA64 is getting 42400 points with OC, at 360W unlimited. That's a 5.7% higher score at 33% higher power draw. Efficiency = 118 points per Watt.

Efficiency would continue to improve with lower power limits, showing there are diminishing returns for allowing this high of a power draw, and proving Intel have only done it to reach a certain performance target (efficiency be damned). But AMD are also guilty of this on their top-end CPU models. The same happens on the GPU market with NVIDIA vs. AMD. It seems that most buyers are fixated on a tiny lead in the launch day reviews to make up their mind about a product, hopefully we'll get away from that a bit. There's more to a GPU/CPU than just the raw performance. If we keep prioritizing performance over everything, we're gonna keep having hot-headed, inefficient, borderline-uncoolable products as a result.

Now that is very interesting! I don't really ever look at things from the angle of per-Watt performance, my main concern being keeping things cool enough and consistent, but I must say that I was pleasantly surprised by the scores I got in CB23, because given the 270W power limit I was expecting something in the range of 38k to 39k, but whoa it's consistently at ~40.1k. Reasoning for my excitement is that animation rendering can keep the CPU under full load for prolonged periods, so if the 270W CB23 temps, stability, and performance are anything to go by, I'm golden!

 

[cagrgoncule15c802e9]

Hi there i am planning to build a system l down below -some parts already ordered:

• CPU: i9-13900KS
• RAM: G.Skill Trident Z5 RGB 32 GB (2x16) 7800 MHz DDR5 CL36 F5-7800J3646H16GX2-TZ5RK
• Motherboard: MSI MEG Godlike z790
• Cooling: CORSAIR iCUE H170i ELITE LCD Display Liquid CPU Cooler 420 mm
• GPU: GIGABYTE GeForce RTX 4090 Aorus Master OC 24GB
• PSU: THERMALTAKE TOUGHPOWER GF3 1650W GOLD 80PLUS
• M.2: 1 x Corsair MP600 Pro 2TB 7000MB-6550MB/s nVME M.2 SSD

There is a program called HQPlayer (audiophile grade upscale software) that i want to run on this system

Link HQPlayer 4 Desktop on Windows:
https://www.signalyst.com/consumer.html
Down below that page is the trial download link:
https://www.signalyst.com/bins/HQPlayer4Desktop-setup-4211 .exe?

The problem is this program HQPlayer uses AVX2 (Advanced Vector Extensions)
i9-13900KS supports AVX2 but z790 Godlike doesnt support igpu (integrated graphics )
Some say i9-KS series can run igpu on a Z790 Msi Godlike motherboard - not like KF series but i am not sure
Does this not supporting igpu affects using AVX2 extensions

Could someo who has a similar systembuild - i mean same processor i9 13900 KS and same motherboard Msi z790 Godlike
do me a favour and install that small ca 170 mb HQplayer trial software and run the program
If the program starts up and runs then everything is ok
If there is no AVX2 support - then the program doesn't start closes down and nothing happens

Thanks in advance and best regards -

 

[IA64]

Why do you want to get the Godlike may I know? complete waste of money. Even the ACE is overkill but it's your money anyway....

As far as I know, the motherboard can't handle two GPUs at the same time so if 4090 is active, you won't be able to use any of the iGPU features.


Edit :

Looks like it's opening on my PC :

4090 Suprim X - 13900K - Z790 ACE

 

[cagrgoncule15c802e9]

Hi again its client version
i want to use desktop version to be tested please check again