Question 13900K, KF, and KS Undervolting wizards... What is your methodology?

[Hulk]

I have a 13900K and have noticed that many others with the same CPU are getting much better power/thermal results than I am at equivalent frequency/performance. I realize that there is a difference in the quality of the silicon of each CPU but that being said I'd like to learn more about how you go about undervolting? The reason I have specifically mentioned the 13900K series is because unless you have some type of extreme cooling most people are generally performance limited by heat/temperature. But some around here are getting much better results than others, like myself. Anyway, onto the specifics of my question.

Someone hands you a 13900K, KF, or KS, in a step-by-step analysis how do you go about optimizing for a given cooling solution? Let's say it's Noctua U12A air cooling in a case than can effectively deal with about 200-225W of heat.

 

[Hulk]

I think CPUz and HWinfo are reporting the same Vcore values. I believe the difference is that CPUz has a longer polling time than default the default setting for HWinfo. So the instant in time the voltage is recorded rarely coincides exactly.

 

[Carfax83]

Jeez, those are some great numbers. I'll post mine for comparison purposes. I don't see Vcore listed in your HWinfo? I hate to keep bothering you but what was that reading as compared to CPUz?

I just opened up HWinfo and I don't see Vcore anywhere. It only shows the VID.

That's why I use CPU-Z for the Vcore.

 

[Carfax83]

I think CPUz and HWinfo are reporting the same Vcore values. I believe the difference is that CPUz has a longer polling time than default the default setting for HWinfo. So the instant in time the voltage is recorded rarely coincides exactly.

I don't know about this. Whenever I mess with the voltage, HWinfo does not register the change so I think it's strictly only VID that it reports.

 

[Hulk]

I don't know about this. Whenever I mess with the voltage, HWinfo does not register the change so I think it's strictly only VID that it reports.

I noticed in your screenshot that Vcore isn't indicated under your motherboard in HWinfo. That is where Vcore should be and it seems to correlate well with CPUz.

 

[alcoholbob]

Definitely wild how wide voltages are with different RPL chips. Some chips can go down quite a bit and others are only stable with nuclear voltage levels.

 

[Rigg]

@Carfax83 Click the button with the arrows pointing out to open up more columns in HWI64. Since you appear to running your 4k desktop without any scaling there is enough room to display all relevant sensor data in a screenshot. The Vcore is displayed under the motherboard sensors.The Vcore reading in CPUZ should align with the Vcore reading from the mobo in HWI64. If you hit the settings button you can change the polling rate. The minimum is 20 ms.

Something seems weird here. Your SVID seems incredibly low for 5.5 ghz. I wouldn't think it would be that low without intervention on your part. A Vcore offset won't affect the SVID. That being said, the Vcore displayed in CPUZ does directly correlate to the 115 mV offset you set against the SVID in HWI64. Which is also kind of weird since it is usually only that close at idle and almost never exact like yours is.

5.5 @ 1.163 v pcore is crazy good compared to my 13700k. I need about 165 mV more than that to hit 5.5 on my P cores. Can't wait to see a proper screenshot. Lottery indeed. You should really stop underclocking this thing if it's this good.

 

[Carfax83]

I noticed in your screenshot that Vcore isn't indicated under your motherboard in HWinfo. That is where Vcore should be and it seems to correlate well with CPUz.

Yeah you were right, it does correlate with CPUz:

 

[Carfax83]

Something seems weird here. Your SVID seems incredibly low for 5.5 ghz. I wouldn't think it would be that low without intervention on your part. A Vcore offset won't affect the SVID. That being said, the Vcore displayed in CPUZ does directly correlate to the 115 mV offset you set against the SVID in HWI64. Which is also kind of weird since it is usually only that close at idle and almost never exact like yours is.

What kind of intervention could change the SVID? Isn't that programmed into the CPU itself from Intel?

5.5 @ 1.163 v pcore is crazy good compared to my 13700k. I need about 165 mV more than that to hit 5.5 on my P cores. Can't wait to see a proper screenshot. Lottery indeed. You should really stop underclocking this thing if it's this good.

But the thing is that it requires a massive jump in power to achieve 5.5ghz, compared to 5.3ghz. My CPU will do 5.3ghz at 235w under stress, but needs around 270ish at 5.5ghz on my current cooling. To me it's not worth the extra power for a performance increase that won't be noticeable to me.

If I had better cooling I would probably do it though. With better cooling the CPU wouldn't need as much wattage to hit 5.5ghz and would probably be in the upper 70s. A good AiO would be best, but I don't trust those things. Heard a lot of horror stories about leaking and pump failures etcetera. Custom loop is also too much work and maintenance for a guy like me.

I'm definitely a fan of air cooling due to its simplicity and reliability. The most I see myself doing in the future is getting Noctua's NH-D15 successor, but they keep delaying it. Or, I may just do what @Herald did and buy two T30 Phanteks which move way more air than the NH-U12a stock fans.

 

[JoeRambo]

There are two Vcore offset modes:

1) Vcore Offset mode is applied at VRM level, basically if CPU is asking for 1.3V, voltage will get lowered, VID as reported by CPU will remain the same, while Vcore is obviously lower
2) VID Offset mode is applied to the VID the CPU is asking ( and thre might be separate VID offset for uncore ). VID is reported as that new value.

1.163V for 5.5 is good chip for sure.

 

[Rigg]

What kind of intervention could change the SVID? Isn't that programmed into the CPU itself from Intel?

You can do it 3 different ways. There is a drop down for SVID behavior near the top of the AI tweaker tab with 5 options. I suspect that because your mobo exposes the SP rating it may set this to "best case scenario" by default on auto. Your SVID aligns with mine when I set it to "best case scenario" and set my core multiplier to 55. It could also be a 13900k vs. 13700k thing.

Right below the Vcore offset in BIOS there is an SVID offset. I haven't played with it much. The smallest SVID offset I can set on my z690 Tuf is 100 mV.

Using the V/F offsets also offsets SVID instead of Vcore.

But the thing is that it requires a massive jump in power to achieve 5.5ghz, compared to 5.3ghz. My CPU will do 5.3ghz at 235w under stress, but needs around 270ish at 5.5ghz on my current cooling. To me it's not worth the extra power for a performance increase that won't be noticeable to me.

Why you gotta be all practical? This triggers me. That CPU is begging to be overclocked.

If I had better cooling I would probably do it though. With better cooling the CPU wouldn't need as much wattage to hit 5.5ghz and would probably be in the upper 70s. A good AiO would be best, but I don't trust those things. Heard a lot of horror stories about leaking and pump failures etcetera. Custom loop is also too much work and maintenance for a guy like me.

I'm definitely a fan of air cooling due to its simplicity and reliability. The most I see myself doing in the future is getting Noctua's NH-D15 successor, but they keep delaying it. Or, I may just do what @Herald did and buy two T30 Phanteks which move way more air than the NH-U12a stock fans.

Custom water cooling is fun but it's a complete waste of money. I think I prefer air too. I need to figure out what to do with my old water cooled rig. I switch hardware too often and I found it too annoying and expensive to upgrade. I sold the 3900x and x570 crosshair but i still have the radeon VII, case, PSU, and water cooling stuff. I also have a hero dark, 32 gb of good B-die, 2tb 960 pro, and 5800x laying around but I think I want to sell it and get an AM5 setup to throw together with the Radeon VII and water cooling stuff. As expensive as the hardware has gotten I'm not sure its worth the leak/failure risk anymore for me. On the other hand I'd rather use it then sell it for penny's on the dollar. Anyway, enough off topic.

I saw an optimum tech vid on youtube where he did some testing with the T30's and they looked pretty good against Noctua's.

 

[Just Benching]

The FC140 is the best air cooler you can buy right now, if you care about upgrading. Was thinking about it but it's kind of a waste for me, the u12 is already superb for my needs

 

[Hulk]

What kind of intervention could change the SVID? Isn't that programmed into the CPU itself from Intel?



But the thing is that it requires a massive jump in power to achieve 5.5ghz, compared to 5.3ghz. My CPU will do 5.3ghz at 235w under stress, but needs around 270ish at 5.5ghz on my current cooling. To me it's not worth the extra power for a performance increase that won't be noticeable to me.

If I had better cooling I would probably do it though. With better cooling the CPU wouldn't need as much wattage to hit 5.5ghz and would probably be in the upper 70s. A good AiO would be best, but I don't trust those things. Heard a lot of horror stories about leaking and pump failures etcetera. Custom loop is also too much work and maintenance for a guy like me.

I'm definitely a fan of air cooling due to its simplicity and reliability. The most I see myself doing in the future is getting Noctua's NH-D15 successor, but they keep delaying it. Or, I may just do what @Herald did and buy two T30 Phanteks which move way more air than the NH-U12a stock fans.

Here's the thing. I wouldn't limit the frequency, limit the temp. There are a lot of loads that will run up to 5.5GHz under much less power than CB. They won't activate all of the structures in the core so temp and power will remain in check but you will get the benefit of 5.5GHz.

Temp limiting is the only way to do this. If you try by cores, way 5.5GHz with 4 cores, etc... there is always some background process grabbing a thread and it will prevent the cores in use from hitting 5.5GHz. Been there tried that.

Water is not as scary as you imagine. I thought the same way until my 280mm AIO dropped to $60 on Amazon and I picked it up. Install was easy as air, it's quiet, and moves more heat than my air. Plus if it does fail I'll know immediately as the CPU will be throttling like mad as I'm set for a max temp of 85C. Now if I get a leak next week I could change my tune!

Your CPU is great. Mine must have just squeaked over the 13900K bar. It won't even run stock without 120% from the power supply in the BIOS, which is why I requested a new one from Intel and they are sending it. It would run 100% when I first bought it and I never overclocked it. For the record in 30 years of buying CPU's I've never returned one. But I paid a lot for this one and it should as least perform to spec with BIOS default settings I think.

Anyway the testing of CPU's for the VID programming is a tightly guarded secret. I have no idea where in the production process it happens but I would guess they do a really quick "super secret" test on it that pretty much instantly spits out a v/f curve to be programmed into the cpu before the lid goes on.

Two lines of thinking of your CPU. One would be overclock its brains out since it looks like a good one. Slap a 420 AIO on it and let it rip.

The other would be to do exactly what you're doing and get 95% of the performance while using less power, generating less heat, and extending the life of the CPU.

 

[Rigg]

Here's the thing. I wouldn't limit the frequency, limit the temp. There are a lot of loads that will run up to 5.5GHz under much less power than CB. They won't activate all of the structures in the core so temp and power will remain in check but you will get the benefit of 5.5GHz.

Temp limiting is the only way to do this. If you try by cores, way 5.5GHz with 4 cores, etc... there is always some background process grabbing a thread and it will prevent the cores in use from hitting 5.5GHz. Been there tried that.

I currently have mine setup to turbo to 5.8 when 1-2 cores are being used. I have it set to 5.6 for 3-8 cores. I have TVB set to down clock to 5.5 @ 75c and 5.3 @ 84c when 8 cores are used. I have an 85c temp limit. It turbos to 58 steady in CB23 single core. It holds steady at 5.6 in all the games I've tried with temps in the 60's. If I run CB23 multi or real bench it oscillates between 5-5.3 (when actually loaded) as cores hit the thermal limit. Voltage ranges from 1.375 - 1.2 depending on load.

 

[Hulk]

I currently have mine setup to turbo to 5.8 when 1-2 cores are being used. I have it set to 5.6 for 3-8 cores. I have TVB set to down clock to 5.5 @ 75c and 5.3 @ 84c when 8 cores are used. I have an 85c temp limit. It turbos to 58 steady in CB23 single core. It holds steady at 5.6 in all the games I've tried with temps in the 60's. If I run CB23 multi or real bench it oscillates between 5-5.3 (when actually loaded) as cores hit the thermal limit. Voltage ranges from 1.375 - 1.2 depending on load.

Very nice. Do you have a board that will tell you the SP rating of your CPU?

 

[Rigg]

Very nice. Do you have a board that will tell you the SP rating of your CPU?

Nope. It's definitely worse than @Carfax83 's CPU. 165 mV at 5.5 ghz difference seems like a lot. I used a bit of voltage manipulation to get there.

 

[coercitiv]

What kind of intervention could change the SVID? Isn't that programmed into the CPU itself from Intel?

VIDs change based on a number of factors, the programmed V/F curve is just a baseline. Watch this.

 

[JoeRambo]

VIDs change based on a number of factors, the programmed V/F curve is just a baseline. Watch this.

+1 for ScatterBencher videos, he has amazing deep dive stuff for Intel and AMD.

 

[Carfax83]

Why you gotta be all practical? This triggers me. That CPU is begging to be overclocked.

Modern CPUs have made overclocking far less tempting than it was in the past. The CPUs come straight out of the factory with a very generous turbo boost, and if you want more Mhz, you better have damned good cooling.

I remember the old days when you could get nearly 1ghz extra out of a CPU on top of factory clocks..

I saw an optimum tech vid on youtube where he did some testing with the T30's and they looked pretty good against Noctua's.

I bought a triple pack from Amazon yesterday, and it should be here tomorrow so I'll let you guys know how much better they are compared to my stock fans. If I can drop even 3c, that would be great!

 

[Carfax83]

Here's the thing. I wouldn't limit the frequency, limit the temp. There are a lot of loads that will run up to 5.5GHz under much less power than CB. They won't activate all of the structures in the core so temp and power will remain in check but you will get the benefit of 5.5GHz.

Temp limiting is the only way to do this. If you try by cores, way 5.5GHz with 4 cores, etc... there is always some background process grabbing a thread and it will prevent the cores in use from hitting 5.5GHz. Been there tried that.

I ordered a triple pack of Phantek T30 120mm fans, and these fans are supposed to be the best for air coolers. If they can shave off a few degrees, I will likely do the temp limiting thing permanently.

 

[Carfax83]

The FC140 is the best air cooler you can buy right now, if you care about upgrading. Was thinking about it but it's kind of a waste for me, the u12 is already superb for my needs

I ordered the Phantek T30 120mm fans. Any tips or suggestions before I install them? They should be arriving tomorrow unless Amazon screws something up.

 

[Just Benching]

I ordered the Phantek T30 120mm fans. Any tips or suggestions before I install them? They should be arriving tomorrow unless Amazon screws something up.

Yeah, put an anchor on top of your case so it doesn't fly away by all the airflow

 

[Carfax83]

Yeah, put an anchor on top of your case so it doesn't fly away by all the airflow

Man, you weren't kidding! They sound like turbines when they spool up! They move a shed load of air though, and now as a result of switching to these fans, stock clocks are far more stable. I can maintain 5.5ghz on the P cores and 4.3ghz on the E cores for about 4 CBR23 cycles before they start to throttle, which is twice that of the Noctua fans. And when they do throttle, only some of the cores are dropping to 5.4ghz while the others stay at 5.5ghz. With the Noctua fans, all of the cores would start to throttle after a while and drop to 5.4 and 5.3ghz.

 

[Just Benching]

Man, you weren't kidding! They sound like turbines when they spool up! They move a shed load of air though, and now as a result of switching to these fans, stock clocks are far more stable. I can maintain 5.5ghz on the P cores and 4.3ghz on the E cores for about 4 CBR23 cycles before they start to throttle, which is twice that of the Noctua fans. And when they do throttle, only some of the cores are dropping to 5.4ghz while the others stay at 5.5ghz. With the Noctua fans, all of the cores would start to throttle after a while and drop to 5.4 and 5.3ghz.

Most likely it starts to throttle cause your front fans dont provide enough fresh air to the t30. Try removing the front panel and test again

 

[Hulk]

Most likely it starts to throttle cause your front fans dont provide enough fresh air to the t30. Try removing the front panel and test again

Just use this for the front fan and you don't need to remove the front panel.

 

[Timur Born]

Drop AC towards 1 (keep DC at stock), something like AC 3 + slight VCore offset might be more stable, though (looking at you, Folding@home). Increase CPU LLC to middle value ("High" in my Gigabyte BIOS, aka step 4 out of 7). If you do per core OC then use TVB to limit boosting above 60-65°C.

And most importantly: Use a power limit slightly above what CB23 uses, no realworld load will ever hit that limit! I am currently using 245W PL1/2.

Don't use silly memory overclocks. 5600 MT CR 1T is enough for almost all realworld applications and runs at 1.15v IMC and dimm voltages for me.