Simulate Turbo Boost when Overclocking

[Kratos]

Hi again, I made a thread about a 360mm AIO for an i9-7940X which was running too hot with a Noctua NH-D14, Final result for who ever read and repplied to that thread = It worked really well, even with XMP 3000 MHz enabled it stays under 65c on extreme loads. 

So now? Well...who couldn't resist doing a bit of overclocking....

I followed Der8auer's Youtube tutorial for the 7920X but ran into serious issues. I had done overclocking before with my old i7-4790K managed to get it up to 4.7 GHz at 1.200v with max temps of 80c (not bad).

The thing is with my old i7 as it was only 4 cores, I just did an OC to all 4 cores. Now 14 cores is a bit more to consider the heat output.

 

So now the title of the thread, I thought if I used the Per Core OC setting in the motherboard it will be just like the turbo boost mode, meaning if it was for example a 4 core say with multipliers 44, 42, 40, 38 it will run at 3.8 GHz using all cores, but it appears on all cores it will run at 4,4 on one, 4,2 on another one and so on. I thought the whole point of using the per core method was to do this just like turbo boost does. 

I haven't got enough cooling or PSU power to get 4.4 GHz on all cores (actually tried and reached 87c on Blender), so not knowing this I set the cores just like intel manually (43,41,41,41,40,40,40,40,40,40,40,40,39,39) changing the last 2 cores to 3.9 GHz instead of 3.8 so I thought it will run at 3.9 GHz on all cores. But it didn't. Also the "auto" multiplier on the cores was set by default at 4.3 GHz on all cores which is a bit too much for my cooling solution. 

Anyone has any idea of how to fix this or if it's even possible?

It's really a shame as I did some single core tests with 4.6 GHz and reached 55c max. If I could only have those 2 first cores at 4.6 GHz and then scale back all the way to 4 GHz when using all cores...that would be just perfect as some of my workloads require fast dual core speeds.

[Jurrunio]

Gigabyte calls it Turbo Ratio (X cores active), so I guess this function exists for Skylake X, it just depends on whether the board maker implemented that into the BIOS

 

https://overclocking.guide/gigabyte-x299-skylake-x-overclocking-guide/#Step_4_Adjust_Turbo_Settings

[Kratos]

Thanks! 

I'll have a look, I've got a MSI X299 Tomahawk. I'll read the article and see if it works. Another idea I had was to use the BCLK to 105 - 110 MHz but don't know if that will scale like turboboost does, as the multipliers will be untouched. 105 MHz will result in a 3.99 GHz on all cores and 4.62 GHz on single core. I will try both and read the article and update with results in case anyone is having this issue.

[Jurrunio]

5 minutes ago, Kratos said:

Thanks! 

I'll have a look, I've got a MSI X299 Tomahawk. I'll read the article and see if it works. Another idea I had was to use the BCLK to 105 - 110 MHz but don't know if that will scale like turboboost does, as the multipliers will be untouched. 105 MHz will result in a 3.99 GHz on all cores and 4.62 GHz on single core. I will try both and read the article and update with results in case anyone is having this issue.

I dont think the Tomahawk is high end enough to have a clock gen, which means you're bound by the BCLK overclocking tied to PCIe issue. Careful here, I've seen people corrupt PCIe SSDs because they are connected to the CPU PCIe lanes. SATA devices arent affected.

[Kratos]

3 minutes ago, Jurrunio said:

I dont think the Tomahawk is high end enough to have a clock gen, which means you're bound by the BCLK overclocking tied to PCIe issue. Careful here, I've seen people corrupt PCIe SSDs because they are connected to the CPU PCIe lanes. SATA devices arent affected.

Oh thanks then! I was a bit lost on this one, I do have an NVME PCIe SSD so thanks! Well, I've heard some complains with the board but I've had 0 issues with it, honestly it was handling 4.4 GHz on all 14 cores if it wasn't for my AIO warming up to 87c. 

I couldn't find anything on MSI's site so I'll have to look in the BIOS if there's some Turbo Ratio.

[X_X]

It sounds like you have per core p-states enabled, try disabling it if you can find it in BIOS settings.

[Kratos]

Thanks for both of your repplies, I did look at those, it appears MSI has the feature under the Turbo Ratio name, that feature is to controll C-States, P-States, T-States and all those, but it is very confusing to use I don't know how to. With the Gigabyte article it was much easier as it was clear "Turbo Ratio X Cores Active".

With the MSI it displays like this:

 

-Number of CPU Cores of Group 1

-Target CPU Turbo Ratio Group 1

 

All the way to group 8, which is confusing since I don't know what these groups are and don't know how 8 "groups" fit with my 14 core, if it was 7 groups I could tell each group is 2 cores, but since it's 8...

 

The description next to each one is the following:

 

-Sets the number of CPU Cores as a group to run target CPU Turbo Ratio. The next group should be more than former one in CPU core number.

 

-Sets the target CPU Turbo Ratio value for assigned CPU Cores group. The target CPU Turbo Ratio value should not be higher than the former one.

 

I suppose it was too much work to just put "CPU Clock 1 Active Core" and that's it....

[Jurrunio]

Actually this is also nice as long as you don't need more than 8 steppings for the multipliers.

 

For example, if you set the number of CPU cores (a) of Group 1 to 4, Target CPU Turbo Ratio (b) Group 1 to 46, (a) of Group 2 to 10, (b) of Group 2 to 30, then you'll have 4 cores reaching 4.6GHz when others have no load, otherwise run 3GHz under load.

[Kratos]

2 hours ago, Jurrunio said:

Actually this is also nice as long as you don't need more than 8 steppings for the multipliers.

 

For example, if you set the number of CPU cores (a) of Group 1 to 4, Target CPU Turbo Ratio (b) Group 1 to 46, (a) of Group 2 to 10, (b) of Group 2 to 30, then you'll have 4 cores reaching 4.6GHz when others have no load, otherwise run 3GHz under load.

I really appreciate the repply, but I just don't get it. Why 10 and then 30? What do those values represent?

If for example I want to run all cores at 4 GHz when all cores are used but I still would like to have 2 cores at 4.4 when using 2 cores, then 4.1 when using 4 cores, and 4.0 GHz when using all how can I do that?

What would be the values I'd have to put on Number of CPU cores of Group X and Target CPU Turbo Ratio Group X? Just as an example for my understanding. I still don't understand how I'm going to scale 14 cores if there are only 8 groups and 8 is not a multiple of 14. Not trying to be rude or anything, it's just that your explanation was too technical for a non-regular Overclocker .

[Jurrunio]

Just now, Kratos said:

I really appreciate the repply, but I just don't get it. Why 10 and then 30? What do those values represent?

30 refers to the multiplier (30x100 = 3000MHz). 10 here should've been 14... not sure what I'm thinking at the time.

 

1 minute ago, Kratos said:

If for example I want to run all cores at 4 GHz when all cores are used but I still would like to have 2 cores at 4.4 when using 2 cores, then 4.1 when using 4 cores, and 4.0 GHz when using all how can I do that?

so Group 1 has 2 and 44, Group 2 has 4 and 41, Group 3 has 14 and 40

[Kratos]

4 minutes ago, Jurrunio said:

30 refers to the multiplier (30x100 = 3000MHz). 10 here should've been 14... not sure what I'm thinking at the time.

 

so Group 1 has 2 and 44, Group 2 has 4 and 41, Group 3 has 14 and 40

Ahh now I get it, So just as an example, if I wanted to do 4.4 GHz on 2, 4.1 GHz on 4, and 4 GHz on the rest, then it would look like this: ?

 

-Number of CPU Cores of Group 1 = 2

-Target CPU Turbo Ratio Group 1 = 44

 

-Number of CPU Cores of Group 2 = 4

-Target CPU Turbo Ratio Group 2 = 41

 

-Number of CPU Cores of Group 3 = 14

-Target CPU Turbo Ratio Group 3 = 40

 

If this is correct, the only thing I'm left to ask is this "The target CPU Turbo Ratio value should not be higher than the former one" this is refering that the Turbo Value should be higher when there are less cores meaning that you can't have 2 cores group 1 at 40 and then 14 cores at group 2 at 44??

Sorry for asking too much questions I just want to make sure of every little step before I fidle with something I don't know on a 1200€ CPU.

[Jurrunio]

1 minute ago, Kratos said:

Ahh now I get it, So just as an example, if I wanted to do 4.4 GHz on 2, 4.1 GHz on 4, and 4 GHz on the rest, then it would look like this: ?

 

-Number of CPU Cores of Group 1 = 2

-Target CPU Turbo Ratio Group 1 = 44

 

-Number of CPU Cores of Group 2 = 4

-Target CPU Turbo Ratio Group 2 = 41

 

-Number of CPU Cores of Group 3 = 14

-Target CPU Turbo Ratio Group 3 = 40

 

That's correct, if that's what MSI's software engineers are also thinking.

 

1 minute ago, Kratos said:

If this is correct, the only thing I'm left to ask is this "The target CPU Turbo Ratio value should not be higher than the former one" this is refering that the Turbo Value should be higher when there are less cores meaning that you can't have 2 cores group 1 at 40 and then 14 cores at group 2 at 44??

Sorry for asking too much questions I just want to make sure of every little step before I fidle with something I don't know on a 1200€ CPU.

Yes I think, that make sense to me. Not like Intel did that either.

[Kratos]

Okay thank you! I'm going to try now and I'll update the results here with what I got to see if that's it.

 

Edit: Yes it was that, however it doesn't scale quite as well in Single Core and 4 cores, since 2 was at 4.4 GHz and on single core tests it was bouncing from one core o another at 4 and 4.1 GHz and sometimes at 4.4. They weren't the favoured cores, that's really the only issue I see but it can be solved, temps didn't go pass 50c. Now with 4 GHz on all cores, temps were about 74c max and 69 average (used a 2h Render scene in Blender). The only thing that did scare me was the Motherboards temps were reading 85c max and 81.5c average, I made another thread about VRM cooling, I'm still waiting for a 90mm Noctua Fan to stack up somewhere near.

 

Thanks Jurrinio, I can now marked your explanation as solved.