Someone with 13700k/13900k assist with an inquiry

[Agall]

Hello, my request is simple if possible. Someone with a 13700k/13900k validate some of the information for me for academic reasons (I don't have an LGA 1700 K series CPU to test this myself):

 

Does the 13700k/13900k allow the user to down configure the cores like it does on AM4/AM5? Example being that I can turn my 7950x3D into a 7600x3D or 7800x3D or 7900x3D by manually reconfiguring the CCDs. 7800x3D being a 1-1-1-1-1-1-1-1, 0-0-0-0-0-0-0-0 CCD configuration in the AMD Overclocking section of the UEFI.

 

If that is possible, does down configuring the 13700k/13900k lower the reported quantity of L3 cache that Task Manager's Performance section reports? I've heard of people turning off E cores, which would give a similar scenario since based on Intel's binning scheme, each E core cluster of 4 E cores enabled adds 3MB of L3 cache, similar to 1 P core.

 

The theory being that if a 13900k can be down configured to a 13600k but with more L3 cache, how much does that matter with respect to performance? Requesting a suite of benchmarks would be too much an ask in this, so I'm simply wondering if it's even the case or if access to that 3MB of L3 cache goes away with it, similar to what happens in Intel's binning scheme.

 

Example:

 

I know Windows does this because it's how I can validate my down configurations, since a pseudo 7700x with CCD1 only will show less L3 cache than a pseudo 7800x3D with CCD0 only. Image shown is simply the laptop I'm currently on with a 5800HS.

 

"Since the 36 MB L3 cache remains untouched, there's more of it accessible per core."

RTX 4090 & 53 Games: Core i9-13900K E-Cores Enabled vs Disabled Review | TechPowerUp

 

I'm unsure if this applies to P cores, however.

 

[RONOTHAN##]

20 minutes ago, Agall said:

Does the 13700k/13900k allow the user to down configure the cores like it does on AM4/AM5? Example being that I can turn my 7950x3D into a 7600x3D or 7800x3D or 7900x3D by manually reconfiguring the CCDs. 7800x3D being a 1-1-1-1-1-1-1-1, 0-0-0-0-0-0-0-0 CCD configuration in the AMD Overclocking section of the UEFI.

You can do one better on LGA 1700, turn on and off specific cores. Only limitation with this is there must be at least one P core enabled. 

 

20 minutes ago, Agall said:

does down configuring the 13700k/13900k lower the reported quantity of L3 cache that Task Manager's Performance section reports?

No, the reported cache amounts remain the same. The cache differences between the chips aren't really enough to cause massive performance differences like they do with the X3D series of chips, but they do technically exist. 

[Agall]

9 minutes ago, RONOTHAN## said:

You can do one better on LGA 1700, turn on and off specific cores. 

 

No, the reported cache amounts remain the same. The cache differences between the chips aren't really enough to cause massive performance differences like they do with the X3D series of chips, but they do technically exist. 

I'm trying very hard to not buy a 13600k and 13900k to test this myself, although I don't care enough to do a proper suite of academic benchmarks to properly gauge whether if the extra 12MB of L3 cache would matter. Its easy to demonstrate large gaps, but demonstrating an extra 50% extra L3 cache's advantage sounds more like a HWUB sort of inquiry.

 

I'll see if it's something the labs would be interested in, I think its right up their ally since its possible.

[Kilrah]

30 minutes ago, Agall said:

does down configuring the 13700k/13900k lower the reported quantity of L3 cache that Task Manager's Performance section reports?

No but L1/L2 go down.

 

 

A cinebench run with 6/8 P/E gives 23K points, pretty on par with what seems a 13600K should give.

[Agall]

10 minutes ago, Kilrah said:

Yup, L1/L2 go down.

 

 

A cinebench run with 6/8 P/E gives 23K points, pretty on par with what seems a 13600K should give.

Noting that L1/L2 cache are per core, which makes sense.

 

Peculiar why Intel finds it necessary to disable 3MB of L3 per P-core or 4x E-core cluster then when binning, if it doesn't need it to access its respective section of the L3 cache.

 

 

Unfortunately, you'd need a 13600k to drop in to test whether that's anymore performance, but thank you for the answer. I posted on LTT lab's suggestion thread since I think it would be a wonderful academic test. 

 

I could see a scenario where Intel creates a 'gaming' series processor with no E-cores and the rest of the die allocated to L3 cache, where they could potentially get up to 80MB if they're spicy enough.

 

Also my testing with cinebench r23 doesn't seem to scale based off L3 cache, being my tests between CCD1 and CCD0. Both CCDs are thermally limited relatively quickly, so CCD1 doesn't boost much higher than the 4.8-4.9GHz all core but does bench higher than CCD0.

[Kilrah]

4 minutes ago, Agall said:

Unfortunately, you'd need a 13600k to drop in to test whether that's anymore performance

And find a cache-heavy workload.

Would be interesting to see the comparison indeed. Intel's designs could also respond completely differently to cache amount than AMD's...

 

[Agall]

On 8/7/2023 at 9:36 AM, Kilrah said:

And find a cache-heavy workload.

Would be interesting to see the comparison indeed. Intel's designs could also respond completely differently to cache amount than AMD's...

 

That's fair. I'd say if someone did test it, then that would conclude whether that was the case or not.

 

If someone did test a 13600k and 13900k (down-configured) with similar clock speed correction and there was no performance difference, then it likely disables the L3 cache as well. If it does have higher performance beyond margin of error, then its likely due to the increased L3 cache.

 

That article might incorrectly assume that disabling E cores doesn't change the L3 cache because it simply isn't reporting it properly to Windows as well.