disable cores to increase cache?

[Cramig88]

so i been playing around with overclocking, got my system stable and all is well.

i have a 5950x 16c/32t, 64g ram across 4 sticks, asus\rog crosshair viii NOT the dark hero - everythings watercooled on a custom loop, video cards too

 

anywho upon digging around the bios (latest version, 5002 from feb 20th or something 2025) i noticed theres an option to disable cores and such.  i was just wondering, is there a way i can disable CCX1 (CCD? sorry little dyslexic) or 0, with the goal of increasing the available cache space for the remaining cores?

basically id like to see if a "5800x3d" could be made out of my 5950x.  unfortunately i DID want the 5800x3d but since i couldnt find one available i settled with the 5950 (from a 5900).

CCX1 actually DOES have the fastest and coolest cores while also running at the lowest voltages\temps.  so yeah, if i disable cores (could u explain how to disable CCX0?) would i have more cache space on CCX1?

 

just trying to see if theres any FPS gain with more cache useable with my cpu.  I could easily live with 8 cores and 16 threads especially if it gives me more overhead for overclocking the cpu.  more speed though isnt really a concern since windows and junk loads up super speedy and games run fine on my 4090.  curious if i could bump the fps up by idk 10% or something.

 

thanks!  whoop!!

[Levent]

IIRC you cant disable CCXs but can disable CCDs (meaning all cores and cache). In that case, you will basically have a 5800X. What makes X3D CPUs special is that they have 96MB L3 cache on single CCD/CCX, that basically mitigates all negative effects distributed L3 cache and having to travel over IF for accessing entirety of L3 cache.

You should have gotten a 5700X3D if FPS was your primary concern.

[Cramig88]

55 minutes ago, Levent said:

IIRC you cant disable CCXs but can disable CCDs (meaning all cores and cache). In that case, you will basically have a 5800X. What makes X3D CPUs special is that they have 96MB L3 cache on single CCD/CCX, that basically mitigates all negative effects distributed L3 cache and having to travel over IF for accessing entirety of L3 cache.

You should have gotten a 5700X3D if FPS was your primary concern.

eh i cant say that im surprised.  kinda figured it would be split between the two rather than distributed.

are there any optimizations to the cpu i can make to disable idk 4 cores on each ccx to increase the available cache on either?  i did notice a few options in the bios that i was "worried" about altering.  idk something about disabling cores and it had a really weird set of options.  something like 0+4 1+5 (just guessing as im not looking at my bios now) but 0+4 was deff an option.  also numa nodes too, although i thought numa was only related to servers with multiple CPU's.  My U2 server for example, its got 24 dimms/DDR3 and 2x xeon 6490's or some BS.  all i remember is its 2.4ghz, 384gb ram and dual processors.  both watercooled cause geez, them 90mm fans are loud as f* - even being in the basement i can hear the darn thing humming away like a jet engine from the top floor!

kinda a bleeped situation, got a 4090, 64g ram, 5950x, ~120fps on super max settings for most if not all AAA titles.  yet my monitor goes up to 240hz...  actually since all my games are under 240 fps does gsync have any benefit or is it limiting my FPS potential?  im not too concerned about latency since i only play single player stuffs anyway

outside of that thanks for the concise and accurate response! 

[SorryBella]

15 minutes ago, Cramig88 said:

~120fps on super max settings for most if not all AAA titles.  yet my monitor goes up to 240hz... 

...and what resolution? Even on 7800X3D 1080p240 is pretty hard to hit with maxxed out settings, you need to optimize your game settings if youre in it for the framerate. You'll be surprised at how low you can go before you start noticing things after 10 minutes.

[Cramig88]

5 hours ago, SorryBella said:

...and what resolution? Even on 7800X3D 1080p240 is pretty hard to hit with maxxed out settings, you need to optimize your game settings if youre in it for the framerate. You'll be surprised at how low you can go before you start noticing things after 10 minutes.

... 5120x1440 maybe im setting my expectations too high for 120fps at 120hz on ultra everything?
still if i can disable not so well performing cores, 4x4, that leaves a lot of cache to utilize

[TatamiMatt]

On 2/28/2025 at 1:52 PM, Cramig88 said:

... 5120x1440 maybe im setting my expectations too high for 120fps at 120hz on ultra everything?
still if i can disable not so well performing cores, 4x4, that leaves a lot of cache to utilize

I believe 1440p ultrawide is actually more pixels than 4K, so in your head, would you be able to get that performance at 4K? If not then you definitely wont with 1440UW

 

EDIT: upon researching to confirm its actually slightly less pixels but much closer to 4K's pixel count than standard 1440p's

[YoungBlade]

7 minutes ago, Cramig88 said:

... 5120x1440 maybe im setting my expectations too high for 120fps at 120hz on ultra everything?
still if i can disable not so well performing cores, 4x4, that leaves a lot of cache to utilize

You can try it. I don't know that I've ever seen testing of a 4+4 configuration for Zen 3. My hunch is that the increased latency from core-to-core communication across the connection between the CCDs would make it slower in practice than a basic 8+0 configuration from turning off one CCD. But maybe the increased cache per core would help? It's hard to say without testing.

[cmndr]

No point to disabling cores in 99% of cases. You're effectively downgrading to some other CPU model. Benchmarks exist and it's very rare for a 5800x class CPU to outdo a 5950x in any meaningful capacity. 

If you want more cache... x3D parts. 

[YoungBlade]

1 hour ago, cmndr said:

No point to disabling cores in 99% of cases. You're effectively downgrading to some other CPU model. Benchmarks exist and it's very rare for a 5800x class CPU to outdo a 5950x in any meaningful capacity. 

If you want more cache... x3D parts. 

To me, the thing that's more telling is that the 5900X basically never beats the 5950X. There are rare games where it does, and it's possible that the cache per core is the reason why, but it could also be because at stock limits in heavily threaded loads, the 5900X clocks higher per core.

 

If increasing cache per core benefits the 5950X in most cases, then the 5900X should beat it in most cases. But it doesn't, so this is unlikely to help.

[cmndr]

21 minutes ago, YoungBlade said:

To me, the thing that's more telling is that the 5900X basically never beats the 5950X. There are rare games where it does, and it's possible that the cache per core is the reason why, but it could also be because at stock limits in heavily threaded loads, the 5900X clocks higher per core.

That's in part because a Zen 3 CCD can be thought of as having UP to 8 cores available while a cut down CCD might only have 6 available. 

If only 6 cores are being highly used on a 5950x... there's still the exact same amount of L3 as if 8 cores are being widely used. There's going to be differences because of how data in L2 cache is handled but... it's pretty minor. There's still the same L3 cache available at the end of the day to share across the total workload.

 

Also a 33% increase in L3 cache per core would have pretty modest effects anyway. The big impact on the X3D parts comes from having 200% more cache and it's not THAT profound. 
 

[Cramig88]

18 hours ago, YoungBlade said:

To me, the thing that's more telling is that the 5900X basically never beats the 5950X. There are rare games where it does, and it's possible that the cache per core is the reason why, but it could also be because at stock limits in heavily threaded loads, the 5900X clocks higher per core.

 

If increasing cache per core benefits the 5950X in most cases, then the 5900X should beat it in most cases. But it doesn't, so this is unlikely to help.

As mentioned I had a 5900 and went to a 5950 cause I couldn’t get my hands on a 5800x3d. In fact I actually “burned” through 3 5900x’s

sure it has less cores but a higher clock, therefore allowing (whatever) size of the cache to be distributed and therefore provide, better performance?

 

I tried multiple methods since this post, Numa nodes, disabling cores, and yes also the 4+0 (no clue what that means) and I benched virtually every logical logical configuration for the cpu

 

from what I learned thru all the testing is the 5950 is actually a solid cpu. Might not be a 5800x3d but my gpu (4090) isn’t cpu bottlenecked. 
yea the x3d variants have a huge cache which is beneficial but between the two (despite multiple sources sayin otherwise) the 5950 was actuallly a better choice. Why?

well, it’s got 16 cores, duh. Meaning it can handle the reduced cache csuse it’s got more cores to process the data faster.  Sure a larger cache would be helpful especially if it’s x3d or. VNAND, but stuffs happening so fast with this kind of horsepower. 

long story short the difference when paired with a 4090 is… insignificant. Plus it’s important to note I do gaming, but also other office/otherwise kind of stuff

so in my use case scenario along with my equipment (plus amd’s sweet OC options) I have a solid balance between the two

[cmndr]

CPU matters less for many (not all) gaming situations than the internet would believe. 

If you're playing at 1440p with a 4090... a 5950x is pretty close to a 5800x3D and not even THAT far from a 9800x3D for many titles. 
Factorio, certain MMOs, etc. shift this but...