Turn off certain cores to boost memory usage.
Turn off certain cores to boost memory usage.
i've been experimenting with overclocking and everything seems to be working fine. my setup is a 5950x16c/32t, 64gb ram spread across four sticks, using asus rog crosshair v8. it's not the dark hero; everything is watercooled on a custom loop, and the graphics cards are also pretty powerful after digging into the bios (latest version around 5002 from early February 2025). i found an option to disable certain cores, like ccd or 0, to free up cache space for the remaining cores. my goal is to see if a "5800x3d" could fit on my 5950x. although i wanted the 5800x3d, i ended up choosing the 5950 from a 5900 because it was available. ccd does have the fastest and most efficient cores while also running at lower voltages. if i disable cores (could you explain how to disable ccd0?) would that give me more cache on ccd1? i'm just curious if using more cache could improve fps by around 10% or so. it's not a priority since windows and most applications run smoothly, and my 4090 handles games fine. i'm wondering if I could push the fps up a bit more with this approach. thanks! whoop!!
Typically you can't turn off CCXs but you can disable CCDs, which covers all cores and cache. In this scenario, you'd essentially be getting a 5800X. What sets X3D CPUs apart is their 96MB L3 cache on each single CCD/CCX unit, reducing the impact of distributed L3 cache issues and eliminating the need to jump over the IF for full cache access. You might have been better off with a 5700X3D if maximizing FPS was your goal.
I don't understand why I'm surprised. It seemed more likely to be split between the two parts instead of spread out. Are there any ways to tweak the CPU settings to turn off those extra cores on each chipset to boost available cache? I saw some options in the BIOS that made me nervous about changing them. I think something like 0+4 and 1+5 might work, though I'm not checking the bios right now. Also, there are Numa nodes, but I thought they were only for servers with multiple CPUs. My U2 server, for example, has 24 DDR3 modules and two Xeon 6490s or similar. It's 2.4GHz, 384GB RAM, dual processors. Both watercooled units are noisy—90mm fans are really loud, even in the basement I can hear them humming like a jet engine from the top floor! This feels like something that needs fixing. My machine has a 4090, 64GB RAM, 5950X, around 120fps on most AAA games at full settings. But my monitor only goes up to 240Hz. Do I really need G-Sync if it's not limiting my FPS potential? I'm not too worried about latency since I mostly play single-player games outside of that. Thanks for the quick and clear answer!
The resolution remains challenging even at high settings like 7800X3D 1080p240. Adjusting your game configurations is essential to maintain a stable frame rate. You might be amazed at how far you can push performance before noticeable delays appear after just ten minutes.
You might want to give it a shot. I haven't seen any tests for a 4+4 setup with Zen 3 before. My guess is that the extra delay from communication between cores could slow things down compared to an 8+0 layout where just one CCD is active. On the other hand, more cache per core might offset some of that lag. It's tough to tell without real-world testing.
What stands out is that the 5900X rarely outperforms the 5950X. There are occasional situations where it does, possibly due to cache per core differences, but it might also be linked to performance under high threading limits—where the 5900X ticks higher. If boosting cache per core helps the 5950X more often, then the 5900X should generally lag behind. This explanation seems unlikely.
The design allows up to eight cores in a Zen 3 CCD, whereas a reduced version might offer only six. Even if just six cores are heavily utilized on a 5950x... configuration, the L3 memory remains comparable to using eight cores. Variations arise mainly from how data is managed in the L2 cache, but overall it's negligible. The final L3 cache stays consistent across all workloads. A 33% boost per core in L3 would still have limited significance. The main benefit for X3D comes from a 200% increase in cache capacity, which isn't that substantial.