Despite supporting only two or three games, Intel APO has been a thorough topic of discussion. Our friends at QuasarZone took a stab at it and made some interesting observations. It turns out that Vulkan-based games (singular) don’t see any performance gains with APO enabled. Tom Clancy’s Rainbow Six Siege becomes slower on the Core i9-14900K when APO is used with Vulkan. At 1080p “Ultra,” the average FPS drops from 631 FPS to 584 FPS when APO is switched on.
In DX11 mode, turning on APO boosts the FPS from 620 FPS to 659 FPS at 1080p “Ultra.” This is explained based on CPU core utilization, which behaves differently in Vulkan and DX11 when using APO. In DX11 mode, the P-cores are better utilized with APO enabled, reducing (or at least trimming) E-core usage. All the game threads remain on the P-cores instead of revolving between the two clusters, improving performance.
In Vulkan API mode, the P-core usage doesn’t change upon enabling APO. Furthermore, the first core sees a decrease in utilization, the exact opposite of what we saw in DX11 mode. Meanwhile, the E-cores continue to have an inconsistent utilization graph, indicating that the scheduling fix isn’t effective in Vulkan.
Assuming that APO shifts the primary render thread from the E to P-cores, you’d think that reducing the former or disabling them would provide maximum performance. But no, that’s not quite it either.
In DX11 mode, Rainbow Six Siege runs the best with all the P/E cores and APO enabled. Even 8P + 4E config is faster than just using the P-cores alone. Ergo, the E-cores are not useless.
In Vulkan mode, Rainbow Six Siege behaves strangely. The Core i9-14900K is the fastest at stock, with all cores enabled and APO disabled. Interestingly, the 8P + 4E config is faster than stock+APO but slower than stock.
In Metro Exodus, the stock 14900K with APO produces the best performance, followed by the 8P+4E (no APO) and the 8P configuration. The stock 14900K (APO off) is notably slower than the rest.
Comments ()