Memory usage affects RAM speed when VRAM is at capacity
Memory usage affects RAM speed when VRAM is at capacity
I’ve been thinking about a few things lately. Recently I got a brand new PC, and the only thing missing is a GPU. I’m considering getting an RTX 4070 Ti, but I’ve heard AMD drivers for VR games aren’t great, and NVidia’s VRAM prices are dropping. Since I often play VRChat, I’ve noticed that during big events with hundreds of players, my graphics card gets nearly full—sometimes over 10GB of RAM is used instead. This makes me wonder how much RAM speed really affects performance. Right now I have two 16GB sticks at 6000MT and CL36, but I’m thinking about adding another set at 6400MT and CL30 once I can afford it, especially after the AGESA update.
RAM velocity isn't the main concern; the bottleneck will come from the PCIe link, which is significantly slower than RAM speeds. If the system relies solely on the full x16 bandwidth for data transfer, performance will cap at around 31.5 GB/s with PCIe 4.0. This is far below the capabilities of DDR5 RAM and even more so than VRAM capacity. Quick RAM won't solve the issue of running out of VRAM, and it has little impact overall. Storage size still plays a role because exceeding it forces the system to use the pagefile, which drastically reduces speed. Capacity is important, but speed remains the critical factor.
Your RAM appears to be very good, likely not the main issue here. It could be the CPU or simply a game that isn't well optimized.
It’s clear you’re focused on the right aspects. For Ryzen models, memory bandwidth becomes the key factor, and a GPU offers significantly more capacity than a CPU would. The 4070 Ti provides five times the bandwidth compared to a theoretical maximum of 6400MT/s for RAM without other CPU constraints. However, the FCLK in Ryzen 7000 chips caps performance at 64GB/s regardless of memory speed, meaning higher memory won’t boost bandwidth much. This also applies to DDR5 variants like 8000 or 6000. Going for a faster setup won’t really help if it can’t access the full system memory bandwidth. The PCIe link limit is another constraint at 32GB/s, so speed improvements there are limited too.
I own a DDR5 setup and planning another upgrade would follow the same standard, indicating it uses PCIe 5.0. Based on this, transmission rates could reach roughly 64GB/s. Are those speeds still relevant? The game's performance is actually limited by its optimization. The issue with VRAM overflow stems from storing numerous assets—each avatar using around 150MB, 50 people equals 7.5GB, plus the game world (~10GB). Additionally, rendering doubles for each eye at high resolutions (1,832 × 1,920) and includes complex effects like particles and varied shaders.
It looks like you're noticing a high resource demand. According to YoungBlade, the issue might stem from insufficient RAM. With 32GB of fast RAM plus VRAM, it seems challenging to load all characters simultaneously.
Current GPUs don’t support PCIe 5.0 because it requires compatibility across the device, motherboard port, and chipset. The RTX 40 line only uses PCIe Gen 4.0, which restricts you to that standard. Even if PCIe Gen 5.0 existed, the data rate from dual-channel DDR5 would still lag behind typical RAM speeds, making it irrelevant.