Limits on VRAM for APU/iGPU devices
Limits on VRAM for APU/iGPU devices
Hello everyone... I remember when the Vega Integrated Graphics launched, we were restricted to just 2GB of VRAM from our system RAM. With a 16GB kit using integrated graphics, that reduced our available memory to 14GB. I’m curious if this situation has changed and the limit has been raised. It would be great if we could expand it to 8GB for a lower-end gaming or emulation PC.
If you turn on the iGPU on your 5800H (Zen 3) it sets aside 2GB of memory by default. The BIOS lets you lower this but advises against it. The optimal choice seems to be turning it off completely and using dGPU instead. It makes little sense to reserve more than needed, since the 2GB is meant for the driver and hardware to recognize as available. If demand goes up beyond that, it will release some from other system resources. On a dGPU, using full system RAM is problematic due to its sluggishness. With iGPU, the same amount of RAM is used, so performance stays consistent.
Exactly. Overloading system memory harms a dedicated GPU due to extra delay and reduced speed. An integrated GPU uses the same VRAM as system memory, so no extra cost arises. Keeping more than 2 GB reserved permanently is pointless for tasks that don’t require it.
My prediction suggests the reservation serves two purposes: 1) guarantees a substantial unbroken RAM space remains available 2) prevents the OS from repeatedly attempting to reclaim memory for minor VRAM adjustments. The concept aims to provide sufficient capacity most of the time, reducing interruptions caused by the system trying to allocate more. Indeed, it has been enhanced—Steam Deck now reserves up to 4GB when heavy VRAM usage is expected. A drawback exists if no RAM is free at the moment needed; hence the reservation can be adjusted according to your needs. Since most integrated GPUs are designed for desktop tasks, the default setting aligns with what’s typically required for the frame buffer alone.
When the iGPU demands more memory than is present, it becomes problematic. Similarly, locking a significant portion of system RAM to the iGPU can also cause issues. The best solutions are either boosting overall RAM capacity or opting for a dedicated GPU, which tends to offer better performance per dollar. Adjusting RAM allocation would resolve both scenarios.
I measured VRAM with nvidia-smi, which is the most reliable method for tracking usage. In theory, the numbers would match an iGPU, but the impact is much more significant with an iGPU.
I considered setting a fixed 8 GB allocation as suggested by OP. On a 32 GB machine it could work, and 24 GB still covers most needs. However, choosing 16 GB would mean permanently reducing your RAM by half—less ideal for performance. A gaming-focused system like a Steam Deck might handle it, but mixing in memory-heavy tasks would likely require frequent adjustments. Unless you have extremely limited space, like in a handheld, I’d prefer a dedicated card to avoid the hassle.
I was only testing with 16GB, but I need to know if it's feasible to allocate more than 2GB for the iGPU. A straightforward yes or no will suffice. Thanks to @Eigenvektor for his positive outlook—it’s much cheaper to upgrade RAM than to purchase a dedicated GPU.
Even the Steam Deck limits you to 4GB RAM, which is logical since running games at 800p on a compact display doesn’t require ultra-high resolution files. The issue lies with certain titles that don’t reduce their resource demands as expected. Essentially, you only allocate memory when necessary, avoiding competition between CPU and RAM. If your system has ample memory, the problems it addresses are minimal, making it unnecessary to increase the limit beyond what’s needed.
According to user feedback, the R7 8700G is priced similarly to an i3-12100F paired with an RX 6600. The i3 model with a dedicated GPU offers about double the performance for the same budget. https://www.youtube.com/watch?v=Ye60Wf8lUt8&t=914s While cheaper CPUs exist with integrated GPUs, it's wise to review options before purchasing. A more affordable CPU-GPU setup might actually perform better.