It is quite well known that the iGPUs that are currently on the market are a huge leap forward compared to what we saw just a few years ago, but this is bringing a very interesting consequence to deal with: the performance of its “supposed” VRAM. How does the processor allocate graphics memory to the iGPU? And above all, how much does performance depend on this type of memory?
Integrated graphics cards on both motherboards and processors have always depended on external memory to be used as video RAM or VRAM, but for some time now the assignments are different and it is convenient to understand why.
Your iGPU is still using system RAM as VRAM
And it does so for a very simple reason: the benefit/cost ratio. The processors that have iGPU and that are not called as such APU because the CPU part prevails over its opposite, usually have a low-level graphics card, competent for certain tasks yes, but not for gaming as such. Therefore, and although this type of device could have hbm-memory to his credit and achieve shoot performance due to final costs for the price of the processor this is not done.
But it is not done for APUs as such and for the same reasons. The technology is there, it’s mature, but it’s expensive to implement and no one in their right mind would pay the extra cost, so the range of processors would end up dying compared to that of the rival. The solution to have VRAM? The one of always, the one of all life: use system RAMthat is, DDR memory as an information exchanger.
The difference is the way of assigning it by your PC, since in the past a very small part was reserved for VRAM, only a few MB, until the potential of the iGPU took off, the games demanded more video memory and it had to change the system.
Graphics APIs and the importance of speed
Some APUs still allow maximum allocation from the BIOS/UEFI, but as such this is tending to go away because Shared Memory is doing a better job these days thanks to DX12 and AMD and Intel drivers.
Normally the parameter that defines the VRAM in BIOS is the DVMT Pre-Allocated VRAM, where as a rule it leaves up to 4,096 MB assigned. Windows and DX12 using what is known as hUMA (particularly at AMD) or Unified Memory Architecture. This system shares the PC’s RAM memory with the iGPU and if our processor and platform allow it, it should be in AUTO, since it will be, as we say, the system that manages to increase or reduce the size of the frame buffer through a consistent memory model and bidirectional (HSA in AMD).
The proportion of RAM used is determined in part by the allocation as shared memory, which at the same time overlaps with the Windows Virtual Memory if there is a case of lack of RAM in the system. The thing to understand here is that the API and driver will oscillate the amount of VRAM allocated in a ratio between 1:4 and 1:2depending on the case.
That is, if the system requires 1 GB of VRAM for the iGPU, our PC should have 4 GB in the best case and 2 GB in the worst. It is by no means an exact rule, as we say it is managed in a completely open and free way (if we indicate so in AUTO), so your iGPU’s VRAM remains and will remain system RAM on PC.