Now with GPUs coming imbalanced from a VRAM vs compute, and more VRAM intensive applications becoming the norm (eg. Ray tracing, Machine Learning) VRAM should play a much larger role in the score each card is given. For example, a card with 48GB of VRAM should score significantly higher than one with 4GB given the same amount of computing capabilities. Users should also have the ability to filter out cards by vram amount, since for some users, lower VRAM cards are useless (eg. Someone trying to run LLMs locally).
Announcement
Collapse
No announcement yet.
Weigh VRAM more heavily in in benchmarks.
Collapse
X
-
We agree that the amount of VRAM is important to an extent.
But it is also nice to have benchmark tests that can run on a broad range of hardware. If we made a test that required 48GB of video RAM, then it wouldn't even launch on 99% of machines.
Having said that, we have starting planning on the next major release of PerformanceTest. More intensive 3D tests are definitely going to be added (and likely we drop the low resolution DirectX9 from the suite).
Comment