Why is the 5090 allegedly having 32GB VRAM a big deal? Can’t they just solder on another VRAM chip if they want?
In: Engineering
You need circuitry in the chip that is able to address all of that memory. That circuitry can get very complicated.
Up until now it’s been Nvidia’s apparent policy to limit the VRAM on consumer cards, which is thought to be to force people doing professional work that needs that extra VRAM to buy the more expensive professional cards.
So it’s this change of policy that is a big deal.
Not just 32GB VRAM, but GDDR7 VRAM, which is faster and more energy efficient than it’s GDDR6x predecessor.
And the 5090 is a card that straddles the boundary between a gaming card and a professional card. Graphics cards nowadays have become a central piece for a content creator, since they’re now heavily used both in video editing, 3D rendering and especially AI. You can never have enough VRAM for each of these applications.
And many of these content creators both work with a GPU like the 5090, and play games with it on the same PC, so having one single PC that can do both is very useful.
EDIT: Oh and yeah, you can’t solder more memory because Nvidia have disabled that with the 40 series onwards. You can do it on the 3090 though.
No. You need a custom software for the GPU to even recognize the memory properly and use it. Think of the individual software for each GPU like a plan for a building, and the GPU can only use the parts of the building it knows about.
The added memory would exist, yes, but the GPU wouldn’t be able to recognize the extra rooms and simply wouldn’t use it, and this is of course beside other obvious issues, like does the mainboard even have extra slots for this, and if it does, did they actually place physical lanes leading to those slots, or are they just “dead space”?
Latest Answers