Graphics cards: when you need one and how much VRAM is enough
Integrated or discrete? How much VRAM for gaming and editing? How to avoid model-name traps.
When is integrated graphics enough?
Modern integrated GPUs (AMD Radeon 890M, Intel Xe2) comfortably handle office work, 4K media and even e-sports titles. Cheap “budget” discrete cards often add nothing on top — while using more power.
When do you need a discrete card?
- Modern AAA games at 1080p/1440p with higher settings.
- Video editing, rendering, 3D modeling.
- Higher resolutions (1440p+) — there VRAM is the deciding factor.
How much VRAM is enough?
- 8 GB — the lower bound; runs out quickly at 1440p. Modern engines use heavy asset streaming, ray tracing and AI frame interpolation (e.g. DLSS 4) — once VRAM is exhausted the system slows down: micro-stutter and texture pop-in.
- 12 GB — the real minimum for AAA at 1440p (Arc B580, RX 9060 XT, RTX 5070).
- 16 GB — comfortable headroom for years.
The trap: model name ≠ performance
In laptops, manufacturers can set TGP (power limits) across a wide range. An RTX 5060 at 60 W will be clearly slower than the same card at 115 W. Always check the full specification of the exact configuration, not just the name.
The trap: PCIe bus
A PCIe 4.0 x8 card (e.g. Arc B580, RTX 5060) in an older board with a PCIe 3.0 bus loses bandwidth — in bandwidth-sensitive titles that is a real performance loss. Check the PCIe generation of the x16 slot in the motherboard.
The trap: 32-bit CUDA on RTX 50
The Blackwell architecture (RTX 50) dropped native 32-bit CUDA support — older compute applications fall back to the CPU and run significantly slower. If you rely on legacy CUDA software, verify compatibility before buying.
Rule of thumb
Office and media — integrated graphics. AAA gaming at 1440p — a card with 12 GB+ VRAM, and in laptops also verify the TGP.