Our recommendation
ASUS ProArt GeForce RTX™ 4070 OC Edition 12GB GDDR6X Graphics Card (PCIe 4.0, 12GB GDDR6X, DLSS 3, HDMI 2.1, DisplayPort 1.4a) is our top-scoring pick in Graphics Card, at 9.8/10 (model: ProArt GeForce RTX™ 4070 OC Edition).
ASUS ProArt GeForce RTX™ 4060 Ti 16GB OC Edition GDDR6 Graphics Card (PCIe 4.0, 16GB GDDR6, DLSS 3, HDMI 2.1a, DisplayPort 1.4a) sits just 0.2 behind at 9.6, close enough that the deciding factor is usually which specification matters more to you (gpu: RTX 4060 Ti).
Both scores are our editors’ assessment of published specifications, pricing and terms against the criteria in our rating methodology — they are not aggregated customer reviews.
Start from your monitor, not the card
The single most useful step before buying is writing down your display's resolution and refresh rate, because that determines how much card you actually need. Buying a flagship GPU for a 1080p 60 Hz panel wastes most of what you paid for; pairing an entry card with a 1440p 165 Hz monitor wastes the monitor.
- 1080p — entry and mid-range cards handle this comfortably, including at high refresh rates in most titles.
- 1440p — the current sweet spot, and where mid-range cards earn their money. High refresh here needs genuine mid-to-upper tier silicon.
- 4K — demands upper-tier hardware, and this is where upscaling stops being optional in demanding titles.
VRAM: the specification that ages worst
Video memory has become the most common reason an otherwise capable card starts stuttering a few years after purchase. Textures, higher resolutions and ray tracing all consume VRAM, and when a card runs out the result is not a gradual slowdown but sharp frame-time spikes and texture pop-in.
Treat VRAM as a longevity decision rather than a performance one. Two cards with similar benchmark scores today can diverge sharply in two years if one has meaningfully less memory. When choosing between a slightly faster card with less VRAM and a slightly slower one with more, the extra memory is usually the better bet for a build you intend to keep.
Ray tracing and upscaling
Ray tracing simulates light physically and costs a great deal of performance. Its practical viability depends on the vendor's upscaling technology, which renders at a lower internal resolution and reconstructs the image — NVIDIA's DLSS, AMD's FSR and Intel's XeSS. Two things worth knowing: upscaling quality differs noticeably between these, particularly at lower internal resolutions, and frame-generation features that insert synthesised frames raise the frame counter without reducing input latency in the same way. If competitive responsiveness matters more to you than visual effects, weight raw rasterisation performance over ray tracing capability.
Will it physically fit, and can your system feed it?
This is where builds fail, and it is entirely avoidable with three measurements.
- Length and thickness. Check your case's maximum GPU clearance against the card's length in millimetres. Many current cards are triple-slot, which can block the slot beneath and foul front-mounted radiators or drive cages.
- Power supply capacity and connectors. Look at the card's recommended PSU wattage and leave headroom for transient spikes, which are brief power surges that can trip a protection circuit on a marginal supply even when average draw looks fine. Confirm you have the right connectors — modern high-end cards use a 12-pin connector rather than multiple 8-pins, and an adapter must be seated fully to be safe.
- Case airflow. A high-wattage card in a poorly ventilated case throttles and gets loud regardless of the cooler on it.
PCIe generation and CPU pairing
Modern cards run on older PCIe slots with a small performance penalty — generally minor at full x16, but potentially significant on cards with a narrow x8 or x4 interface, which are more sensitive to older slots. More importantly, a fast GPU paired with a much older processor will be limited by that processor at lower resolutions, where the CPU does proportionally more work. If your build is several generations old, check whether the CPU is the actual constraint before spending on the card.
Board partners, cooling and noise
The same GPU is sold by several manufacturers with different coolers, clock speeds and prices. Factory overclocks typically deliver a few per cent, which is rarely worth a large premium. What is worth paying for is a cooler that keeps the card quiet under sustained load, since noise is the difference you will notice daily. Look for reviews that publish noise levels and temperatures under extended load rather than short benchmark runs, and check whether the card has a zero-RPM mode that stops the fans at idle.
New, used and mining cards
The used market offers real savings and real risk. Cards used for cryptocurrency mining ran at sustained load, which is not automatically damaging — steady temperatures are gentler than repeated thermal cycling — but fans and thermal paste wear, and warranties are often non-transferable or already expired. If buying used, prefer sellers who accept returns, test immediately under sustained load while watching temperatures, and check for physical signs such as sagging, damaged connectors or replaced thermal pads. Verify the card is genuine and matches its claimed model in system information.
On timing: GPU generations arrive every couple of years, and prices on the outgoing generation usually fall when a successor launches. If you are not in a hurry, checking whether a new generation is imminent can save a meaningful amount.
Specifications, pricing and availability in this category change quickly, and performance varies by title and system configuration. Check current independent benchmarks at your own resolution, and confirm dimensions and power requirements against your case and supply before buying.
