The GeForce RTX 4090 is a flagship card from NVIDIA’s Ada Lovelace generation, built around 16,384 CUDA cores and 24GB of GDDR6X memory on a 384-bit bus. Its headline numbers are that 24GB frame buffer, a board power rating of roughly 450 watts, and a physical size that is larger than most cards sold before it.
This guide reads the spec sheet line by line, explains which workloads can actually fill 24GB, covers the power connector and case-fit issues that trip up new owners, and lists the cautions that apply when a card is bought second-hand. Check the manufacturer’s current documentation for exact figures on any specific partner model, since factory-overclocked variants differ from the reference design.
RTX 4090 Specs at a Glance
The table below summarizes the published reference specifications. Partner cards from other brands can run higher clocks, use different coolers, and carry slightly different power limits, so treat these as the baseline.
| Spec item | Published reference value | What it means in practice |
|---|---|---|
| Architecture | Ada Lovelace | Fourth-generation Tensor cores and third-generation RT cores |
| CUDA cores | 16,384 | The main shader count; the largest in the GeForce 40 series |
| Video memory | 24GB GDDR6X | Roughly double what mainstream gaming cards of the era carried |
| Memory bus | 384-bit | Memory bandwidth of about 1 terabyte per second |
| Board power | About 450W | Needs a strong power supply and good case airflow |
| Power connector | 16-pin 12VHPWR (adapter included) | Needs careful, fully seated installation |
| Interface | PCIe 4.0 x16 | Works in newer and older slots; older slots rarely limit it much |
| Display outputs | Three DisplayPort, one HDMI on the reference design | Supports high-resolution, high-refresh monitors |
What the Core Count and Clock Speeds Tell You
CUDA cores are the general-purpose processing units of an NVIDIA GPU. The 4090’s count of 16,384 is the figure most people quote, but cores alone do not decide real-world speed. Clock speed, memory bandwidth, cache size and software support all shape the result. Ada Lovelace added a much larger L2 cache than the previous generation, which helps the chip avoid waiting on memory.
The boost clock on the spec sheet, around the mid-2 gigahertz range, is a typical figure rather than a guarantee. Actual clocks move up and down with temperature, power limits and the workload. This is why two cards with the same chip can behave slightly differently when one has a better cooler.
The card also carries dedicated hardware beyond shaders. Fourth-generation Tensor cores run DLSS upscaling and frame generation, and third-generation RT cores handle ray-tracing calculations. Dual hardware video encoders with AV1 support matter for streamers and video editors who export in modern formats.
Reading the 24GB Memory Figure
The 4090’s 24GB of GDDR6X is its most distinctive feature, and it is also the most misunderstood. Video memory holds textures, geometry, frame buffers and, for compute work, entire datasets or model weights. When a workload exceeds the available memory, the system spills into slower system RAM or simply fails with an out-of-memory error.
Where 24GB is actually used
Most current games fit comfortably in well under 24GB even at high resolution. The workloads that can genuinely use the full capacity are professional and enthusiast tasks: large 3D scenes in path-traced renderers, high-resolution video editing with heavy effects, and running or fine-tuning large language and image-generation models locally. For local AI, memory capacity often decides which models can be loaded at all, not just how fast they run.
Why bandwidth matters as much as capacity
The 384-bit bus and fast GDDR6X give the card bandwidth near one terabyte per second. High bandwidth helps at 4K and above, with large textures, and in compute tasks that stream data constantly. A card with the same capacity but a narrower bus would behave differently, so read both numbers together.
When extra VRAM is wasted
A player at 1080p or 1440p who mostly plays esports titles will rarely approach even half of 24GB. In that case the extra memory is headroom for the future, not a present-day benefit. Readers who only game at modest resolutions can find a lower-tier card sufficient.
Power Draw, the 12VHPWR Connector and Your PSU
NVIDIA rates the reference board at about 450 watts, and its published guidance for the system power supply has typically been in the 850 watt class. That figure covers the whole PC, including a high-end processor, so a system with a power-hungry CPU and many drives benefits from extra margin. Choose a unit from a reputable maker with a good efficiency rating and enough 12V capacity.
The card uses the 16-pin 12VHPWR connector, which carries power through a compact plug rather than three or four older 8-pin plugs. The included adapter converts several 8-pin cables into the 16-pin form. Newer power supplies ship with a native cable. Whichever you use, there are a few safe habits worth following:
- Push the connector in until it is fully seated and flush; a partly inserted plug can create resistance and heat.
- Avoid sharp bends right next to the plug. Leave a few centimeters of straight cable before any curve, and check the clearance before closing the side panel.
- Do not reuse a cable that shows discoloration, melted plastic or loose pins.
- Follow your power supply maker’s instructions for which cable to use, and do not mix cables between different supply brands.
Power-connector failures drew a lot of attention after launch, and the guidance from manufacturers has centered on full insertion and bend clearance. Check the manufacturer’s current documentation for the latest installation advice for your specific card and cable.
Size, Cooling and Case Fit
The reference cooler occupies about three slots, and many partner designs are thicker and longer still. Before buying, measure the space inside the case for length, thickness and the extra room needed for the power plug on the side or top edge of the card. Narrow mid-tower cases and small-form-factor builds often cannot take the larger designs.
Weight is another factor. Heavy coolers can sag in the PCIe slot over time, so a support bracket is a sensible precaution when the card is mounted horizontally. Airflow also matters: a card of this class releases a lot of heat, so front intake and rear or top exhaust fans should be balanced and unobstructed.
Buying Used: Cautions and Checks
Because the card is no longer the newest flagship, many readers look at the second-hand market. A used 4090 can be a sound way to get 24GB, but the risks are real and worth working through before paying.
Inspect the card and connector
Ask for clear photos of the 16-pin socket on the card and the cable. Any browning, melted plastic or bent pins is a reason to walk away. Check the fan blades, the backplate and the display outputs for damage.
Confirm the warranty situation
Warranty coverage varies by brand and region, and some warranties do not transfer to a second owner. Ask for the serial number and the proof of purchase, then confirm the terms with the manufacturer rather than relying on the seller’s word.
Watch for scams and mismatched listings
Be wary of listings that are far below typical market levels, sellers who refuse to show the card running, or products described as a 4090 that turn out to be a different model. Use a payment method with buyer protection, and run a stress test and a memory check as soon as the card arrives, within the return window.
Who the RTX 4090 Suits
The card suits readers who need high-resolution gaming with ray tracing enabled, creators working with heavy video or 3D projects, and developers who want to run large models locally. For many other people a card one or two tiers down covers the need with a smaller case, lower power use and a gentler load on the power supply. Match the card to the resolution of your monitor and the software you actually run, not to the largest number on the box.
Related guides
- What Does Manage Gpu Performance Counters Mean On Nvidia Hardware
- Low Profile Nvidia Graphics Cards For Small Form Factor Pcs
- Nvidia App Error Your System Configuration Is Not Supported
Frequently Asked Questions
How much VRAM does the RTX 4090 have?
The reference RTX 4090 carries 24GB of GDDR6X memory on a 384-bit bus. Partner models use the same capacity, so the number to compare between brands is cooling, clocks and power limits rather than memory size.
What power supply does an RTX 4090 need?
NVIDIA’s published guidance has generally pointed to an 850 watt class unit for a typical system, with more headroom for powerful processors. Check your card maker’s current documentation, because partner designs can ask for a different minimum.
Is 24GB of VRAM overkill for gaming?
For most games at 1080p and 1440p, yes, since they use far less than 24GB. The extra capacity is more relevant to 4K and higher, heavy texture packs, creative software and local AI work.
Will an RTX 4090 fit in my case?
It depends on the model. Measure the available length, the thickness in slots, and the clearance for the 16-pin plug, then compare with the dimensions the card maker publishes for that exact version.
Write Your Review
No reviews yet. Be the first to share your experience!