For years, AMD and Nvidia have traded blows as the sole manufacturers of consumer-grade desktop graphics cards. Although Intel's Arc series brought Team Blue to the fray, the Nvidia GeForce RTX 40 series dominated our best GPU lists, and for good reason. Nvidia used CES 2025 to announce the GeForce RTX 50 series of GPUs, codenamed (and powered by) Blackwell. Named after the mathematician David Blackwell, RTX 50 brings to the table GDDR7 memory, PCIe 5.0 support, and advanced AI technologies for improving in-game performance.

Multiple Nvidia RTX 40 Super series GPUs from different manufacturers
Which GPU should you buy in each $100 price bracket?

There's something here for everyone, regardless of whether you're a budget user or a gaming enthusiast

1

When will Nvidia GeForce RTX 50 series GPUs be available?

Sooner than you think

Render of the RTX 5070 from Nvidia
Source: Nvidia

Nvidia announced the Blackwell generation of GeForce RTX 50 series graphics cards during CES 2025. They start shipping at the end of January 2025, though not all will be immediately available.

What Nvidia GeForce RTX 50 GPUs are launching?

There's an RTX 50 for everyone

An image showing a Zotac Gaming GeForce RTX 4070 Super Trinity Black Edition GPU's backplate and heatsink.

Nvidia is launching four Blackwell-powered GPUs to start with. The Nvidia GeForce RTX 5090 is the most powerful GPU available, at least according to marketing slides released by the company at CES 2025. To put into perspective just how much of a bump these cards are compared to RTX 40 predecessors, Nvidia claims the RTX 5070 will outperform the RTX 4090 ... through the power of artificial intelligence. DLSS has been a wonderful addition to gamer arsenals, allowing rigs to handle far higher settings than they typically would relying on raw computing power alone. Nvidia is turning all this up to 11 for Blackwell.

Price

CUDA Cores

TFLOPS

GDDR7

Power

Nvidia GeForce RTX 5090

$1,999

21,760

318

32 GB

575 W

Nvidia GeForce RTX 5080

$999

10,752

171

16 GB

360 W

Nvidia GeForce RTX 5070 Ti

$749

8,960

133

16 GB

300 W

Nvidia GeForce RTX 5070

$549

6,144

94

12 GB

250 W

Prices are as eye-watering as we expected. The flagship RTX 5090 will set you back $1,999, which will almost score you a decent set of wheels. The most affordable RTX 5070 is $549, though you can bet Nvidia will bring the 5060 to market to tackle AMD and Intel in the more affordable segment. Interestingly, the Intel Arc B580 is a solid performer at 1440p, costing just $249, so we should hopefully see some aggressive pricing from AMD and Nvidia. You may also require a new power supply if planning a PC around RTX 50 cards as the RTX 5090 can suck up to 575 W of power alone.

jensen-holding-blackwell
Nvidia's GeForce RTX 50 series promises RTX 4090 performance for $549

It's like an RTX 4090 only much more powerful and with considerably more RAM.

7

What's new with Blackwell?

Quite a lot actually

DeepCool PM500D non modular psu and cables side view

Many GPU enthusiasts were disappointed with the Ada Lovelace family because Nvidia decided to bring over the GDDR6X and GDDR6 memory from older GPUs. Thankfully, Nvidia made the switch to leverage GDDR7 memory. For reference, the GeForce RTX 4090 had a memory bandwidth of 1.01 TB/s, which is the same as that of the last-gen GeForce RTX 3090 Ti. In contrast, the memory bandwidth of the RTX 5090 is a whopping 1.79 TB/s, which is a staggering 79% upgrade! And you won't have any trouble storing data on these cards with the RTX 5090 having 32GB of VRAM to work with.

It's as we move through the stack do we encounter some issues with the 5070 having just 12 GB of VRAM. This isn't quite what we expected from Nvidia, as everyone and their grandparents have been calling on Nvidia to load its lower-end cards up with more memory. On the flip side, Nvidia has continued to utilize newer manufacturing processes over the last couple of GPU generations and the Blackwell series is no different. Team Green’s GPU family uses the TSMC 4N process node, based on the TSMC 5 manufacturing process. This brings notable improvements to efficiency and allows the company to push its cards harder.

Although the Ada Lovelace GPUs will work with PCIe 5.0 motherboards, they aren’t able to fully utilize the high-speed interface standard. However, the Blackwell GPUs can leverage the lightning-fast data transfer speeds provided by PCI Express Gen 5.0. Likewise, the GeForce RTX 50 graphics cards are a step up from their predecessors in terms of display capabilities. We've got the DisplayPort 2.1 standard, which offers much higher bandwidth, refresh rate, and resolution support than the last-gen DisplayPort 1.4. Intel also loaded up the B580 with DisplayPort 2.1 and I expect AMD to follow suit.

An image of an AMD GPU.
AMD Radeon RX 8000 series: All of the rumors so far

With Team Red reportedly planning to exit the high-end GPU market, the upcoming RX 8000 cards sound pretty tempting for mid-range gamers