NVIDIA's "Blackwell" era has officially arrived, succeeding the RTX 4000 "Ada" generation that has been around since 2022. For some, it's a bittersweet goodbye; for others, including the owners of RTX 4060 GPUs, it's a welcome change.
NVIDIA's new RTX 5000 graphics cards were revealed at CES 2025 alongside some impressive claims — an RTX 5070 can outperform an RTX 4090?! — and some less impressive price tags. Let's dig into the information shared by NVIDIA at its keynote speech, as well as take a look at when you can expect these GPUs to land and where you can buy them on launch day.
Recent updates
January 9, 2025: I've updated this guide with all of the new RTX 5000 information revealed at CES 2025. — Cale Hunt
NVIDIA officially announces its new RTX 5000 GPUs
NVIDIA CEO Jensen Huang officially kickstarted CES 2025 with a keynote speech on January 6, where he revealed the company's next generation of "Blackwell" RTX 5000 graphics cards.
The flagship RTX 5090 and its less powerful RTX 5080 sibling are expected to launch first on January 30, 2025, with the RTX 5070 Ti and RTX 5070 following up shortly after in February.
Stripping back the hardware to reveal the Blackwell GPU, NVIDIA lists 92 billion transistors, 4,000 AI TOPS, 380 ray-tracing TFLOPS, and 125 shader TFLOPS. Those are massive numbers. Compared to the last-gen Ada GPUs, AI TOPS have tripled while RT FLOPS have doubled.
Huang made a bold claim, stating that the RTX 5070 can achieve RTX 4090 performance, all for well less than half the price. However, he followed that statement by saying it's all impossible without AI, which isn't too surprising.
All-new DLSS 4 and Reflex 2 with Frame Warp were revealed to be the main drivers behind the massive performance boost. Together, they push the capabilities of these GPUs far beyond what the physical hardware can actually do.
Is that a good thing? It depends on how you look at it. There are concerns floating around about visual fidelity and latency suffering with such a heavy reliance on frame generation, but I'm ultimately reserving judgment until we get the new cards in for testing.
NVIDIA RTX 5000 specifications
NVIDIA has announced the RTX 5090, RTX 5080, RTX 5070 Ti, and RTX 5070 GPUs for both desktop and laptop PCs.
The flagship RTX 5090 card unsurprisingly has the most impressive specs, arriving with 32GB of GDDR7 VRAM with 1.792TB/s memory bandwidth. That's nearly twice as much bandwidth as the RTX 4090, with more and faster GDDR76 VRAM.
Here's a look at the four new cards announced by NVIDIA.
| Spec | NVIDIA RTX 5090 | NVIDIA RTX 5080 | NVIDIA RTX 5070 Ti | NVIDIA RTX 5070 |
|---|---|---|---|---|
| Architecture | Blackwell | Blackwell | Blackwell | Blackwell |
| Memory | 32GB GDDR7 | 16GB GDDR7 | 16GB GDDR7 | 12GB GDDR7 |
| Memory Bandwidth | 1.792TB/s | 960GB/s | 896GB/s | 672GB/s |
| Memory Bus | 512-bit | 256-bit | 256-bit | 192-bit |
| CUDA Cores | 21,760 | 10,752 | 8,960 | 6,144 |
| RT TFLOPS | 318 | 171 | 133 | 94 |
| AI TOPS | 3,352 | 1,801 | 1,406 | 988 |
| NVIDIA Encoder (NVENC) | 3x 9th Gen | 2x 9th Gen | 2x 9th Gen | 1x 9th Gen |
| NVIDIA Decoder (NVDEC) | 2x 6th Gen | 2x 6th Gen | 1x 6th Gen | 1x 6th Gen |
| TGP | 575W | 360W | 300W | 250W |
| Required PSU | 1,000W | 850W | 750W | 650W |
How much do NVIDIA RTX 5000 GPUs cost?
With all of the leaks and rumors coming out ahead of NVIDIA's official announcement, most PC enthusiasts were only waiting to hear the price of the new Blackwell GPUs.
The RTX 5090 is unsurprisingly the most expensive at a $1,999 MSRP. The RTX 5080 drops to half that at $999, with the RTX 5070 Ti following up at $749.
The RTX 5070, which NVIDIA claims can match the performance of an RTX 4090 in part thanks to all the new AI advancements, comes in at $549.
Keep in mind that these are the prices NVIDIA has set for its reference cards. Custom RTX 5000 GPUs built by NVIDIA's partners will likely come in at different prices, with some higher and some lower than NVIDIA's MSRP.
When do NVIDIA RTX 5000 GPUs launch?
The RTX 5090 and RTX 5080 are expected to launch first on January 30, 2025. That includes NVIDIA's Founders Edition reference cards and custom cards from partners like ASUS, Colorful, Gainward, GALAX, Gigabyte, INNO3D, KFA2, MSI, Palit, PNY, and ZOTAC.
Laptops with mobile versions of these two GPUs are expected from March 2025. Expect to see models from Acer, ASUS, Dell, Gigabyte, HP, Lenovo, MECHREVO, MSI, and Razer.
The RTX 5070 Ti and its RTX 5070 sibling are expected to launch sometime in February 2025.
What is NVIDIA's DLSS 4?
NVIDIA's Deep Learning Super Sampling (DLSS) technology has been around since the launch of the first RTX GPUs in 2018, and as the name suggests, it's been learning and improving the whole time. DLSS 4 is available to any users of RTX 20-series through 50-series cards, but the available features differ due to hardware performance.
For example, the latest RTX 5000 GPUs are the only ones to get the new DLSS 4 Multi Frame Generation. Enhanced DLSS Frame Generation will also be available for RTX 50- and 40-series GPUs only. Older cards get access to enhanced DLSS Ray Reconstruction, DLSS Super Resolution, and Deep Learning Anti-Aliasing (DLAA).
So what exactly is the new DLSS 4 Multi Frame Generation? It's an improvement on the DLSS Frame Generation tech first accompanying DLSS 3, relying on the Blackwell GPU's specialized architecture to boost neural rendering abilities. Thanks to a new AI model, Multi Frame Generation takes the single generated frame and multiplies it.
NVIDIA says this results in "consistent and optimal frame pacing" despite the extra frames being generated. This new tech relies heavily on the fifth-gen Tensor cores found in the Blackwell GPUs, which is why the feature isn't coming to RTX 4000 cards. The tool also lowers VRAM usage, which should be a good thing considering NVIDIA didn't wow anyone with VRAM amounts.
NVIDIA used Cyberpunk 2077 running with full ray tracing to highlight the improvements. The game sits around 25-30 FPS with DLSS disabled. With DLSS 4 enabled, the frame rate climbs to 235 and beyond, even with ray tracing enabled.
More than 75 games and apps that are already compatible with NVIDIA's frame generation will be compatible with DLSS 4's Multi Frame Generation on launch day, with "many more on the way."
What is NVIDIA Reflex 2 with Frame Warp?
NVIDIA's Reflex technology was introduced in 2020 as a tool to reduce latency. In general, it works by syncing the CPU and GPU, resulting in your input actions being recognized faster. That's especially important in competitive games, but it can benefit any gamers. NVIDIA says that more than 90% of gamers use Reflex (when available).
New for 2025 is NVIDIA Reflex 2. It takes a different approach to lowering latency, which NVIDIA calls Frame Warp. Here's an explanation of how it works from NVIDIA itself:
"Reflex 2 Frame Warp takes this concept from research to reality. As a frame is being rendered by the GPU, the CPU calculates the camera position of the next frame in the pipeline, based on the latest mouse or controller input. Frame Warp samples the new camera position from the CPU, and warps the frame just rendered by the GPU to this newer camera position. The warp is conducted as late as possible, just before the rendered frame is sent to the display, ensuring the most recent mouse input is reflected on screen."
Shifting and warping a frame thus results in missing pixels, which are filled in with a "latency-optimized predictive rendering algorithm." A lot of the technical stuff starts to go over my head, but the bottom line? You should expect an easier time aiming and tracking enemies in competitive play.
