As expected, Nvidia revealed its new RTX 50 series desktop graphics cards during its CES 2025 keynote, which took place in the early hours of the UK. The current lineup includes four cards: a $1,999/£1,939 RTX 5090, a $999/£979 RTX 5080, a $749/£729 RTX 5070 Ti, and a $549/£539 RTX 5070. Each “Blackwell” graphics card is equipped with GDDR7. memory, impressive frame rate claims, and DLSS 4 multi-frame generation. The two high-end cards will arrive on January 30, while the high-end offerings are scheduled for February.
Claims of “relative performance” for some of these graphics cards are far-fetched, so let’s start with an idea of the expected frame rates for each card and a quick explanation of DLSS 4 multi-frame generation before getting into the architectural improvements.
The $1,999 RTX 5090 is unsurprisingly positioned as the ultimate graphics card, a prosumer model with 32GB of GDDR7. It is based around the GB202 GPU with 92 billion transistors, compared to 76 billion on the 4090, and includes 21,760 CUDA cores. Nvidia’s slides promise double the frame rates of the RTX 4090, or comfortable 4K 240Hz with full RT and maximum settings in games that support the Multi-Frame Generation (MFG) feature.
To explain before going further, MFG adds up to three generated frames for each traditionally rendered frame. This results in higher frame rates and therefore visual fluidity, but not necessarily higher. performance as we generally call it, because latency depends on the “base” frame rate rather than the final “output” frame rate. The concept is similar to the original Frame Generation (FG) functionality on the RTX 40-series cards and, as such, leverages developer onboarding and Reflex latency mitigation technology 2 from Nvidia to work properly.
Surprisingly, while the RTX 5090 has a total rated graphics power of 575W – right against the PCIe 5.0 limit of 600W and 125W more than the RTX 4090 – the Founders Edition design is actually thinner than the RTX 4090 FE . It’s a two-slot board that can actually fit into small form factor PC cases (with the required 1000W PSUs), with two continuous-circulation fans and the PCB centrally located. It will be fascinating to see how the new design works, considering that early prototype 5090 coolers were apparently four-slot monsters!
The $999 RTX 5080 is designed to double the frame rates of the RTX 4080, with Nvidia promising beyond 2x results in games like Cyberpunk 2077, Alan Wake 2, and Black Myth: Wukong – all RT-heavy titles. The 5080 is based on the smaller GB203 chip and comes with 16GB of GDDR7, 10752 CUDA cores and a power rating of 360W – with an 850W PSU recommended. The price matches that of the RTX 4080 Super, so it will be interesting to see how these models compare in terms of apples-to-apples performance when FG or MFG is not used.
| Nvidia | RTX5090 | RTX5080 | RTX 5070Ti | RTX5070 |
|---|---|---|---|---|
| GPU | GB202 | GB203 | GB203 | GB205 |
| Cores | 21760 | 10752 | 8960 | 6144 |
| Increase clock | 2.41GHz | 2.62GHz | 2.45GHz | 2.51GHz |
| TOPS tensor core | 3352 | 1801 | 1406 | 988 |
| RT-core TFLOPS | 318 | 171 | 133 | 94 |
| Memory | 32 GB GDDR7 | 16 GB GDDR7 | 16 GB GDDR7 | 12 GB GDDR7 |
| Memory bus width | 512 bits | 256 bits | 256 bits | 192 bits |
| Memory bandwidth | 1,792 GB/s | 960 GB/s | 896 GB/s | 672 GB/s |
| Total graphics power | 575W | 360W | 300W | 250W |
| PSU Recommendation | 1000W | 850W | 750W | 650W |
| Power connector | 600 W PCIe 5.0 (4x 8 pin) | 450 W PCIe 5.0 (3x 8 pin) | 300 W PCIe 5.0 (2x 8 pin) | 300 W PCIe 5.0 (2x 8 pin) |
| Price | $1999/£1939 | $999/£979 | $749/£729 | $549/£539 |
| Release date | January 30 | January 30 | FEBRUARY | FEBRUARY |
February’s two cards, the RTX 5070 Ti 16GB and 5070 12GB, hit prices of $749 and $549. The 5070 Ti uses a scaled-down version of the GB203 GPU with 8,960 CUDA cores, while the 5070 uses the GB205 with 6,144 cores. It’s disappointing to see the 12GB of GDDR7 on the $549 RTX 5070 card in the year of our lord 2025, but Nvidia should have equipped the 5070 with a 256-bit bus (or used a mixed VRAM allocation) to reach 16 Go so that the move is not unexpected.
In terms of performance, it’s the same theme: the 5070 Ti is supposed to double the frame-rate of the 4070 Ti, while the 5070 does the same for the 4070. However, Nvidia also pointed out that this means the 5070 should therefore offer the same frame rates as the RTX 4090, which would be a huge boost at around a third of the price! Since this relies on an uneven comparison between MFG and FG, we will have to wait for proper third-party testing to see if this claim is truly confirmed and what the difference looks like in non-FG and even non-DLSS/non-models. RT titles.
In addition to confirming its desktop lineup, Nvidia also announced RTX 50 series laptop GPUs. These will arrive starting in March and will include smaller memory allocations. These include the RTX 5090 24GB, RTX 5080 16GB, RTX 5070 Ti 12GB, and RTX 5070 8GB.
Manage cookie settings
Now that we’ve covered the lineup, it’s worth quickly mentioning some of the architectural changes that contribute to both faster overall performance and key gains in RT and AI performance. These improvements come from more robust SM with higher throughput and tighter integration of the Tensor core, reordering improvements, and acceleration of shader execution for FP4 compute. There is also improved hardware flip measurement needed for multi-frame generation, new RT cores with doubled ray triangle intersection rate, and better compression. The cards also support the PCIe 5.0 interface and are equipped with DisplayPort 2.1b UHBR20 ports, matching AMD’s previous generation cards.
If you’re interested in learning more about the Blackwell architecture, Nvidia’s blog has more details on these architectural details, and we expect the usual tech press like Gamers Nexus or TechSpot to offer more detailed information on these changes soon. once they have the opportunity. to digest the information.
As for our results, we plan to release more information on the DLSS 4 multi-frame generation in the near future, so watch out for that, and of course we can’t wait to submit the new GPUs to our suite of refined testing as soon as possible. .
