Image Credit: NVIDIA
While NVIDIA’s Blackwell technology has been in data centers for a while now, we have yet to lay eyes on the consumer version of this massive silicon slab. The upcoming 50-series of graphics cards are still largely a mystery, but as we get closer and closer to their official reveal, more potentially true rumors come to the fore. So here’s the most plausible stuff we think will feature in the 50-series.
A Faster Release Process
For some time now, NVIDIA has started off revealing and releasing the top two cards of the first wave, and then spacing out the release of the mid-rang and entry-level cards. So you’d expect that we’d see the 5090 and 5080 first, and then the 5070, 5060, and 5050 would be released over the following months.
However, some rumors are suggesting that we’ll see all the cards revealed and released together, or much closer together. We might be looking at:
January 2025: Launch of the RTX 5090 and RTX 5080.
February 2025: Introduction of the RTX 5070 and possibly the RTX 5070 Ti.
March 2025: Release of the RTX 5060 and RTX 5060 Ti
It’s possible that we might still get a late 2024 reveal, but that seems less likely now that we’re at the end of the year without an event announced by NVIDIA.
Spec Rumors
There are numerous “leaks” and speculation about the sorts of specs we’ll be seeing with the 50-series, but we’ve tried to tabulate the most credible ones here.
Model | CUDA Cores | VRAM | Memory Bus | TDP (W) | Architecture | Chip |
RTX 5090 | 21,760 | 32GB GDDR7 | 512-bit | 600 | Blackwell | GB202 |
RTX 5080 | 10,752 | 16GB GDDR7 | 256-bit | 400 | Blackwell | GB203 |
RTX 5070 | ~8,000 | ~12GB GDDR7 | 192-bit | 220 | Blackwell | GB205 |
RTX 5060 | ~6,000 | ~8GB GDDR7 | 128-bit | 170 | Blackwell | GB206 |
RTX 5050 | ~4,000 | ~6GB GDDR6 | 128-bit | 100 | Blackwell | GB207 |
As you can see, this time around it seems there might be a much larger gap between the top two cards. If the spec rumors are in fact true, the 5090 will be almost twice the card the 5080 is. From there, the scaling is more gentle.
Unfortunately, rumors suggest that 8GB and smaller cards will still exist in the Blackwell lineup, and that’s one rumor we sincerely hope isn’t true. VRAM is cheap, and NVIDIA should offer no less than 12GB of it, even on the lowest end card, in our opinion.
Power consumption might see a bump across the board, and the 5090 in particular might have a massive appetite for wattage, but with potentially 50-70% better performance than a 4090. We expect a new 16-pin connector for 50-series cards, which is good news because the VHPWR connector has been a bit of a disaster.
It’s All About AI
Overall, the 50-series probably won’t be a massive leap in general performance, but we expect that NVIDIA’s R&D into AI will pay off in dividends for these cards. We expect AI-based features like DLSS and frame generation to get a decent improvement, but also local AI jobs such as running LLMs and image generation.
However, with rumors of those small VRAM allocations, that could put a damper on AI tasks on anything below a 5080, which would be a real shame.
As for the laptop variants of these GPUs, so far there’s been little to nothing leaked or rumored, but we expect once the first wave of cards are official, those details will also begin to emerge.
