leaks hint at a monstrous RTX 5090

The release of Nvidia’s next RTX 50X0 graphics cards is now fast approaching, and all leakers are on alert. This is particularly the case of kopite7kimia pseudonym well known to hardware enthusiasts since it is one of the most reliable sources in the industry. Recently, it returned to business with a series of revelations on the technical specifications of the future RTX 5080 and RTX 5090: here’s what to expect for the next high-end GPUs from the green giant.

kopite7kimi had already started to distill some information on this new generation called Blackwell, in homage to the eminent mathematician David Blackwell. Last September, he notably revealed the theoretical energy consumption of these future cards – and unsurprisingly, they were astronomical, with a TGP of 400W for the RTX 5080 and… 600W for the RTX 5090.

More recently, in a series of tweets spotted by PCWorld, he revealed the rest of the forecast technical sheet for these two models. And if these predictions are correct, we can expect a power gain that is anything but negligible.

Twitter-tweet” data-width=”500″ data-dnt=”true”>

GeForce RTX 5090
PG144/145-SKU30
GB202-300-A1
21760FP32
512-bit GDDR7 32G
600W

— kopite7kimi (@kopite7kimi) September 26, 2024

Real monsters of power

The RTX 5090 will apparently be the first consumer card (relatively speaking, obviously) to pass the symbolic bar of 20,000 CUDA cores – the computing subunits that take on the majority of graphics rendering tasks. According to kopite7kimi, she will embark 21,720 CUDA coresan increase of almost 25% compared to the 16,384 of the RTX 4090. A more than substantial gain which, in practice, should manifest itself as a largely perceptible increase in raw performance.

The move upmarket is, however, less impressive on the 80 series models (10,752 cores for the RTX 5080 compared to 9,728 for the 4080).

The trend is the same for streaming processors, the elements which manage the distribution of tasks among the CUDA cores. Nvidia’s future flagship will apparently have some 170, compared to 128 for the RTX 4090. The RTX 5080, for its part, will carry 84 – 8 more than the 4080. Again, this is a significant increase which should allow these cards to manage very large amounts of data in parallel. Good news for fans of very high resolution gaming with raytracing, but also for all professionals who will use it to process scientific data, train AI models, and so on.

Same observation regarding memory. The Blackwell generation will be the first to accommodate GDDR7 memory, with its monstrous theoretical bandwidth of 32 GT/s – 60% more than the GDDR6 of the RTX 40X0. In addition, the size of the memory bus that connects the GPU to its VRAM increases, with 512 bits for the RTX 5090 compared to 448 for the 4090. This should manifest itself in a significant performance gain, and even lower latency than before. .

If kopite7kimi’s information is accurate, which has almost always been the case in recent generations, the future flagship will therefore offer quite astonishing performance; no worries in terms of raw power. This is less obvious for the RTX 5080, where the gain seems much more modest.

What about the Price?

The only missing data is unfortunately the price; impossible to know what to expect at the moment. But we can still expect an increase at least as significant as that of performance. Nvidia still has a near-monopoly in this segment, it will have absolutely no reason to be aggressive on pricing, and it would not be surprising if the RTX 5090 becomes the first “consumer” card to cross the bar 2000 € at the exit. To find out more, we’ll see you at CES next January.

🟣 To not miss any news on the WorldOfSoftware, , .