NVIDIA Unveils GeForce RTX 50 Blackwell, Up To 3X Sooner Than RTX 4090

It was again in September of 2022 that NVIDIA first revealed the GeForce RTX 4000 sequence. The GeForce RTX 4090 has been the reigning GPU efficiency champ, primarily unchallenged, ever since then. It simply abdicated the throne to its successor, although: the GeForce RTX 5090 is right here, bringing with it the patron model of the Blackwell GPU structure.

Introduced tonight at NVIDIA’s keynote in a surprisingly windy Las Vegas, the GeForce RTX 50 sequence represents NVIDIA’s newest efforts at bringing its usually datacenter-focused chip growth efforts again house to avid gamers demanding the newest rendering know-how and the very best GPU efficiency. So with out additional ado, let’s truly speak in regards to the GPUs.

Clearly, first up is the GeForce RTX 5090, and crucial specification for lots of you is the worth: $1999 USD. That is a rise over even the $1599 of the GeForce RTX 4090, but it surely’s not simply inflation accountable, as the brand new GPU is a whopping 92 billion transistors. You may simply get your cash’s value, although, because the specs are staggering: 170 Streaming Multiprocessors (up from 128 in RTX 4090), and a full 32GB of GDDR7 reminiscence operating at 28 Gbps. That delivers a complete reminiscence bandwidth only a hair beneath 1.8 TB/second, roughly an 80% improve from the earlier era.

Thanks nearly completely to the massive dimension of the GPU and architectural adjustments, the GeForce RTX 5090 apparently provides 3400 AI TOPS—absolutely 2.5 instances that of the GeForce RTX 4090, going by NVIDIA’s personal numbers. In fact, AI TOPS aren’t that related to gaming workloads, which is what nearly all of you expensive readers are most likely most serious about. NVIDIA did not produce any

Along with the top-end GeForce RTX 5090, NVIDIA additionally introduced the GeForce RTX 5080, RTX 5070 Ti, and RTX 5070. The GeForce RTX 5080 is actually one-half of a GeForce RTX 5090, with 16GB of RAM on a 256-bit bus, and it comes with an identical price ticket of $999. It does function barely larger clock charges on its reminiscence—a dizzying 30 Gbps—so the efficiency might be a bit greater than half of the flagship.

The GeForce RTX 5070 Ti is an extra step down at $749; it additionally comes with 16GB of RAM, after which the GeForce RTX 5070 with 12GB of GDDR7 reminiscence is attractively priced at $549. NVIDIA claims that even the GeForce RTX 5070 is quicker than the GeForce RTX 4090, though we expect that is more likely to be each in a path-traced workload and taking DLSS 4 under consideration.

Certainly, DLSS 4 is actual and NVIDIA says that it has moved away from convolutional neural networks towards a brand new new transformer mannequin to extend each picture high quality and efficiency. Nonetheless, the large function this time round is multi-frame era. DLSS 4 is able to growing body charges significantly by producing as much as three frames between the precise classically-rendered frames. Jensen Huang says that out of 4 full-resolution 4K output frames—roughly 33 megapixels of picture information—solely 2 megapixels must be computed historically.

That is not the one AI function NVIDIA premiered tonight. RTX Neural Shaders are small AI networks embedded inside shader code. NVIDIA says that they’ll enhance each picture high quality and efficiency with a smaller reminiscence footprint in comparison with typical shader code, like with NVIDIA’s Neural Texture Compression. Additionally, NVIDIA RTX Neural Face seems to be a brand new middleware library that makes use of AI to enhance the realism of life-like, interactive faces rendered in real-time.

Arguably probably the most thrilling new function of ‘Blackwell’ is definitely the introduction of one thing known as RTX Mega Geometry. NVIDIA claims that this function permits the GeForce RTX 50 sequence to ray-trace one-hundred instances the triangles of Ada Lovelace GPUs. We’re looking forward to extra particulars on Mega Geometry, however Huang did not truly point out it in his transient GeForce RTX 50 presentation.

Lastly, NVIDIA introduced the second era of its Reflex enter lag discount know-how. Supposed to be used at the side of DLSS 4, that is supposedly the key sauce that permits the multi-frame era know-how to work with out making each recreation really feel like a whole lag-fest. The unique model of Reflex provides tangible advantages by way of enter lag discount, so we’re keen to check NVIDIA’s claims right here.

Unusually, NVIDIA additionally revealed the cell GeForce RTX 50 lineup tonight. These vary from the top-end GeForce RTX 5090, outfitted with 24GB of reminiscence and delivering 1824 TOPS, all the way down to the GeForce RTX 5070 with 8GB of RAM and producing 798 TOPS. As with the earlier era, it seems like NVIDIA is actually utilizing GPUs one tier down for the cell elements, though the GeForce RTX 5090 (cell) would not map on to any of the desktop elements, so it is going to be attention-grabbing to see for positive.

So far as the desktop GPUs go, all of those elements are on the best way to retailer cabinets this month. In the meantime, laptops outfitted with the brand new GeForce RTX 50 cell GPUs will likely be coming round towards the top of this quarter, or in March.