NVIDIA's Performance Claims and Parting Thoughts

NVIDIA as part of its launch press deck and technical deep-dive provided various first-party performance claims. We begin with the ones related to the cooling solution. Thanks to the various silicon-level and architecture-level advancements on the RTX 5090, paired with the new Double Flow Through cooling solution, NVIDIA claims to have scored a technological leap in terms of how loud the RTX 5090 can get for the kind of power it can draw.


The NVIDIA GeForce RTX 5090 is an absolute beast on paper, unlocking experiences never thought possible, such as 4K @ 240 Hz, or gaming at 8K with 60 FPS, all thanks to the new NVIDIA DLSS 4 and Neural Rendering technologies. The card provides a gargantuan 3,352 AI TOPS, or about 75 times that of a Copilot+ NPU, 105 TFLOP/s FP32, and 318 RT TFLOP/s. The card comes with 32 GB of video memory across a mammoth 512-bit GDDR7 memory interface, with a speed of 28 Gbps, for 1.792 TB/s of memory bandwidth. It can max out the most demanding of today's games at 4K.


DLSS 4 Multi Frame Generation adds a whole new paradigm to this product. With it enabled, the card is able to more than quadruple its frame rate, making it capable of 4K @ 240 Hz, with reduced latency from NVIDIA Reflex.


With the previous generation GeForce RTX 4090 as the baseline, this is how NVIDIA claims the RTX 5090 fares, with the latest available DLSS and or RT technologies enabled on both cards.


The new generation Blackwell streaming multiprocessor (SM), along with its 5th Gen Tensor core lay the hardware groundwork for neural sharing, a revolutionary concept where generative AI works collaboratively with conventional raster 3D graphics rendering, rather than for generating intermediate frames.


NVIDIA knows more than anyone just how memory sensitive generative AI workloads can be, especially when the GPU is trying not just to accelerate it, but also render real time 3D graphics in tandem. To give both sufficient bandwidth, the company opted for the fastest graphics memory technology available, GDDR7, which leverages PAM3 signaling to yield data rates as high as 28 Gbps, double that of the first graphics cards to implement GDDR6.

NVIDIA has knocked it out of the park with the industrial design of its new GeForce RTX 5090 Founders Edition. None of its AIC partners has a card this sleek, from what we've seen, and if NVIDIA's claims on cooler noise hold up, then it would be a no-brainer to choose this card over custom designs. More so, because NVIDIA will offer this card at its baseline price of USD $1,999. NVIDIA's performance claims for the RTX 5090 see it perform along expected lines, with a roughly 35% gen-on-gen increase in raster 3D performance without DLSS, over the RTX 4090. DLSS 4 Multi Frame Generation is exclusive to Blackwell, and adds a whole new dimension to the product, more than quadrupling frame-rates. We hope you are thoroughly appetized with what you've seen so far. We can't wait to tell you more, later this week.