Ada Lovelace brings new 3rd generation RT Cores and 4th generation Tensor cores, both of which bring insane improvements. This architecture is 4x faster in ray-tracing and has a 2x uplift in performance/watt compared to Ampere. NVIDIA says ray tracing performance has improved 16x over just four years in contrast to Turing. The company is also launching the all-new DLSS 3 that’s exclusive to Ada Lovelace.
The World’s Fastest GPU
The RTX 4090 is built on the AD102 GPU, which is unparalleled in terms of performance, however it’s not the full-fat version of the chip. NVIDIA is saving that for later. Regardless, the GPU features a massive 24GB of Samsung-made GDDR6X memory running at 24Gbps, amounting up to an effective bandwidth of 1152GB/s across a 384-bit bus. Packed with 16384 CUDA Cores, this powerhouse has 75 billion transistors in total, more than 2.5x than the RTX 3090 Ti. The on-paper specifications already crush world records with the raw FP32 Compute Performance clocking in at 90 TFLOPs. Being built on TSCM‘s cutting-edge N4 (5nm) process node, the clock speeds go as high as 2.52GHz with base speeds of 2.23GHz . All this power does come at the cost of, well, power. The RTX 4090 has a TDP of 450W and will require at least a 850W power supply to boot. Arriving on the 12th of October, the RTX 4090 is all yours for $1599.
The Design
The RTX 4090 FE (Founders Edition) has almost the exact same design as the RTX 3090 FE. The only difference lies in the front fan which overlaps the bounds of the GPU. In addition, we are greeted by a new font namely NVIDIA Sans NALA which has a slightly reduced thickness as shown in the image below: It is true that fortune favors the bold, however, there is no need to fix something that isn’t broken. A complete redesign would have met with unwanted criticism. Besides, thermal testing is also a major hurdle when designing a GPU. So that is why NVIDIA opted to stick with the old Ampere GPU design. The RTX 4090 features a simplistic design akin to last-Gen with a single fan on the front. The cooler is much more beefier as mentioned above with the GPU spanning almost 3-4 slots. Many AIB partners have also taken the stage and showcased their next-Gen AIB models for the RTX 4000 series already. ASUS has an all-new design for their RTX 40-series cards, including an overhauled ROG STRIX variant that looks even more beastly and somehow gamer-y than the current generation.
Performance
In the performance category, NVIDIA left no loose ends with the RTX 4090 outclassing every GPU in its league by almost twice the numbers. The RTX 4090 as compared to the RTX 3090 Ti is 2x faster in Flight Simulator, 3x faster in Portal RTX and 4x faster in Racer RTX (simulation demo). The actual raster performance has increased by 1.5-2.2x against the RTX 3090 Ti. However, the new DLSS 3.0 takes that lead up to 4x showcasing the importance of NVIDIA’s AI upscaling solution in demanding games.
DLSS 3.0
NVIDIA has introduced a new version of DLSS named ‘DLSS 3.0’ which is a major jump as compared to DLSS 2.0. DLSS 3.0 renders entire new frames instead of just pixels offering 4x better performance over traditional rasterization. Jensen mentioned that DLSS 3 is such a major leap that often CPUs are not able to keep up with the framerates resulting in a CPU-bound scenario even in AAA titles. Sadly DLSS 3.0 will not officially support the Turing and Ampere platforms and will be exclusive to the latest RTX 40-series only. DLSS 3.0 is supported by 35 games with the first few coming as soon as October.
Dual AV1 Encoders
The RTX 4090 features 2 of NVIDIA’s new 8th Gen NVIDIA Encoders (NVENC) with added supported for AV1 encoding. As per NVIDIA, AV1 improves encoding efficiency by 40% along with 40% more bitrate. To top it off, Discord is getting AV1 support later this year.
RTX Improvements
NVIDIA since Turing (RTX 2000), has seen a 16x increase in RT performance in just 4 years. Ada Lovelace is 4x faster in Raytracing games as compared to Ampere.
Worth Your Money?
All in all, the RTX 4090 is a massive upgrade over last-gen complying with the initial test results. NVIDIA opted for TSMC’s 5nm process node which is much more expensive, not to mentioned advanced, than Samsung‘s 8nm. Even after this, the price is increased by just $100, which is not much if you’re already paying $1500+ for a GPU. In addition, the new DLSS 3.0 will be a great addition to your arsenal in the battle for every last frame. Therefore, the RTX 4090 is looking like a great choice if you’re looking for high-end performance at a ‘reasonable’ price. But if the price tag is a little too steep for you, NVIDIA also launched the RTX 4080 starting from just $899. That GPU offers performance on par with the RTX 3090 Ti and even comes in two different variants to suit your needs.
