Nvidia launches GeForce RTX 40 series GPUs that are up to 4x faster

During its GeForce Beyond keynote today, Nvidia introduced the GeForce RTX 40 series powered by the latest Ada Lovelace architecture. The Ada Lovelace GPU is built on a new 4nm process from TSMC, which Nvidia says is optimized for GPUs, and it packs a lot of improvements across its shader cores, RT cores, and Tensor cores. The GPUs also come with new GDDR6X memory.

Nvidia GeForce RTX 40 series specs

Right off the bat, Nvidia’s Ada Lovelace GPU comes with some impressive specs. On the shader side, the Nvidia Ada Lovelace GPU comes with new streaming multiprocessor that supports up to 83 TFLOPS of performance, and that’s for shading alone. It can deliver up to twice the performance of the existing Nvidia GPU, which is a pretty significant upgrade.

For ray tracing, the Nvidia GeForce RTX 40 series comes with third-generation RT cores, which promise up to 191 effect TFLOPS of ray tracing performance, an increase of 2.8x compared to the previous generation. Finally, for artificial intelligence, these GPUs come with fourth-generation Tensor Cores, which deliver a whopping 1.32 Tensor PFLOPS (petaflops), a massive 5x increase compared to the previous generation when using FP8 acceleration.

On top of all of this, Nvidia has enabled a new feature called Shader Execution Reordering. As the name suggests, this allows the GPU to reorder shader workloads on the fly in a way that takes better advantage of the GPU. Typically, workloads are delivered to the GPU sequentially, and not all of them use the full power of the GPU. By reordering them, these GPUs can process more of these workloads at once and use their full power to deliver better performance. Nvidia claims this improves ray-tracing performance by as much as 3 times, and it can increase frame rates in games by 25%.


An illustration depicting the process of using Shade Execution Reorderint to take full advantage of a GPU's performance

Despite this increase in performance, thanks to the new 4nm process designed in collaboration with TSMC, Nvidia is twice promising the power efficiency of the previous generation. That increase in performance shouldn’t come with a huge increase in power consumption, which is good news considering how power-hungry modern GPUs already are.

The GPUs also come with dual Nvidia Encoders (NVENC), which cut export times for videos in half, in addition to supporting AV1 encoding. Intel was the first to all support for AV1 encoding in its Arc GPUs, but now, Nvidia is catching up.

DLSS-3

One of the big new features of the Nvidia GeForce RTX 40 series is support for DLSS 3, the third generation of Nvidia’s deep-learning supersampling technology. DLSS 3 comes with even more improvements to increase framerates even further, overcoming CPU limitations by allowing the GPU to render entire frames all by itself using DLSS Frame Generation. In CPU-bound games like Microsoft Flight Simulatoryou can get twice the framerate thanks to DLSS Frame Generation.

There’s also the new Ada Optical Flow Accelerator, which allows the neural network in DLSS 3 to predict movement in a scene, increasing frame rates even further while maintaining image quality. Using DLSS 3, Nvidia says you can get up to four times the performance in games compared to traditional rendering techniques.

An illustration depicting all the key components of DLSS 3

On top of that, DLSS 3 integrates Nvidia Reflex, reducing system latency by allowing the CPU and GPU to work more closely together. Nvidia says this can double system responsiveness.

More than 35 games and apps have already announced support for DLSS 3, including Microsoft Flight Simulator, Cyberpunk 2077, and tools like Unity and Unreal Engine. Additionally, Nvidia also announced Portal RTX, a new version of Portal designed with ray tracing and DLSS 3 in mind.

Nvidia GeForce RTX 4090 and 4080

All of these features are part of the next-generation GPUs from Nvidia, which is debuting the the Nvidia GeForce RTX 4090, followed by the RTX 4080 a few weeks later. The Nvidia GeForce RTX 4090 promises up to four times the performance of the GeForce RTX 3090 Ti in fully ray-traced games thanks to DLSS 3, or twice the performance while using the same 450W of power.

Meanwhile, the Nvidia GeForce RTX 4080 comes in two versions, one with 16GB of memory and one with 12GB. For the 16GB model, Nvidia promises up to twice the performance of an Nvidia GeForce RTX 3080 Ti, and more performance than an RTX 3090 Ti while using less power. The 12GB version is still faster than the RTX 3090 Ti, but it uses more power to do so. Of course, this is all assuming you’re using DLSS 3. Here are the specs for these cards:

GPU model CUDA cores Memory Price
Nvidia GeForce RTX 4090 16,384 24GB GDDR6X $1,599
Nvidia GeForce RTX 4080 16GB 9,728 16GB GDDR6X $1,199
Nvidia GeForce RTX 4080 12GB 7,680 12GB GDDR6X $899

The Nvidia GeForce RTX 4090 will be the first one to launch, coming on October 12th. The two RTX 4080 configurations will be available in November. Nvidia says you can expect partners such as Asus, Colorful, MSI, and Zotac to also launch their own versions of these cards. As you might have heard, EVGA is no longer working with Nvidia, so don’t expect to see anything from that company.

Nvidia RTX 6000

If you’re not a gamer, Nvidia’s Ada Lovelace is debuting in the new Nvidia RTX 6000 professional GPU. This GPU comes with 48GB of GDDR6 memory and all the improvements of Ada Lovelace. It promises up to twice the throughput of the previous generation from its RT cores, and the fourth-generation Tensor Cores also twice deliver the performance in AI training. Additionally, for XR workloads, Nvidia is touting up to 3x the video encoding performance.


Nvidia RTX 6000 GPU

The Nvidia RTX 6000 will be available starting in December, though Nvidia didn’t mention pricing.

Leave a Comment

Your email address will not be published.

S-F-U