The NVIDIA GeForce RTX 40 series has finally been announced. Featuring a new Ada Lovelace architecture and manufactured on the custom TSMC 4N node, the GeForce RTX 40 series is up to 4X faster than last-gen, and up to 2X higher efficiency. Ada Lovelace packs quite a number of new things, including the new 3rd Gen RT cores, 4th Gen Tensor cores and new Streaming Multiprocessors. Let’s get into it.
NVIDIA Ada Lovelace: Beyond Fast
2X+ shader throughput
NVIDIA Ada Lovelace is a pretty massive leap over Ampere. It offers more than double the peak throughput with 83 Shader TFLOPs from the GeForce RTX 4090, versus the GeForce RTX 3090 Ti’s 40 Shader TFLOPs. Helping this along is the new Shader Execution Reordering (SER) technology that dynamically reorganizes these inefficient workloads into more efficient ones to improve shader performance by up to 2X, and frame rates by up to 25%.
2.8X peak raytracing throughput
The new 3rd Gen RT cores in Ada Lovelace deliver 2x the ray-triangle intersection throughput, which bumps up peak RT-TFLOP performance by up to 2.8X. The GeForce RTX 4090 packs 191 RT-TFLOPs, a good 2.4X bump over the GeForce RTX 3090 Ti. The GeForce RTX 40 series will also feature Opacity Micro-Map Engines and Micro-Mesh Engines, accelerating performance when used together with new software techniques to enable faster raytracing even as raytraced scenes get increasing complex and detailed.
5X Tensor throughput
Along with the new shaders and RT cores, NVIDIA are also packing in the latest 4th Gen Tensor cores into Ada Lovelace. With an all-new 8-bit FP8 Tensor Engine, NVIDIA increased throughput by 5X. This in turn enhances performance in DLSS-enhanced titles, on top of the AI-powered noise reduction features in NVIDIA Broadcast. Versus the last-gen GeForce RTX 3090 Ti’s 320 Tensor-TFLOPs, the GeForce RTX 4090 is over 4X faster, doling out 1.32 Tensor-PFLOPs. Yes, petaFLOPs. This is arguably the largest upgrade coming from Ampere, which lets NVIDIA claim the huge leap in efficiency.
Speaking of DLSS, NVIDIA also released NVIDIA DLSS 3 with Frame Generation technology, powered by Ada Lovelace’s new Optical Flow Accelerator. Pixel motion data from subsequent frames are fed into the DLSS neural network to generate new frames on the GPU itself, ensuring up to a 2X performance bump, even when dealing with a CPU bottleneck.
2X faster video encoding
Ada Lovelace will also up the bar for content creators and streamers with two 8th Gen NVENCs, delivering twice the performance versus last-gen. Content creators who rely on Davinci Resolve, Premiere Pro and even Jianying, will get up to 2X faster encoding thanks to the dual NVENC encoders. Pretty nice. On top of that, the new encoders support AV1 encoding, which lets you stream at 40% higher quality at the same bitrate. With OBS and Discord adding AV1 support this year, GeForce RTX 40 series users will be looking at improved quality streaming and video calls.
NVIDIA GeForce RTX 40 Series starts from RM4150
Interestingly enough, it seems that NVIDIA isn’t intending to entirely replace the entire product stack yet. The GeForce RTX 40 series will coexist along the older GeForce RTX 30 series. In any case, anyone looking for the latest and greatest will be looking to pay in excess of RM4,150. The highlight of course is the GeForce RTX 4090, which will start from RM8,400, according to NVIDIA.
NVIDIA GeForce RTX 4090
The GeForce RTX 4090 packs 16,384 CUDA cores running at up to 2.52GHz, mated to 24GB GDDR6X over a wide 384-bit memory bus. NVIDIA claims a TGP of 450W, which will require a trio of 8-pin PCIe power connectors, or a PCIe Gen 5 cable. Performance is claimed to be in excess of 4X the GeForce RTX 3090 Ti, which is a massive upgrade, especially when you consider that the last-gen card was also rated at 450W.
Essentially, what we are looking at with the GeForce RTX 40 series is a doubling of performance at the same power draw, which is pretty great. I am not sure how well will that translate to games, although if you look at the graph above, the gains in excess of 4X in the latest games does look very promising. However, I also really like that NVIDIA kept the older games around to illustrate more modest gains of around 2X. Imagine being so confident in your product that you can be honest with the marketing for once.
NVIDIA GeForce RTX 4080
Meanwhile the less power-hungry and more reasonably-priced GeForce RTX 4080 will come with 16GB of GDDR6X memory. The GeForce RTX 4080 16GB coming in from RM6,300, delivers up 2X the performance of the GeForce RTX 3080 Ti, while drawing 10% less power. The lesser 12GB version that starts from RM4,730 has been renamed as the GeForce RTX 4070 Ti, will still allegedly beat the GeForce RTX 3090 Ti by a comfortable margin.
NVIDIA GeForce RTX 4070 Ti
The contentious GeForce RTX 4080 12GB GeForce RTX 4070 Ti is now finally official, with a price tag of $799. That’s $100 off the original MSRP for the card, but whether it’s good value or not is up to you. In any case, you will be getting a nice improvement over the GeForce RTX 3080 12GB’s performance, at lower power levels and also at the same MSRP. If you are coming from an older card, this might be a pretty good idea. However worth noting is that the MSRP will probably be hard to come by, as NVIDIA won’t be making a Founders Edition card. Not like it really matters to us Malaysians, since we never get the FE cards anyway. Malaysian MSRP is set to start from RM4,150, according to NVIDIA’s landing page.
Here’s a breakdown of the specs:
NVIDIA GeForce RTX 40 Series Specs and Price in Malaysia
| NVIDIA GeForce RTX 4070 Ti | NVIDIA GeForce RTX 4080 | NVIDIA GeForce RTX 4090 | |
| GPU | AD104, TSMC 4N 7680 CUDA cores 60 RT cores 240 Tensor cores | AD103, TSMC 4N 9728 CUDA cores 76 RT cores 304 Tensor cores | AD102, TSMC 4N 16,384 CUDA cores 128 RT cores 512 Tensor cores |
| Boost Clocks | 2.61GHz | 2.51GHz | 2.52GHz |
| Performance | 40 Shader-TFLOPs 82 RT-TFLOPs 641 Tensor-TFLOPs | 49 Shader-TFLOPs 113 RT-TFLOPs 780 Tensor-TFLOPs | 83 Shader-TFLOPs 191 RT-TFLOPs 1321 Tensor-TFLOPs |
| Memory | 12GB GDDR6X (21Gbps) 192-bit memory bus | 16GB GDDR6X (21Gbps) 256-bit memory bus | 24 GDDR6X (21Gbps) 384-bit memory bus |
| Interface | PCIe 4.0 x16 | PCIe 4.0 x16 | PCIe 4.0 x16 |
| TGP rating | 285W | 320W | 450W |
| Price | From RM4,150 | From RM6,300 | From RM8,400 |
| Malaysian AIB Prices | TBC | TBC | Click here |
| Availability | 5th January 2023 | November 2022 | 12th October 2022 |
Of course, these starting prices would be nigh impossible to get, if the GeForce RTX 30 series is anything to go by. Also, we don’t get Founders Edition cards in Malaysia, so we won’t get that sexy thicc new Founders Edition cards. For a more realistic look at the GeForce RTX 40 series cards in Malaysia, check out GIGABYTE’s GeForce RTX 40 series designs, or head on over to Lowyat.net to see what most of NVIDIA’s partners have announced.
[UPDATE 15/10/22: NVIDIA “unlaunch” the GeForce RTX 4080 12GB]
NVIDIA has unlaunched the GeForce RTX 4080 12GB, and will leave the GeForce RTX 4080 (16GB) as the only GeForce RTX 4080 to avoid “confusion”. The GeForce RTX 4080 (16GB) will go on to launch this 16th November.
[UPDATE 4/1/23: NVIDIA launches the GeForce RTX 4070 Ti]
The “unlaunched” GeForce RTX 4080 12GB is now officially named the GeForce RTX 4070 Ti, with prices set to start from $799.