Summary
- Nvidia has introduced the RTX 4080 Super and RTX 4070 Ti Super graphics cards to replace their predecessors, offering improved performance in areas such as image and video generation.
- The RTX 4080 Super is a more affordable option compared to the regular RTX 4080, with enhancements in raytracing, AI, and shaders, making it an excellent choice for gamers.
- The RTX 4070 Super features more cores and 16GB of vRAM, enabling better performance and the ability to load larger AI models, making it ideal for high-quality outputs.
During this year's CES announcements, Nvidia had some good news for consumers eyeing up new graphics cards. If you were particularly looking at picking up an RTX 4080 or an RTX 4070 Ti, you'll find that you can no longer pick up either aside from old stock; instead, they're being replaced by the RTX 4080 Super and the RTX 4070 Ti Super, respectively. There will also be an RTX 4070 Super, though the regular RTX 4070 will still be kept around.
Nvidia has shown off these cards in contrast to an upgrade path that it envisions for more most users, and so hasn't given us data to compare these cards to the ones that they're replacing. On paper, though, we can already visualize some improvements, and the RTX 4080 Super is likely to be one of the best GPUs for gamers.
Nvidia RTX 4080 Super: More cores and faster memory
The Nvidia RTX 4080 Super packs major upgrades in performance when compared to the RTX 3080 Ti and the RTX 2080 Super, which are two cards that Nvidia sees consumers upgrading from. It has 1.7x faster image generation when compared to the RTX 3080 Ti in Stable Diffusion, and 1.5x faster video generation in Stable Video Diffusion.
|
RTX 4080 Super |
RTX 4080 |
RTX 3080 Ti |
RTX 2080 Super |
|
|
Architecture |
Ada Lovelace |
Ada Lovelace |
Ampere |
Turing |
|
Shader Cores |
10,240 CUDA Cores |
9,728 CUDA Cores |
10,240 CUDA Cores |
3,072 CUDA Cores |
|
Ray Tracing Cores |
80 RT Cores (3rd Gen) |
76 RT Cores (3rd Gen) |
80 RT cores (2nd Gen) |
48 RT Cores (1st Gen) |
|
Tensor Cores |
320 Tensor Cores (4th Gen) |
304 Tensor Cores (4th Gen) |
320 Tensor Cores (3rd Gen) |
384 Tensor Cores (2nd Gen) |
|
Theoretical Performance |
52 Shader TFLOPS, 121 RT TFLOPS, 836 AI TOPS |
49 Shader TFLOPS, 113 RT TFLOPS, 780 AI TOPS |
34 Shader TFLOPS, 320 AI TOPS |
11 Shader TFLOPS, 34 RT TFLOPS, 89 AI TOPS |
|
DLSS version |
3 |
3 |
2 |
2 |
|
NV Encoders |
AV1 / H.264 |
AV1 / H.264 |
H.264 |
H.264 |
|
Framebuffer |
16GB G6X |
16GB G6X |
12GB G6X |
8GB G6 |
|
Memory Bandwidth |
736 GB/s (256-bit bus) |
716.8 GB/s (256-bit bus) |
912.4 GB/s (384-bit bus) |
496 GB/s |
|
Average Gaming Power |
246W |
251W |
352W |
227W |
|
Video Playback Power |
22W |
21W |
27W |
17W |
|
Idle Power |
15W |
13W |
13W |
10W |
|
TGP |
320W |
320W |
350W |
250W |
The Nvidia RTX 4080 Super costs $999, considerably cheaper than the $1200 of the regular RTX 4080. With slight improvements in raytracing, AI, and shaders too, the RTX 4080 Super is a better option than the regular 4080 for a decent chunk of change less. The RTX 4080 Super will be available to purchase from January 31st.
Nvidia RTX 4070 Ti Super: More cores and faster memory
The RTX 4070 Ti Super comes in at a price of $799, which is the same price as the regular 4070 Ti. However, this time around it has 16GB of vRAM instead of 12GB, along with more cores aimed at 1440p 144Hz gaming. More vRAM means better capabilities to hold AI models in memory too, so that consumers can do things like run LLMs and image generation on their own devices with larger and higher quality models.
|
RTX 4070 Ti Super |
RTX 4070 Ti |
RTX 3070 Ti |
RTX 2070 Super |
|
|
Architecture |
Ada Lovelace |
Ada Lovelace |
Ampere |
Turing |
|
Shader Cores |
8,448 CUDA Cores |
7,680 CUDA Cores |
6,144 CUDA Cores |
2560 CUDA Cores |
|
Ray Tracing Cores |
66 RT Cores (3rd Gen) |
60 RT Cores (3rd Gen) |
48 RT Cores (2nd Gen) |
40 RT Cores (1st Gen) |
|
Tensor cCores |
264 Tensor Cores (4th Gen) |
240 Tensor Cores (4th Gen) |
192 Tensor Cores (3rd Gen) |
320 Tensor Cores (2nd Gen) |
|
Theoretical Performance |
44 Shader TFLOPS, 102 RT TFLOPS, 706 AI TOPS |
44 Shader TFLOPS |
22 Shader TFLOPS, 43 RT TFLOPS, 174 AI TOPS |
9 Shader TFLOPS, 73 RT TFLOPS, 89 AI TOPS |
|
DLSS Version |
3 |
3 |
2 |
2 |
|
NV Encoders |
AV1 / H.264 |
AV1 / H.264 |
H.264 |
H.264 |
|
Framebuffer |
16GB G6X |
12GB G6X |
8GB G6X |
8GB G6 |
|
Memory Bandwidth |
672 GB/s (256-bit bus) |
504 GB/s (192-bit bus) |
608 GB/s (256-bit bus) |
448 GB/s (256-bit bus) |
|
Average Gaming Power |
226W |
226W |
240W |
205W |
|
Video Playback Power |
17W |
20W |
20W |
15W |
|
Idle Power |
12W |
12W |
12W |
11W |
|
TGP |
285W |
285W |
290W |
215W |
The RTX 4070 Ti Super will be available to purchase from January 24th, and will completely replace the regular RTX 4070 Ti. While you'll be able to purchase old stock from companies, eventually RTX 4070 Ti units will go out of stock and production will be replaced by the RTX 4070 Ti.
Nvidia RTX 4070 Super: More cores means better performance
The only card that isn't replacing an existing card in Nvidia's lineup is the RTX 4070 Super, which packs 20% more cores than the regular RTX 4070. With 16GB of vRAM as well consumers picking up one of these will find again benefit from being able to load larger AI models into memory for higher-quality outputs.
|
RTX 4070 Super |
RTX 4070 |
RTX 3070 |
RTX 2070 |
|
|
Architecture |
Ada Lovelace |
Ada Lovelace |
Ampere |
Turing |
|
Shader Cores |
7,168 CUDA Cores |
5,888 CUDA Cores |
5,888 CUDA Cores |
2,304 CUDA Cores |
|
Ray Tracing Cores |
56 RT Cores (3rd Gen) |
46 RT Cores (3rd Gen) |
46 RT Cores (2nd Gen) |
36 RT Cores (1st Gen) |
|
Tensor Cores |
224 Tensor Cores (4th Gen) |
184 Tensor Cores (4th Gen) |
184 Tensor Cores (3rd Gen) |
288 Tensor Cores (2nd Gen) |
|
Theoretical Performance |
36 Shader TFLOPS, 82 RT TFLOPS, 568 AI TOPS |
29 Shader TFLOPS |
20 Shader TFLOPS, 40 RT TFLOPS, 163 AI TOPS |
8 Shader TFLOPS, 24 RT TFLOPS, 63 AI TOPS |
|
DLSS Version |
3 |
3 |
2 |
2 |
|
NV Encoders |
AV1 / H.264 |
AV1 / H.264 |
H.264 |
H.264 |
|
Framebuffer |
12GB G6X |
12GB G6X |
8GB G6 |
8GB G6 |
|
Memory Bandwidth |
504 GB/s (192-bit bus) |
504 GB/s (192-bit bus) |
448 GB/s (256-bit bus) |
448 GB/s (256-bit bus) |
|
Average Gaming Power |
299W |
299W |
215W |
177W |
|
Video Playback Power |
16W |
16W |
20W |
18W |
|
Idle Power |
11W |
11W |
11W |
10W |
|
TGP |
220W |
200W |
220W |
185W |
The RTX 4070 Super costs $599 and will be available for purchase from January 17th, offering RTX 3090-level performance while consuming less power and being significantly cheaper. It's a great card for anyone looking to game at 1080p at high refresh rates or in 1440p.
Other AI improvements
New GPUs aren't all that Nvidia has in store to show off at this year's CES. The company has been making strides in AI, and it showed off the Avatar Cloud Engine in collaboration with Convai, a way to converse with NPCs in real-time using generative AI capabilities. If this sounds familiar, it's because it's an expansion on Kairos, a tech demo that Nvidia showed off at Computex last year.
In a new demo, Nvidia showed how you could speak with an NPC out loud and have it respond to you, complete with matching facial expressions based on the sentiment of the replies it gives. It uses Audio2Face and Riva Automatic Speech Recognition to understand what a user says, transform it into a prompt, and then generate a response that the NPC says to you. While still a demo, there are incredible use cases for this kind of technology in the future.