NVIDIA RTX 4070 Super
Honestly, I’m quite surprised that NVIDIA decided to price the NVIDIA RTX 4070 Super at USD 599. Compared to the now EOL RTX 4070 Ti, the Super easily becomes the better buy and at the same time, makes the RTX 4080 seem a little better off. That being said, even the RTX 4080 is now EOL, making the RTX 4080 Super the only 80-class card. So for folks in India, the RTX 4070 Ti is a great buy if you’re looking at 1440p gaming and even some 4K gaming if you can tweak the right settings. As always, what remains to be seen is if the retail cards in India follow a similar pricing. And we all know how that pans out.
We are talking about the NVIDIA RTX 4070 Super so hold on to your pitchforks until you finish reading this review. Graphics cards have been used for GPGPU software development for a while and now there’s a lot of AI development happening at the end-user as well. So we are looking towards making a couple of additions to the way we review graphics cards. In this review, we’ll showcase its capabilities in gaming, a bit of content creation and the usual analysis around power consumption.
NVIDIA positions the RTX 4070 Super right between the souped up Ti variant on one side and the base variant on the other with the super basically looking like a slightly younger 4070 ti. Everything from the perspective of the specifications looks really close to the 4070 Ti but it is a lot cheaper. So you can see where this review is leading. Nevertheless, let’s see how it actually performs.
The RTX 4070 SUPER uses an AD104 GPU but it’s not the absolute top-end variant. The RTX 4070 Ti uses a more complete AD104 and has more functional parts available. The NVIDIA RTX 4070 Super is positioned right between the 4070 Ti and the 4070 with 56 Streaming Multiprocessors and 7168 CUDA Cores. Its Shader performance of 36 FLOPs is a significant leap over the RTX 4070, indicating robust shading capabilities crucial for rendering complex scenes. The 224 Tensor Cores with 568 AI TOPS showcase its AI processing prowess, beneficial for AI-enhanced features in games and other AI-driven tasks. This is what folks interested in AI development should be most concerned with, aside from the memory on the board. Moreover, with 56 RT Cores, it offers advanced ray tracing capabilities.
In comparison, the RTX 4080, with 76 Streaming Multiprocessors and 9728 CUDA Cores, overshadows the 4070 SUPER in raw processing power, which is reflected in its higher Shader performance of 49 FLOPs. Its higher Tensor and RT Core count will likely translate to more efficient AI processing and more realistic ray-traced lighting effects. However, this comes at the cost of higher power consumption at 320 W, compared to the 220 W of the 4070 SUPER.
The RTX 4070 SUPER’s boost clock of 2475 MHz is competitive, ensuring that it can handle peak performance demands during intense gaming sessions. That’s 495 MHz of headroom and is very similar to what the RTX 4070 has to offer. And like all AD104 GPUs, it features a memory configuration with a 192-bit interface handling 12 GB of GDDR6X memory. This is good enough to fit some small LLMs so this is a good card for folks starting out on AI development courses. As for gaming, there is a nice balance between bandwidth and volume, catering well to modern games that demand both across varying resolutions.
| NVIDIA RTX 4080 | NVIDIA RTX 4070 Super | NVIDIA RTX 4070 Ti | NVIDIA RTX 4070 | NVIDIA RTX 3080 | NVIDIA RTX 3070 Ti | |
| Code Name | AD103-300 | AD104-350 | AD104-400 | AD104-250 | GA102-200 | GA104-400 |
| Graphics Processing Cluster | 7 | 5 | 5 | 5 | 7 | 6 |
| Texture Processing Cluster | 38 | 28 | 30 | 23 | 34 | 24 |
| Streaming Multiprocessors | 76 | 56 | 60 | 46 | 68 | 48 |
| CUDA Cores | 9728 | 7168 | 7680 | 5888 | 8704 | 6144 |
| Shader FLOPS | 49 | 36 | 40 | 29 | 30 | 22 |
| Tensor Cores | 304 | 224 | 240 | 184 | 272 | 192 |
| AI TOPS | 780 | 568 | 706 | 466 | 238 | 174 |
| RT Cores | 76 | 56 | 60 | 46 | 68 | 48 |
| RT FLOPS | 113 | 82.1 | 93 | 67 | 58 | 43 |
| Texture Units | 304 | 224 | 240 | 184 | 272 | 192 |
| ROP Units | 112 | 80 | 80 | 64 | 96 | 96 |
| Base Clock | 2205 MHz | 1980 MHz | 2310 MHz | 1920 MHz | 1440 MHz | 1575 MHz |
| Boost Clock | 2505 MHz | 2475 MHz | 2610 MHz | 2475 MHz | 1710 MHz | 1770 MHz |
| Memory Clock | 1400 MHz | 1313 MHz | 1313 MHz | 1313 MHz | 1188 MHz | 1188 MHz |
| Memory Data Rate | 22.4 Gbps | 21 Gbps | 21 Gbps | 21 Gbps | 19 Gbps | 19 Gbps |
| L1 Data Size per SM | 128 KB | 128 KB | 128 KB | 128 KB | 128 KB | 128 KB |
| L2 Cache Size | 64 MB | 48 MB | 48 MB | 36 MB | 5 MB | 4 MB |
| Total Video Memory | 16 GB | 12 GB | 12 GB | 12 GB | 10 GB | 8 GB |
| Video Memory Type | GDDR6X | GDDR6X | GDDR6X | GDDR6X | GDDR6X | GDDR6X |
| Memory Interface | 256-bit | 192-bit | 192-bit | 192-bit | 320-bit | GDDR6X |
| Total Memory Bandwidth | 716.8 GB/s | 504 GB/s | 504 GB/s | 504 GB/s | 760.3 GB/s | 608.3 GB/s |
| Process Node | TSMC 4N | TSMC 4N | TSMC 4N | TSMC 4N | Samsung 8nm | Samsung 8nm |
| Total Graphics Power | 320 W | 220 W | 285 W | 200 W | 320 W | 290 W |
In gaming workloads, these specifications suggest that the RTX 4070 SUPER is a potent card for high-fidelity gaming at high resolutions and frame rates. Its robust AI and ray tracing capabilities will enable it to deliver immersive experiences with advanced visual effects. Moreover, its relatively modest power consumption for its class implies a performance-per-watt proposition that could be attractive to gamers who would rather not see a hefty power bill at the end of the month.
With a core configuration very similar to the RTX 4070 Ti, this does bring into question the need for a 4070 Ti to exist. And a few days before this review, NVIDIA has officially marked the RTX 4070 Ti as End-of-Life (EOL). So you can only get RTX 4070 Ti cards as long as stocks last. Overall, the RTX 4070 SUPER is poised as a high-performance card that offers a sweet spot for gamers and creators who demand high performance without the extremes of the top-tier offerings.
We’ve been running CUDA related benchmarks ever since the first Fermi cards back in 2011 and we’ve seen how graphics cards have evolved from being used for just gaming to being used for gaming and software development and now for even AI. And it seems like with every new GPU microarchitecture, there’s a lot more AI use cases that are being streamlined and incorporated into video game engines. At the core of its AI capabilities are the Tensor Cores, which are pivotal for AI processing. These Tensor Cores are specifically designed to accelerate deep learning performance, which can be seen in features like DLSS (Deep Learning Super Sampling). DLSS leverages AI to upscale lower-resolution images in real-time, enabling higher frame rates without compromising image quality.
NVIDIA’s AI suite also extends to voice and image processing. NVIDIA Broadcast is a set of AI-powered tools that enhance live streams and video calls with virtual backgrounds, auto framing, and noise removal. Of late, even Windows has incorporated similar features under the collection titled Windows Studio Effects. However, these leverage any NPU that’s present and are not vendor-locked like how NVIDIA’s AI features tend to be.
For creators and professionals, the card’s AI features translate to faster content creation and simulation. AI-assisted editing software can utilise the GPU to expedite workflows, such as automatic image tagging, object recognition, and even advanced video effects that traditionally would take hours to render. Some of the plugins that are being developed for popular software such as Adobe Premiere Pro can save a lot of time that would have otherwise been spent on tedious tasks.
Having a lot of memory on the GPU allows for local AI models to be stored and that’s going to open up a lot more use cases in the near future. Aside from using AI to improve visual fidelity for upscaled or generated frames, we’re now seeing how AI is making its way into interactive elements within video games. AI-driven game development tools can use the cards like the RTX 4070 SUPER for tasks such as procedural generation, NPC behaviour, and realistic physics simulations, thus enhancing the gaming experience. At CES 2024, NVIDIA showcased how far NVIDIA ACE had come along. NVIDIA ACE is a stack of tools that allows NPCs (Non Player Characters) within video games to come up with dialogues on the fly while taking into consideration real-time changes within the game. And the faces of the NPCs along with the bodies will move and emote according to the dialogue being spoken. Exciting times are ahead!
The card is being compared against only the recently re-tested graphics cards and not any of the older graphics cards that are no longer available with us. Here’s the rig it was tested on.
TEST RIG
Processor – AMD Ryzen 9 7950X
CPU Cooler – Corsair H115i RGB PLATINUM
Motherboard – ASUS ROG CROSSHAIR X670E HERO
RAM – 2x 16 GB Kingston Renegade FURY 6000 MT/s (Set to 5200 MT/s)
SSD – WD Black SN850X 2 TB NVMe SSD
PSU – Cooler Master MWE 850 V2 Gold
We’re starting off with gaming performance in video games without RTX enabled. All of these FPS numbers are at 1440p resolution.
Switching on to RTX performance in synthetic benchmarks as well as video games we see a close similarity to the RTX 4070 Ti across most benchmarks with barely any difference in performance.
Peak temperatures of the RTX 4070 Super were mostly hovering around 63-64 degrees Celsius during gaming. Only the intensive synthetic benchmarks were able to take it up any further. Every card is hovering around 70 degrees Celsius under gaming load so there’s nothing of note to talk about here.
The NVIDIA GeForce RTX 4070 will be available for sale starting this week. Listing have already been up on some of the US online stores for a few days. In India, the card will be priced at INR 63,000 inclusive of all taxes. The official launch MSRP for the RTX 4070 Super is USD 599 which is the same as the RTX 4070 so you can see why the RTX 4070 Ti has no reason to exist considering how close the RTX 4070 Super is to the 4070 Ti.
Honestly, I’m quite surprised that NVIDIA decided to price the NVIDIA RTX 4070 Super at USD 599. Compared to the now EOL RTX 4070 Ti, the Super easily becomes the better buy and at the same time, makes the RTX 4080 seem a little better off. That being said, even the RTX 4080 is now EOL, making the RTX 4080 Super the only 80-class card. So for folks in India, the RTX 4070 Ti is a great buy if you’re looking at 1440p gaming and even some 4K gaming if you can tweak the right settings. As always, what remains to be seen is if the retail cards in India follow a similar pricing. And we all know how that pans out.
