By Brady Betzel

Nvidia has been on fire over the past year, not just with standard generational upgrades to the RTX line of GPUs, but by steadily advancing AI and multimedia processing, both of which are vitally important to multimedia creators. For everyone from video editors to computer scientists, Nvidia is handing over powerful GPU-based machine learning and language learning model processing to the masses for under $1,000.

For this review, I tested the Nvidia RTX 4070 Super Founder’s Edition, which retails for $599. The Founder’s Edition series of Nvidia GPUs refers to GPUs that are designed, built and sold by Nvidia directly or through a third party. By contrast, third-party vendors, like Asus, for example, have released their own versions. One example of that is the Nvidia RTX 4070 Ti, which I have previously reviewed.

The Nvidia RTX 4070 Super Founder’s Edition shows off a 20% increase in CUDA cores when compared to the RTX 4070 Founder’s Edition — increasing CUDA cores (single precision) from 5,888 to 7,168. In addition to increased CUDA core count, there are other general upgrades compared to the 4070 FE:

Yikes! That’s a lot of ones and zeros. But overall, you can see that there is a good improvement in specs while maintaining about the same power draw. The Nvidia RTX 4070 Super Founder’s Edition increases total graphics power by only 20W — a pretty amazing accomplishment considering the Super version retails for the same $599 price tag as the original RTX 4070 Founder’s Edition did. But what do those numbers really mean to a multimedia content creator? Actually, a lot.

With the fast and furious advancement in AI plugins and tools, multimedia applications, like Adobe’s Premiere Pro and After Effects and Blackmagic’s DaVinci Resolve, are seeing 1.5x and 2x increases in processing of AI-based effects like Magic Mask and AV1 encoding speeds. And for the future of large language models (LLMs), local processing is steadily increasing. Want to make an “AI” that will generate thousands of blog posts based on your own writing style and grammar from 20 years of product reviews? Start a local LLM using Hugging Face to grow your own local machine learning farm. (Actually, maybe I should do that. But for now, just know that a human is writing this review!) These advancements continue to let me know that I really have a lot to learn and figure out.

I have run the Nvidia RTX 4070 Super FE through some of my standard testing in Premiere and Resolve for speed comparisons.

Blackmagic Resolve
Up first is Resolve 18.6.4. If you are comparing my testing to other cards, keep in mind that advancements in underlying software efficiency, drivers, firmware updates, etc. will also improve, so increased speeds are not always tied to new hardware. That said, based on a UHD, 3840×2160 timeline, I edited the following clips together:

• ARRI RAW: 3840×2160 24fps – 7 seconds, 12 frames
• ARRI RAW: 4448×1856 24fps – 7 seconds, 12 frames
• BMD RAW: 6144×3456 24fps – 15 seconds
• Red RAW: 6144×3072 23.976fps – 7 seconds, 12 frames
• Red RAW: 6144×3160 23.976fps – 7 seconds, 12 frames
• Sony a7siii: 3840×2160 23.976fps – 15 seconds

I then added Blackmagic’s noise reduction, sharpening and grain. Finally, I replaced the noise reduction with Neat Video’s noise reduction. From there I exported multiple versions: a DNxHR 444 10-bit OP1a MXF file, DNxHR 444 10-bit MOV, H.264 MP4, H.265 MP4, AV1 MP4 (Nvidia GPUs only), and then an IMF package using the default settings:

These results are better than I was hoping for. When I look back at my RTX 4070 Founder’s Edition Resolve 18.1.2 results, I see that the 4070 Super FE speeds are 5 to 20 seconds faster — around 10% to 15%. Not as fast as the RTX 4090, naturally, but not far off either. You can find my RTX 4070 results here.

Adobe Premiere
I then ran similar tests inside Adobe Premiere Pro 2024 (24.1), exporting using Adobe Media Encoder. The video assets are the same as the ones I used in Resolve, but I used Adobe’s noise reduction, sharpening, and grain filters instead.

Previously, when testing the RTX 4070 Ti, I also used Nvidia’s provided Resolve testing projects that run through some of the AI plugins, like Magic Mask. The UHD Magic Mask processing is essentially real-time processing. It is a really efficient way to create a garbage mask around a person or object without spending a lot of time. While it’s not perfect and not ready for prime-time rotoscoping, it is a great AI tool to have in your toolbelt. Throughout my editing career, I remember more of what editors say they won’t do instead of what they will do. Magic Mask helps me solve a lot more problems quickly and raise my hand when another editor doesn’t know how to rotoscope.

Additional Tests
I brought back more tests that Nvidia provides to help me see where the improvements in AI plugins are headed:

Wedding Styles (:50): Vimeo 2160p – :38

• OFX: Light Rays, Gaussian Blur w/Masks, OFX: Glow, OFX: Styles, OFX: Sketch

Bride Face Refinement (:24): Vimeo 2160p – :34 / ~18fps

• Face Refinement, Beauty, OFX: Glow, OFX: Gaussian Blur

Optical Flow – Enhanced Better 50% Retime (1:07): Vimeo 2160p – :53 / ~32fps

• 50% Speed, Optical Flow – Enhanced Better

SuperScale2x 4k Source (:17): Vimeo 2160p – :12 / 40 fps

• 4k ProRes 2x Super Scale

SuperScale4x HD Source (1:16): Vimeo 2160p – 01:06 / 28fps

• HD H.264 4x Super Scale

SpeedWarp (:30) – MOV/H.264 – :12 /max 223fps

• 50% – Optical Flow – Enhanced Better

Scene Edit Detect (:40) ~ :06

For comparison, here are the RTX 4070 Ti testing times:

Wedding Styles (:50): Vimeo 2160p – :37

Bride_Face Refinement (:24): Vimeo 2160p – :33 / ~18fps

Optical Flow – Enhanced Better 50% Retime (:08): Vimeo 2160p – :36 / ~6fps

SuperScale2x 4k Source (:17): Vimeo 2160p – :18 / 26.5fps

SuperScale4x HD Source (1:16): Vimeo 2160p – 1:35 / 19.5fps

SpeedWarpExample (:30) – MOV/H.264 – :19 / ~112 – 125fps

Scene Edit Detect (:40) ~ :05

H.265 vs AV1 Export Testing:

Benchmarks
Blackmagic Proxy Generator:

Proxies: H.265/10-bit 1080p

Sony a7iii UHD (3840×2160) .mp4

• 46 files
• 160fps

Red RAW – Various Resolutions

• 2 files
• 28fps

Blender Benchmark CPU samples per minute:

1. Monster: 173.900929
2. Junkshop: 125.927716
3. Classroom: 90.111217

Blender Benchmark GPU samples per minute:

1. Monster: 3276.977971
2. Junkshop: 1534.755221
3. Classroom: 1632.598232

Neat Video HD: GPU-only, 71.6 frames/sec

Neat Video UHD: GPU-only, 16.8 frames/sec

OctaneBench 2020.1.5: Total Score – 723.62

PugetBench for Premiere Pro 1.0.0, Premiere Pro 24.1.0:

• Standard Overall Score: 11,050
• Extended Overall Score: 7,402
• LongGOP Score (Standard): 96
• Intraframe Score (Standard): 89
• RAW Score (Standard): 239
• GPU Effects Score (Standard): 73
• LongGOP Score (Extended): 66
• Intraframe Score (Extended): 76
• RAW Score (Extended): 88
• GPU Effects Score (Extended): 68

PugetBench for DaVinci resolve 0.93.1, DaVinci Resolve 18.6.4

• Standard Overall Score: 1910
• 4k Media Score: 133
• GPU Effects Score: 112
• Fusion Score: 328

Stable Diffusion Using the Automatic1111 WebUI:

Test Configuration:

Stable Diffusion CheckPoint: v1-5-pruned-emaonly.safetensors

Sampling Method: Euler_a

Sampling Steps: 50

Batch Count: 10

Batch Size: 1

Image Width: 768

Image Height: 768

Prompt: Beautiful render of a Tudor-style house near the water at sunset, fantasy forest. photorealistic, cinematic composition, cinematic high detail, ultra-realistic, cinematic lighting, depth of field, hyperdetailed, beautifully color-coded, 8k, many details, chiaroscuro lighting, ++dreamlike, vignette.

Total: 12.6 images per minute

VRay – CPU: 19881 vsamples

VRay – GPU RTX: 2217 vrays

VRay – GPU CUDA: 2936 vpaths

The Nvidia RTX 4070 Super FE is a valuable generational upgrade for Nvidia users. If you’re still hanging on to an RTX 3090, the 4070 Super is a great upgrade. If you’re on the 4070 Founder’s Edition, it’s probably not worth upgrading unless you have $599 plus tax burning a hole in your pocket.

Summing Up
Modern GPUs are starting to blaze trails through media creation, AI generation and — more specific to video and color correction — AI-based plugins. In my 40-year-old brain, I see a future with lots of AI tools aiding artists in their quests. Will it replace us? This is a common question that I don’t have the answer to. But last year and the beginning of 2024 have been extra-tough in the entertainment industry. Fingers crossed that it picks up soon, or we risk losing human-based originality in arts and entertainment.

The speed at which Nvidia is empowering AI plugins and toolsets is incredible, and when work picks up, having an Nvidia 4070 Super in your computer system will be a huge advantage.

Brady Betzel is an Emmy-nominated online editor at Margarita Mix in Hollywood, working on shows like Life Below Zero and Uninterrupted: The Shop . He is also a member of the Producers Guild of America. You can email Brady at bradybetzel@gmail.com. Follow him on Twitter @allbetzroff.