Group take a look at: NVIDIA GeForce RTX 40 Collection and Tremendous Collection GPUs

Which GeForce RTX 40 Collection GPU provides the very best bang to your buck for CG work? To seek out out, Jason Lewis places NVIDIA’s present shopper GPUs – together with the brand new Tremendous Collection playing cards – by way of an exhaustive set of real-world exams.

Our newest group take a look at is a type of ‘half two’ to final 12 months’s roundup of GeForce RTX 40 Collection GPUs, during which I pitted NVIDIA’s present Ada Technology shopper playing cards towards older playing cards from the Ampere and Turing generations in a sequence of real-world CG exams.

The unique evaluate featured the primary three playing cards from the GeForce RTX 40 Collection to be launched: the GeForce RTX 4090, GeForce RTX 4080 Ti and GeForce RTX 4080.

Since then, NVIDIA has launched its mid-cycle refresh, the ‘Tremendous Collection’ playing cards: the GeForce RTX 4080 Tremendous, GeForce RTX 4070 Ti Tremendous and GeForce RTX 4070 Tremendous.

On this evaluate, we will likely be placing the brand new Tremendous playing cards by way of the identical take a look at because the originals, utilizing current variations of the CG functions used for benchmarking.

Soar to a different a part of this evaluate
Technology focus: GPU architectures and APIs
Specifications
Testing procedure
Benchmark results
Other considerations
Verdict

Which GPUs are included within the group take a look at?
On this evaluate, we will likely be specializing in the excessive finish of NVIDIA’s GeForce RTX 40 Series: the unique GeForce RTX 4090, GeForce RTX 4080 and GeForce RTX 4070 Ti, and two of the brand new Tremendous Collection, the GeForce RTX 4080 Tremendous and GeForce RTX 4070 Tremendous.

Sadly, I wasn’t in a position to acquire the third Tremendous Collection card, the GeForce RTX 4070 Ti Tremendous, and as earlier than, we received’t be wanting on the mid-range and entry-level playing cards: the GeForce RTX 4070, GeForce RTX 4060 Ti and GeForce RTX 4060.

That is NVIDIA’s present technology of GPUs, primarily based on its Ada Lovelace architecture, and supposed to exchange the earlier GeForce RTX 30 Series, primarily based on the Ampere architecture, and GeForce RTX 20 Series, primarily based on the Turing architecture.

For comparability, we may even be testing two playing cards from the GeForce RTX 30 Collection, the GeForce RTX 3090 and GeForce RTX 3070, and one from the GeForce RTX 20 Collection, the GeForce RTX 2080 Ti.

We may even be testing three of NVIDIA’s workstation playing cards, the Ampere-generation RTX A6000, and the Turing-generation Titan RTX and Quadro RTX 8000.

Know-how focus: GPU architectures and APIs

Earlier than I get to the evaluate itself, here’s a fast recap of among the technical phrases that you’ll encounter in it. When you’re already acquainted with them, you could wish to skip forward.

Like NVIDIA’s previous-generation Ampere and Turing GPUs, the present Ada Lovelace GPU structure options three sorts of processor cores: CUDA cores, designed for rasterization and common GPU computing; Tensor cores, designed for machine studying operations; and RT cores, supposed to speed up ray tracing.

In an effort to reap the benefits of the RT cores, software program has to entry them by way of a graphics API: within the case of the functions featured on this evaluate, both DXR (DirectX Raytracing), utilized in Unreal Engine, or NVIDIA’s OptiX, utilized in most offline renderers.

In lots of renderers, the OptiX rendering backend is offered as a substitute for an older backend primarily based on NVIDIA’s CUDA API. The CUDA backends work with a wider vary of NVIDIA GPUs and software program functions, however OptiX allows hardware-accelerated ray tracing, and often improves efficiency.

Specs

First, let’s run by way of the specs of the brand new Tremendous Collection GPUs on take a look at. Yow will discover specs for the GeForce RTX 4090, GeForce RTX 4080 and GeForce RTX 4070 TI within the unique evaluate.

The GeForce RTX 4080 Super is principally only a common 4080, however with a totally unlocked AD103 processor, bumping the CUDA core depend to 10,240, up from the 9,728 cores of the unique 4080. The Tensor and RT core counts are bumped to 320 and 80, up from 304 and 76. Reminiscence bandwidth can also be barely elevated to 736.3 GB/s, up from 716.8 GB/s.

Within the Founders Version, the GeForce RTX 4080 Tremendous makes use of the identical three-slot cooler design of the unique 4080, the identical 16-pin 12VHPWR energy connector, and has the identical TDP of 320 W.

The GeForce RTX 4070 Super is an upgraded GeForce RTX 4070, utilizing the identical AD104 GPU, however with a way more vital bump in computing cores. The CUDA core depend rises to 7,168 CUDA cores, up from 5,188, whereas the Tensor and RT core counts rise to 224 and 56, up from 184 and 46. Reminiscence bandwidth is unchanged, at 504.2 GB/s.

Within the Founders Version, the GeForce RTX 4070 Tremendous makes use of a two-slot cooler design that’s nearly an identical to the vanilla GeForce RTX 4070, simply barely bigger in dimension, and makes use of the identical 16-pin 12VHPWR connector. Nevertheless, its TDP is larger: 220 W, up from 200 W for the unique.

One ultimate factor to notice in regards to the Tremendous Collection GPUs is the burden of the GeForce RTX 4080 Tremendous. At 4.68 lbs, it’s as heavy as the unique RTX 4080, and nearly as heavy because the RTX 4090. I might suggest utilizing a GPU brace to assist assist that weight, and stop the motherboard bending and cracking over time. GPU braces and helps could be discovered on-line for a couple of {dollars}: for my part, a small funding within the structural integrity of your GPU.

Click on the picture to view it full-size.

Testing process

For the take a look at machine, I’m nonetheless utilizing the reliable Xidax AMD Threadripper 3990X system that I reviewed in 2020. Though it’s now 4 years outdated, it’s nonetheless a particularly succesful system and doesn’t look like a bottleneck for any of the GPUs examined.

The present model of the take a look at system has the next specs:

CPU: AMD Threadripper 3990X
Motherboard: MSI Creator TRX40
RAM: 64 GB of three,600 MHz Corsair Dominator DDR4
Storage: 2TB Samsung 970 EVO Plus NVMe SD / 1 TB WD Black NVMe SSD / 4 TB HGST 7,200 rpm HD
PSU: 1300W Seasonic Platinum
OS: Home windows 11 Professional for Workstations

The one GPU not examined on the Threadripper system was the GeForce RTX 3070. I not have entry to a desktop RTX 3070, so testing was finished utilizing the cellular RTX 3070 within the Asus ProArt Studiobook 16 laptop computer from this current evaluate.

In that evaluate, I decided that throughout a spread of exams, the cellular RTX 3070 was round 10% slower than its desktop counterpart, so right here, I added 10% to the scores to approximate the efficiency of a desktop card. It isn’t a really perfect methodology, nevertheless it will get us to the suitable ballpark.

For testing, I used the next functions:

Viewport efficiency
3ds Max 2024, Blender 3.6, Chaos Vantage 2.1.1, D5 Render 2.3.4, Fusion 360, Maya 2024, Modo 16.0v2, Omniverse Create 2022.3.1, SolidWorks 2022, Substance 3D Painter 9.0.0, Unigine Neighborhood 2.16.0.1, Unity 2022.1, Unreal Engine 5.3.1 and 4.27.2

Rendering
Arnold for Maya 5.1.0, Blender 3.6 (Cycles renderer), KeyShot 11.2.0, LuxCoreRender 2.6, Maverick Studio 2022.5, OctaneRender 2022.1 Standalone, Redshift 3.5.24 for 3ds Max, SolidWorks Visualize 2022, V-Ray GPU 6 for 3ds Max Hotfix 3

Different benchmarks
Axiom 3.0.1 for Houdini 19.5, Cinema 4D v2023.1 (Pyro solver), Metashape 1.8.5, Premiere Professional 2022

Artificial benchmarks
3DMark Velocity Approach 1.0 and Port Royal 1.2, Cinebench 2024.0.1, CryEngine Neon Noir Ray Tracing Benchmark, OctaneBench 2020.1.5, Redshift Benchmark v3.5, Unity Enemies Demo, V-Ray Benchmark v5.02.01

All benchmarking was finished with NVIDIA Studio Drivers put in for the GeForce RTX GPUs and workstation drivers put in for the RTX A6000 and Quadro RTX 8000. Yow will discover a extra detailed dialogue of the drivers used later in the article.

Within the viewport and modifying benchmarks, the body price scores signify the figures attained when manipulating the 3D belongings proven, averaged over 5 testing classes to remove inconsistencies. In the entire rendering benchmarks, the CPU was disabled so solely the GPU was used for computing.

Testing was finished on a correct productiveness monitor setup, consisting of a pair of 27″ 4K screens working at 3,840 x 2,160px and a 34” widescreen show working at 3,440 x 1,440px. All three shows had a refresh price of 144Hz. When testing viewport efficiency, the software program viewport was constrained to the first show (one of many 27″ screens): no spanning throughout a number of shows was permitted.

Benchmark outcomes

Viewport efficiency
The viewport benchmarks embody various key DCC functions – general-purpose 3D software program like 3ds Max, Blender and Maya, extra specialist instruments like Substance 3D Painter, CAD packages like SolidWorks and Fusion 360, and real-time 3D functions like D5 Render, Unity and Unreal Engine.

Within the viewport benchmarks, the efficiency of the GeForce RTX 4080 Tremendous isn’t a lot totally different to its predecessor, the GeForce RTX 4080: solely barely forward in most exams, and even falling behind in a couple of. When you common the entire figures, it’s 3-5% sooner than the vanilla 4080.

Though I by no means had the possibility to check a vanilla GeForce RTX 4070, the GeForce RTX 4070 Tremendous appears to supply a way more vital efficiency increase, since in these exams, it’s basically the brand new GeForce RTX 4070 Ti. It pulls barely forward of the particular 4070 TI in some exams, and falls barely behind in others.

One other factor to notice is the Unreal Engine scores. Within the unique group take a look at, I famous that there was a wierd efficiency anomaly with the Metropolis and Valley of the Ancients scenes. This appears to have been a software program difficulty with Unreal Engine 5.1 itself, since shifting to Unreal Engine 5.3 has mounted the difficulty, with the GeForce RTX 40 Collection GPUs offering a a lot bigger uplift in efficiency over the the older playing cards.

Rendering
Subsequent, we have now a set of GPU rendering benchmarks, carried out with an assortment of the extra fashionable GPU renderers, rendering single frames at 4K or larger resolutions.

The rendering exams comply with an identical sample to the viewport benchmarks. The efficiency of the GeForce RTX 4080 Tremendous is sort of the identical because the GeForce RTX 4080, pulling barely forward in most exams, however falling behind in some – probably not a shock, contemplating that its CUDA, Tensor and RT core counts are solely barely larger, and that it has the identical 16 GB of VRAM.

The GeForce RTX 4070 Tremendous is actually the brand new GeForce RTX 4070 Ti, coming barely forward of the unique card in some exams, however falling barely behind in others.

Nevertheless, it’s price noting that for rendering, the GeForce RTX 4090, with its 24 GB body buffer, nonetheless reigns supreme right here by an enormous margin.

Different benchmarks
The following benchmarks take a look at using the GPU for extra specialist duties. Premiere Professional makes use of the GPU for video encoding; photogrammetry software Metashape makes use of the GPU for picture processing and 3D mannequin technology; and Houdini plugin Axiom and Cinema 4D’s Pyro solver each use the GPU for fluid simulation.

The miscellaneous exams comply with an identical sample to the viewport and rendering exams: the GeForce RTX 4080 Tremendous is only a smidge sooner than the GeForce RTX 4080, whereas the GeForce RTX 4070 Tremendous is actually the brand new GeForce RTX 4070 Ti. The variations in efficiency are even smaller than within the earlier two classes.

Neither of the brand new GPUs fares nicely with the Cinema 4D Pyro simulation, once more on account of reminiscence constraints. The one GPUs that basically carry out nicely are the 48 GB workstation playing cards.

Artificial benchmarks
Lastly, we have now an assortment of artificial benchmarks. They don’t precisely predict how a GPU will carry out in manufacturing, however they’re a good measure of its efficiency relative to different GPUs, and the scores could be in comparison with these out there on-line for different playing cards.

The artificial benchmarks affirm we already know in regards to the GeForce RTX 4080 Tremendous, rating it barely larger than the usual GeForce RTX 4080.

The curiosity is the GeForce RTX 4070 Tremendous: not like within the earlier exams, among the artificial benchmarks have it coming in considerably behind the GeForce RTX 4070 Ti. This is the reason I’m not an enormous fan of synthetics, as they’re usually tuned in a method that doesn’t mirror real-world utilization.

Different concerns

Energy consumption
To check the ability utilization of the GeForce RTX 40 Collection, I measured the ability consumption of your complete take a look at system on the wall outlet, utilizing a P3 Kill A Watt meter. Because the take a look at machine is a power-hungry Threadripper system, my figures will likely be larger than most DCC workstations.

For this group take a look at, I measured each energy and present drawn. Present (Amperage) is commonly neglected by reviewers, however it may be a vital determinant of what number of machines you’ll be able to run on a single circuit.

Most US homes run 15A circuits from the primary panel, and plenty of circuit breakers are rated for 80% of their most load, so a 15A circuit with a regular breaker shouldn’t exceed 12A for steady utilization. In my exams, the present drawn by the take a look at system approached 6A when the extra power-hungry GPUs have been put in. If the wall retailers in your house workplace are linked by a single circuit, this might decide whether or not you’ll be able to run two workstations concurrently, significantly if you consider screens and lights.

The unique GeForce RTX 40 Collection GPUs have been a reasonably good step up in energy effectivity over the GeForce RTX 30 Collection, and the brand new Tremendous Collection playing cards enhance effectivity even additional.

The GeForce RTX 4080 Tremendous exhibits a marked lower in energy utilization from the unique GeForce RTX 4080 in each take a look at other than Unreal Engine 5.

The GeForce RTX 4070 Tremendous additionally has decrease or equal energy utilization to the unique GeForce RTX 4070 Ti in nearly each take a look at: once more, the exception being Unreal Engine 5.

Drivers
Lastly, a observe on the Studio Drivers with which I benchmarked the GeForce RTX GPUs. NVIDIA now provides a choice of Studio or Game Ready Drivers for GeForce playing cards, recommending Studio Drivers for DCC work and Sport Prepared Drivers for gaming. In my exams, I discovered no discernible distinction between them by way of efficiency or show high quality. My understanding is that the Studio Drivers are designed for stability in DCC functions, and whereas I haven’t had any actual points when working DCC software program on Sport Prepared Drivers, if you’re utilizing your system primarily for content material creation, there isn’t any purpose to not use the Studio Drivers.

Click on the picture to view it full-size.

Verdict

Each the GeForce RTX 4080 Tremendous and GeForce RTX 4070 Tremendous are marked enhancements over their predecessors.

The GeForce RTX 4080 Tremendous is just barely sooner than the unique GeForce RTX 4080, nevertheless it additionally makes use of rather less energy, and its largest draw is its value: at $999, its MSRP is $200 decrease.

The GeForce RTX 4070 Tremendous shocked me by how nicely it carried out: in my exams, it matches or barely exceeds the unique GeForce RTX 4070 Ti, and whereas I didn’t take a look at a vanilla GeForce RTX 4070, I can solely assume that the variations could be much more vital. The 4070 Tremendous is smaller than the 4070 Ti and makes use of much less energy, and at $599, its launch value is $200 decrease.

My solely criticism is GPU reminiscence. Each of the Tremendous playing cards have the identical reminiscence capability as their predecessors, and – as these group exams have proven many occasions over – reminiscence capability is essential for DCC work. I might have favored to see the GeForce RTX 4080 Tremendous getting a bump up from 16 GB to twenty GB, even when it meant preserving the unique MSRP of $1,199. I notice that gaming continues to be the precedence marketplace for these GPUs, however alternative at all times advantages the patron.

Total conclusion
Each the GeForce RTX 4080 Tremendous and GeForce RTX 4070 Tremendous are nice GPUs for content material creation. They carry out extraordinarily nicely in most of my exams, solely struggling in people who require a number of VRAM.

However regardless of this, in case you have the price range for it, my suggestion for heavy DCC work would nonetheless be the unique GeForce RTX 4090. Its super-fast AD102 GPU – and, extra importantly, its 24GB of GPU reminiscence – will deal with absolutely anything you’ll be able to throw at it.

Lastly, I wish to thanks for taking the time to cease by. I hope this evaluate has been useful, and in case you have any questions or recommendations, let me know on the e-mail handle beneath.

Hyperlinks

Read more about the GeForce RTX 40 Series GPUs on NVIDIA’s website

Concerning the reviewer

Jason Lewis is a Senior Laborious Floor Artist at Lightspeed LA, a Tencent America growth group, and CG Channel’s common {hardware} reviewer. You possibly can see extra of his work in his ArtStation gallery.

Contact Jason at jason [at] cgchannel [dot] com.

Acknowledgements

Stephenie Ngo of NVIDIA
Chloe Larby of Grithaus Agency
Stephen G Wells
Adam Hernandez

Have your say on this story by following CG Channel on Facebook, Instagram and X (formerly Twitter). In addition to having the ability to touch upon tales, followers of our social media accounts can see movies we don’t submit on the positioning itself, together with making-ofs for the newest VFX films, animations, video games cinematics and movement graphics tasks.