Introduction

Today NVIDIA releases the GeForce RTX 4070 Ada—we have the review. The new RTX 4070 (non Ti) is the most affordable graphics card in the RTX 40-series so far, which also makes it the most important one. At a starting price of $600, and with the promise of maxed out 1440p gaming, or 4K gaming with fairly high settings the card could bring fresh wind to the gaming segment. You also get the force multiplier that is DLSS 3, to achieve significantly higher framerates.

The GeForce RTX 4070 in this review has a lot in common with the recently launched RTX 4070 Ti, in that it's based on a cut-down version of the same silicon, and offers the same 12 GB of GDDR6X memory; but at a much lower wattage class. In fact, many custom-design RTX 4070 cards, including some factory-overclocked ones, make do with just one 8-pin PCIe power connector (a 225 W power configuration when you count in the PCIe slot). Of course, NVIDIA also lets board partners use the newer ATX 12VHPWR power connector that can provide a lot more power.



The GeForce "Ada Lovelace" graphics architecture that the RTX 4070 is based on, debuts the third generation of RTX, NVIDIA's ground-breaking technology that ups realism in games by fusing real-time ray traced elements with classic raster 3D graphics. Even this bit of ray tracing requires enormous compute power, and so the company created dedicated hardware inside the GPU that takes care of these workloads. Ada debuts the 3rd generation RT core with a generational uplift in ray tracing intersection performance; and 4th generation Tensor cores, which accelerate AI deep-learning neural nets, by tapping into even newer capabilities. The Ada CUDA core, plus higher GPU clock-speeds, and a completely redesigned memory sub-system with larger on-die caches, make up the new architecture. Putting it all together is the new TSMC 4N (5 nm with 4 nm-class characteristics) process.

The GeForce RTX 4070 that we test in this review is based on the same AD104 silicon that powers the RTX 4070 Ti, but while the latter maxes out all available hardware on the silicon, the former is heavily cut down. The RTX 4070 has just 46 out of 60 streaming multiprocessors (SM) physically present; which works out to an identical shader count to its predecessor, of 5,888 CUDA cores. It also features 184 Tensor cores, 46 RT cores, 184 TMUs, and 64 ROPs (out of 80 present). Thankfully, the memory itself is unchanged, you get 12 GB of 21 Gbps GDDR6X across the chip's 192-bit memory bus; yielding 504 GB/s bandwidth, which is generationally higher than the 448 GB/s of the RTX 3070.

While NVIDIA didn't release a Founders Edition for the GeForce RTX 4070 Ti, they've engineered one for the RTX 4070. The Founders Edition is no longer a pure reference design, but a slightly more premium one, that still comes at MSRP.

The RTX 4070 Founders Edition design is based on the same Dual Axial Flow-through philosophy as the RTX 3080 FE, or even the latest RTX 4090 FE, but in a much more compact form. While NVIDIA does allow its board partners to use legacy 8-pin power connectors, the Founders Edition relies on the 16-pin 12VHPWR connector, but features the same clocks and power limit as the other cards we're testing today.

Architecture

The Ada graphics architecture heralds the third generation of the NVIDIA RTX technology, an effort toward increasing the realism of game visuals by leveraging real-time ray tracing, without the enormous amount of compute power required to draw purely ray-traced 3D graphics. This is done by blending conventional raster graphics with ray traced elements such as reflections, lighting, and global illumination, to name a few. The 3rd generation of RTX introduces the new higher IPC "Ada" CUDA core, 3rd generation RT core, 4th generation Tensor core, and the new Optical Flow Processor, a component that plays a key role in generating new frames without involving the GPU's main graphics rendering pipeline.


The GeForce Ada graphics architecture driving the RTX 4070 leverages the TSMC 5 nm EUV foundry process to increase transistor counts. At the heart of this GPU is the new AD104 silicon, with a fairly high transistor count of 35.8 billion, which is more than double that of the previous-generation GA104. The GPU features a PCI-Express 4.0 x16 host interface and a 192-bit wide GDDR6X memory bus, which on the RTX 4070 wires out to 12 GB of memory. The Optical Flow Accelerator (OFA) is an independent top-level component. The chip features one NVENC and one NVDEC unit.

The essential component hierarchy is similar to past generations of NVIDIA GPUs. The AD104 silicon features 5 Graphics Processing Clusters (GPCs), each of these has all the SIMD and graphics rendering machinery, and is a small GPU in its own right. Each GPC shares a raster engine (geometry processing components) and two ROP partitions (each with eight ROP units). The GPC of the AD104 contains six Texture Processing Clusters (TPCs), the main number-crunching machinery. Each of these has two Streaming Multiprocessors (SM), and a Polymorph unit. Each SM contains 128 CUDA cores across four partitions. Half of these CUDA cores are pure-FP32; while the other half is capable of FP32 or INT32. The SM retains concurrent FP32+INT32 math processing capability. The SM also contains a 3rd generation RT core, four 4th generation Tensor cores, some cache memory, and four TMUs. There are 12 SM per GPC, so 1,536 CUDA cores, 48 Tensor cores, and 12 RT cores; per GPC. There are five such GPCs, which add up to 7,680 CUDA cores, 240 TMUs, 240 Tensor Cores, and 60 RT cores. Each GPC contributes 16 ROPs, so there are 80 ROPs on the silicon. The RTX 4070 is carved out of the AD104 by disabling an entire GPC worth 6 TPCs, and an additional TPC from one of the remaining GPCs. This yields 5,888 CUDA cores, 184 Tensor cores, 46 RT cores, and 184 TMUs. The ROP count has been reduced from 80 to 64. The on-die L2 cache sees a slight reduction, too, which is now down to 36 MB from the 48 MB present.


The 3rd generation RT core accelerates the most math-intensive aspects of real-time ray tracing, including BVH traversal. Displaced micro-mesh engine is a revolutionary feature introduced with the new 3rd generation RT core. Just as mesh shaders and tessellation have had a profound impact on improving performance with complex raster geometry, allowing game developers to significantly increase geometric complexity; DMMs is a method to reduce the complexity of the bounding-volume hierarchy (BVH) data-structure, which is used to determine where a ray hits geometry. Previously, the BVH had to capture even the smallest details to properly determine the intersection point. Ada's ray tracing architecture also receives a major performance uplift from Shader Execution Reordering (SER), a software-defined feature that requires awareness from game-engines, to help the GPU reorganize and optimize worker threads associated with ray tracing.


The BVH now needn't have data for every single triangle on an object, but can represent objects with complex geometry as a coarse mesh of base triangles, which greatly simplifies the BVH data structure. A simpler BVH means less memory consumed and helps to greatly reduce ray tracing CPU load, because the CPU only has to generate a smaller structure. With older "Ampere" and "Turing" RT cores, each triangle on an object had to be sampled at high overhead, so the RT core could precisely calculate ray intersection for each triangle. With Ada, the simpler BVH, plus the displacement maps can be sent to the RT core, which is now able to figure out the exact hit point on its own. NVIDIA has seen 11:1 to 28:1 compression in total triangle counts. This reduces BVH compile times by 7.6x to over 15x, in comparison to the older RT core; and reducing its storage footprint by anywhere between 6.5 to 20 times. DMMs could reduce disk- and memory bandwidth utilization, utilization of the PCIe bus, as well as reduce CPU utilization. NVIDIA worked with Simplygon and Adobe to add DMM support for their tool chains.


Opacity Micro Meshes (OMM) is a new feature introduced with Ada to improve rasterization performance, particularly with objects that have alpha (transparency data). Most low-priority objects in a 3D scene, such as leaves on a tree, are essentially rectangles with textures on the leaves where the transparency (alpha) creates the shape of the leaf. RT cores have a hard time intersecting rays with such objects, because they're not really in the shape that they appear (they're really just rectangles with textures that give you the illusion of shape). Previous-generation RT cores had to have multiple interactions with the rendering stage to figure out the shape of a transparent object, because they couldn't test for alpha by themselves.


This has been solved by using OMMs. Just as DMMs simplify geometry by creating meshes of micro-triangles; OMMs create meshes of rectangular textures that align with parts of the texture that aren't alpha, so the RT core has a better understanding of the geometry of the object, and can correctly calculate ray intersections. This has a significant performance impact on shading performance in non-RT applications, too. Practical applications of OMMs aren't just low-priority objects such as vegetation, but also smoke-sprites and localized fog. Traditionally there was a lot of overdraw for such effects, because they layered multiple textures on top of each other, that all had to be fully processed by the shaders. Now only the non-opaque pixels get executed—OMMs provide a 30 percent speedup with graphics buffer fill-rates, and a 10 percent impact on frame-rates.


DLSS 3 introduces a revolutionary new feature that promises a doubling in frame-rate at comparable quality, it's called AI frame-generation. While it has all the features of DLSS 2 and its AI super-resolution (scaling up a lower-resolution frame to native resolution with minimal quality loss); DLSS 3 can generate entire frames simply using AI, without involving the graphics rendering pipeline. Later in the article, we will show you DLSS 3 in action.


Every alternating frame with DLSS 3 is hence AI-generated, without being a replica of the previous rendered frame. This is possible only on the Ada graphics architecture, because of a hardware component called the optical flow accelerator (OFA), which assists in predicting what the next frame could look like, by creating what NVIDIA calls an optical flow-field. OFA ensures that the DLSS 3 algorithm isn't confused by static objects in a rapidly-changing 3D scene (such as a race sim). The process heavily relies on the performance uplift introduced by the FP8 math format of the 4th generation Tensor core. A third key ingredient of DLSS 3 is Reflex. By reducing the rendering queue to zero, Reflex plays a vital role in ensuring that frame-times with DLSS 3 are at an acceptable level, and a render-queue doesn't confuse the upscaler. A combination of OFA and the 4th Gen Tensor core is why the Ada architecture is required to use DLSS 3, and why it won't work on older architectures.

Packaging

The Card

NVIDIA's Founders Edition looks stunning—it could be an Apple product. Compared to the GeForce 30 Series, NVIDIA has made small improvements to the design language, making it an even cleaner design than before. As with Ampere, the card is designed for airflow to go through the card—that's why there's two fans. One sucks in cool air from the bottom, is pushed through the card and then blown out towards the case top on the other side.


While the RTX 4080 and RTX 4090 were huge cards, the RTX 4070 is rather compact.


From left to right: RTX 3060 Ti, RTX 3070, RTX 4070, RTX 3070 Ti, RTX 3080.


Dimensions of the card are 24 x 11 cm, and it weighs 1033 g.


Installation requires two slots in your system.


Display connectivity includes three standard DisplayPort 1.4a ports and one HDMI 2.1a (same as Ampere).

The new 8th Gen NVENC now accelerates AV1 encoding, besides HEVC. You also get an "optical flow accelerator" unit that is able to calculate intermediate frames for videos, to smooth playback. The same hardware unit is used for frame generation in DLSS 3.


The card uses the new 12+4 pin ATX 12VHPWR connector, which is rated for up to 600 W of power draw. An adapter cable from 2x PCIe 8-pin is included (which is rated for up to 300 W). The card's default power limit is 200 W. Of course the 3x and 4x 8-pin to 16-pin adapter cables from other Ampere cards will work with the RTX 4070, but the card won't need or use that much power.

Teardown

Disassembly is similar to earlier Founders Edition graphics cards. First pop off the top cover, it's attached magnetically—great idea.


Now remove several Torx screws.


With the back cover removed, we have to disconnect two flat-ribbon cables. Use your fingers where possible, instead of metal tools. Flip the connector latch up, as displayed in the second photo, then carefully pull out the cable. Remove the screws on the slot cover and you can remove the heatsink from the PCB.


NVIDIA has installed four heatpipes that move heat away from the GPU surface.


The cooler has copper base and provides cooling for the memory chips and VRM circuitry, too.

High-resolution PCB Pictures

These pictures are for the convenience of volt modders and people who would like to see all the finer details on the PCB. Feel free to link back to us and use these in your articles, videos or forum posts.


High-resolution versions are also available (front, back).

Circuit Board (PCB) Analysis

GPU voltage is a six-phase design, managed by a uPI uP9512R controller.


OnSemi NCP302150 DrMOS components are used for GPU voltage; they are rated for 50 A of current each.


Memory voltage is a two-phase design, managed by a uPI uP9529Q controller.


For memory, OnSemi NCP302150 DrMOS with a 50 A rating are used again.


The GDDR6X memory chips are made by Micron and carry the model number D8BZC, which decodes to MT61K512M32KPA-21:U. They are specified to run at 1313 MHz (21 Gbps GDDR6 effective).


NVIDIA's AD104 graphics processor is the company's third Ada Lovelace GPU. It is built using a 5 nanometer process at TSMC Taiwan, with a transistor count of 35.8 billion and a die size of 295 mm².