Introduction

PowerColor's Radeon RX 9070 Hellhound is the brand's custom-design take on AMD's latest performance-segment graphics card. Launching at $550, the RX 9070 is positioned as a value-focused alternative to the higher-tier RX 9070 XT. However, with only a narrow $50 price gap between the two, premium custom designs like the Hellhound must justify their place in the market. Designed for high-performance 1440p gaming, including ray tracing, the RX 9070 debuts alongside NVIDIA's GeForce RTX 5070 at the same $550 MSRP.



The Radeon RX 9070 Series is built on AMD's cutting-edge RDNA 4 graphics architecture, delivering significant improvements in performance per compute unit (CU), ray tracing, and AI acceleration. These advancements make the RX 9070 XT and non-XT more competitive in modern gaming. Enhanced ray tracing performance reduces the performance cost of enabling RT effects, while improved AI acceleration brings machine learning closer to gaming applications. A prime example is FSR 4, AMD's latest ML-based upscaler, which offers superior image quality enhancements across all performance tiers compared to previous iterations.

At the heart of the RX 9070 is the 4 nm Navi 48 silicon, which features several process-level advancements over its predecessors. Unlike NVIDIA's Blackwell generation, which retains the same process node as Ada, AMD has upgraded Navi 48 to the TSMC N4P node, boosting both clock speeds and efficiency. Additionally, Navi 48 is a monolithic chip, eliminating the chiplet-based approach seen in Navi 31. This means that the GPU, memory controllers, and Infinity Cache are all built on a single 4 nm die, complemented by RDNA 4's power management and IPC optimizations.

The RX 9070 comes with 56 compute units (CUs), translating to 3,584 stream processors, 112 AI accelerators, 56 RT accelerators, and 224 TMUs. It also features 112 ROPs—an upgrade from Navi 32's 96 ROPs. The card is equipped with 16 GB of GDDR6 memory running at 20 Gbps across a 256-bit memory bus, providing 640 GB/s of bandwidth. What is disappointing, though, is that this is still older generation 20 Gbps GDDR6, which yields 640 GB/s of memory bandwidth. Our recent RTX 5070 testing has shown that memory size trumps bandwidth in ray tracing workloads, and AMD has given the RX 9070 a larger on-die cache than the 48 MB NVIDIA gave the RTX 5070, so things could get interesting.

The PowerColor Radeon RX 9070 Hellhound features a custom cooling solution designed to maintain optimal performance under heavy loads. Its triple-fan, dual-slot cooler utilizes a large aluminium fin-stack heatsink and precision-engineered heat pipes. Unlike some competing models which adopt the 12V-2x6 power connector, the Hellhound retains a more traditional dual 8-pin power input. PowerColor has tuned the RX 9070 Hellhound with a factory overclock, pushing its Game Clock beyond AMD's reference 2400 MHz specification. You also get a small lighting element and a dual BIOS feature with optional "quiet" BIOS. The PowerColor Radeon RX 9070 Hellhound is priced at USD 630, which is a $80 increase over the AMD MSRP of $550.

AMD RDNA 4 Architecture

The new Radeon RX 9070 XT and RX 9070 are powered by the 4 nm Navi 48 silicon. Unlike the Navi 31 and Navi 32 before it, this is a traditional monolithic die built entirely on the TSMC N4P (4 nm EUV) process node. It has an area of 356.5 mm², but with an impressive transistor count of 53.9 billion. To put this number into context, the previous "big Navi" GPU, the Navi 31 powering the RX 7900 XTX, has 57.8 billion transistors (just 8% more), despite having 96 CU, a larger Infinity Cache, and a wider 384-bit memory bus. Navi 31 has a die area of 529 mm² (304 mm² for the 5 nm GCD, with the rest being 6 nm MCDs), so you can see the kind of strides AMD made with die-area. At the top-level, the Navi 48 features a PCI-Express 5.0 x16 host interface and a 256-bit wide GDDR6 memory bus. With 20 Gbps GDDR6 memory chips, the RX 9070 XT gets 640 GB/s of memory bandwidth. This is made up for with a faster 64 MB Infinity Cache, and a critical improvement AMD made to its memory management system, called Out-of-Order Memory Queuing.

The Navi 48 silicon physically has 64 CU spread across 4 shader engines. With each CU containing 64 stream processors, the chip has a total of 4,096 stream processors. All 64 CU are enabled on the RX 9070 XT, while the RX 9070 has 56 out of 64 of them enabled. The maxed-out Navi 48 contains 4,096 stream processors, 128 AI accelerators, 64 RT accelerators, 256 TMUs, and 128 ROPs. The chip also features 64 MB of 3rd Gen Infinity Cache, and a 256-bit wide GDDR6 memory interface, driving 16 GB of 20 Gbps memory on both the RX 9070 XT and RX 9070. With 56 CU enabled on the RX 9070, it has 3,584 stream processors, 112 AI accelerators, 56 RT accelerators, 224 TMUs, and an untouched ROP count of 128, besides the exact same memory configuration as the RX 9070 XT.


At the heart of the RDNA 4 graphics architecture is the new dual compute unit, with a vastly improved memory sub-system, improvements made to the scalar units, a new technology called dynamic register allocation, and improvements to CU efficiency and engine clocks. Each CU has two scheduler blocks, driving a 192 KB general purpose register (GPR), an 8 KB scalar GPR, 32 FMA ALUs, and 32 FMA+INT ALUs. There are also 8 transcendental logic units. RDNA 4 introduces the concept of dual SIMD32 vector units, for even more parallelism. The Scalar Unit comes with support for newer Float32 ops. Schedulers are updated with accelerated spill/fill operations. Instruction prefetching is improved. The new generation AI Accelerator comes with two 16-bit and four 8-bit/4-bit dense matrix compute rates, support for 4:2 structured sparsity for doubling throughput, and matrix loads with transpose. AMD has incorporated many technologies from its CDNA 3 Radeon Instinct AI ML accelerators on the AI Accelerators of RDNA 4, including enhanced and power-optimized WMMA, improvements to the ops per CU, support for FP8, E4M3 and E5M2 formats, and 4:2 structured sparsity.


The new generation AI Accelerator comes with two 16-bit and four 8-bit/4-bit dense matrix compute rates, support for 4:2 structured sparsity for doubling throughput, and matrix loads with transpose. AMD has incorporated many technologies from its CDNA 3 Radeon Instinct AI ML accelerators on the AI Accelerators of RDNA 4, including enhanced and power-optimized WMMA, improvements to the ops per CU, support for FP8, E4M3 and E5M2 formats, and 4:2 structured sparsity.


The new generation Ray Accelerator comes with double the box and triangle intersection resources as RDNA 3 RT accelerator, support for hardware instance transforms, improvements to the RT stack management, BVH8 node compression, and a revolutionary feature called oriented bounding boxes. To contain the number of rays really needed to be tested against an object, modern ray tracing technologies use something called a bounding box, which defines a region in which a geometry has to be tested against rays. Most of the time, the geometry is of a vastly different shape and smaller than the shape of a bounding box, which introduces false intersections, and wastes ray testing resources. AMD innovated a way to turn this bounding box into a 3D shape by giving it a Z-axis component, so the bounding box is oriented closer to the shape of the object to be tested, reducing the number of rays needed to be tested against it.


This graph highlights the contribution of various components toward the 100% generational ray traversal performance gain, allowing AMD to make do with a CU count of 64, with RDNA 3 being the baseline.


Both ray tracing and ML acceleration are memory sensitive applications, so AMD innovated a revolutionary change to its memory management system with the introduction of new out-of-order memory. All math is executed in waves on an RDNA GPU, and mutual dependencies between waves can cause memory request stream misses, as one wave's memory request queue waits for the other wave to complete its task. This is solved with a new out-of-order (relaxed ordering) memory management. This graph highlights the contribution of various components toward the 100% generational ray traversal performance gain, allowing AMD to make do with a CU count of 64, with RDNA 3 being the baseline.


On AMD, a fairly big chunk of the ray tracing stack continues to be executed on shaders, but the company has made advances to ensure the cost of ray tracing on the shader resources of the GPU is minimal, with the introduction of Dynamic Registers to improve parallelism.


The new Radiance 2 Display Engine comes with major hardware updates that reduce GPU idle power draw in multi-monitor setups. The engine also comes with hardware flip-metering support (something NVIDIA also introduced with Blackwell, and which enables Multi-Frame Gen on the RTX 50-series). Flip-metering improves video frame pacing to the GPU and reduces CPU overhead for video playback. There is also a display engine level hardware image sharpening component that drives Radeon Image Sharpening. As for I/O, you get contemporary DisplayPort 2.1a and HDMI 2.1b, the maximum bitrate is UHBR 13.5.


Navi 48 comes with a dual media engine, each of the two can perform concurrent encoding and decoding, and so the implementation resembles an NVIDIA GPU having two each of NVENC and NVDEC units. The new generation media engine offers a 25% increase in H.264 low-latency encode quality, and an 11% improvement in HEVC encode quality. AV1 encode and decode get B-frames support, vastly improving bitrates. The media engine posts a 50% generational performance uplift (measured in encoder/decoder frame rates), with reductions in memory overhead.

AMD FSR 4 and New AMD Software Features

With the Radeon RX 9000 series, AMD is debuting FSR 4, the latest version of its performance enhancement suite. FSR 4 introduces the biggest tech update since FSR 2, which also makes it exclusive to the RX 9000 series. With the RDNA 4 generation of GPUs offering significantly higher ML compute throughput, AMD designed a new ML-based upscaler to reconstruct details of the upscaled frame. In that sense, FSR 4 is technologically close to DLSS. RDNA 4 GPUs also leverage AI DNNs to de-noise ray tracing and path tracing, unburdening the shaders.

An AI ML based upscaler requires ground truth data, and just as NVIDIA goes through the painstaking task of training game-specific ML models that can later be shipped to end-users through game developers and as driver updates; AMD has set up a large FSR 4 data-center of its own, powered by Radeon Instinct MI300-series AI GPUs for the task. On the client side, the ML-based upscaler needs the dedicated AI acceleration hardware from the RDNA 4 family. The frame generation technology appears to be carried over from FSR 3 Frame Generation, and is essentially a smart frame interpolation technology.


AMD in its presentation provided examples of image quality for FSR 4, claiming that in some cases, FSR 4 presents geometry more accurately than even native resolution. While AMD didn't detail it, FSR 4 comes with preparation for neural rendering. Given that Microsoft has now standardized neural shaders, and the ability for applications to directly address AI acceleration hardware on the GPU via the DirectX API, AMD could in the near future implement neural rendering, however, the company hasn't detailed a timeline. FSR 4 is ready for neural rendered objects.


Fluid Motion Frames gets an update with FMF 2.1, with improved image quality, reduced ghosting, and improved temporal tracking. The company also updated its API-agnostic image sharpening tool, with the introduction of Radeon Image Sharpening 2 (RIS 2).


AMD provides a slick driver frontend application called simply "AMD Software." AMD improved the way in which users can report bugs, artifacts, or corruption in games, through a voluntary (opt-in) feature called AMD Image Inspector.


As you submit a screenshot or video sequence to AMD, the company uses an AI model to identify rendering bugs, or corruption in the display. The entire driver-level game optimization, testing, and bug fixing process has been revamped with AI doing the heavy lifting.


The company also introduced AMD Chat, which is an AI chatbot that runs locally, and is accelerated by the GPU. You can make plain language queries on configuring AMD Software features and settings; and also chat with it on AMD technologies. This is essentially AMD's answer to NVIDIA's ChatRTX, but we don't know if the utility can be trained with custom datasets.


For streamers, AMD Software simplifies connecting their gameplay with their social media and streaming accounts. You can then take advantage of the new dual media engine of RDNA 4 that improves H.264 and HEVC image quality, and use other utilities such as AMD Noise Suppression, which is a 2-way AI-based audio filter.

AMD AI Apps Manager consolidates all your AI applications into a simple launcher-like interface resembling a game launcher.

Packaging

The Card

The PowerColor RX 9070 Series Hellhound design matches the style of the company's Radeon RX 7000 Series Hellhound cards. Black is the primary color, with gray and white highlights sprinkled on the card. The metal backplate that has a cutout for air to flow through.


Dimensions of the card are 33.0 x 13.0 cm, and it weighs 1076 g.


Installation requires two slots in your system. We measured the card's width to be 40 mm.


Display connectivity includes three standard DisplayPort 2.1a and a HDMI 2.1b.

With RDNA 4, AMD put effort to improve its standing with game streamers and creative professionals. It's done this by giving Navi 48 a dual VCN solution, so the GPU has two concurrent hardware accelerators for encoding and decoding. Perhaps the biggest changes at the silicon level is that AMD improved the encoding quality of its hardware H.264 and HEVC codecs. This was a niche complaint streamers had with AMD GPUs, and would avoid the brand altogether. The company also updated its AV1 hardware acceleration with support for B-frames, which are frames that lack image information, but math data that let the decoder reconstruct image data by comparing with the image data from adjacent I-frames containing it. This technique vastly improves streaming bitrates since half the frames lack image data.


PowerColor has installed a fixed blue lighting element behind the eye of the hound print on the back of the card, on the eye—very hard to notice. A second zone of the same color is located near the back end of the card.


The card uses two 8-pin PCIe power connectors, which are specified for up to 300 W, plus 75 W from the slot, the actual power limit of the card is much lower of course.


This BIOS switch lets you toggle between the default performance BIOS and a second "quiet" BIOS. Also note here, the LED switch, which lets you turn off the lighting easily, without any software installation needed.

Teardown

The PowerColor cooling solution uses four heatpipes. The heatsink provides cooling not only for the GPU, but also for the memory chips and VRM circuitry.


The backplate protects the PCB against damage during handling and installation.

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).