RDNA 3: Farewell, Monolith, and Hello, Chiplets!
The Radeon RX 7900 XTX and its predecessor, the Radeon RX 7900 XT, are not only the first entries in AMD’s new Radeon 7000 series of graphics cards, but also the first GPUs based on AMD’s RDNA 3 microarchitecture.
The use of “chiplets” distinguishes this new approach. Let us explain: Processors and graphics cards have traditionally been designed as single large chips, in what are known as “monolithic” designs. Chiplets, on the other hand, are multiple smaller chips packaged together as a single component. Monolithic chip designs have a performance advantage, but chiplets compensate for this with scalability and cost savings.
Chiplet designs have been around for decades, but their use has recently increased as they offer unique solutions to problems facing the tech industry today, particularly in manufacturing efficiency.
AMD builds graphics cards with two different types of chips in RDNA 3 GPUs. AMD refers to the larger of the two chips as a Graphics Compute Die (GCD). The GCD houses the majority of the functional hardware, such as shader cores, TMUs, ROPs, ray-tracing cores, display controllers, and media engines. The majority of the GPU’s cache, including the Infinity Cache and memory controllers, is stored separately in what AMD calls Memory Cache Dies (MCDs).
So far, the cards we’ve seen rely on a single GCD with multiple MCDs. The Radeon RX 7900 XTX has six of these MCDs, each with two 32-bit memory controllers and 16MB of Infinity Cache, for a total of 96MB of Infinity cache and a 384-bit memory interface. The MCDs are then linked to the GCD via ultra-fast interconnects capable of supporting up to 5.3TB/s of bandwidth.
AMD is able to use different fabrication processes for the various components by splitting these components up. The more important component, the GCD, is built on a 5nm TSMC process and measures 330mm2. The MCDs were built on a 6nm TSMC process and each measure 37mm2, totaling 222mm2 for all MCDs. The combined area of all MCDs and the GCD is 552mm2, with a total of 57.7 billion transistors.
As previously stated, monolithic chip designs have a performance advantage over chiplets, but this advantage is difficult to quantify and may be razor-thin. What is clear is that chiplets allow more resources to be crammed into a component while also helping to keep costs low. AMD cited this as a key reason for selling the Radeon RX 7900 XTX for less than $1,000.
An Examination of the Radeon RX 7900 XTX
The Radeon RX 7900 XTX’s GCD contains 96 Compute Units (CUs) each with 64 Streaming Processors and four TMUs, for a total of 6,144 Streaming Processors and 384 TMUs. AMD is also continuing to add one Ray Accelerator to each CU, giving the RX 7900 XTX 192 and 96 of these processing elements, respectively. AMD claims that the Ray Accelerators have been reworked to improve performance in select games by up to 1.8x over the previous generation.
Clock speeds have not increased significantly since the previous generation, with the Radeon RX 7900 XTX sitting only slightly higher (around 200MHz) than a stock AMD Radeon RX 6950 XT. The base clock of the Radeon RX 7900 XTX is 2,300MHz, and the boost clock is 2,500MHz. Given the relatively minor gains in clock speed, the performance gains we will see in benchmarks will come from architectural improvements and the more substantial increase in core count.
AMD also increased the number of ROPs significantly, from 128 on the RX 6950 XT to 192 on the Radeon RX 7900 XTX. The Radeon RX 7900 XTX also has significantly increased memory capacity and bandwidth, with 24GB of GDDR6X memory connected via the previously mentioned 384-bit memory interface.
There are a few more minor feature updates on the AMD Radeon RX 7900 XTX that you should be aware of. The card can support up to an 8K display at 165Hz or two 8K displays at 60Hz thanks to AMD’s Radiance Display Engine technology, which supports dual-link bandwidth of up to 54Gbps over DisplayPort 2.1.
Second, AMD updated its media engines to include AV1 encoding and decoding support, as well as up to two simultaneous 8K 60Hz encodes or decodes. Finally, in addition to the usual HDMI port and two DisplayPort connections, AMD has included a USB Type-C port on the Radeon RX 7900 XTX’s rear I/O panel. This could be useful as USB Type-C becomes more popular as a video connection.
The card we’re testing is based on AMD’s reference design for the Radeon RX 7900 XTX. The card measures 4.7 by 11.3 by 2.1 inches (HWD), making it both larger and smaller than Nvidia’s GeForce RTX 4080 or GeForce RTX 4090. The card includes three fans for cooling and two eight-pin PCI Express power connectors.
Competition and Test Setup
AMD’s Radeon RX 7900 XTX, as well as all other cards listed below, were tested on our updated 2022 GPU testbed, which includes an Intel Core i9-12900K processor running at stock clock speeds on an Asus ROG Maximus Z690 Hero motherboard. The processor is cooled by a Corsair Hydro Series H100X water cooler, and the system includes 32GB of 5,600MHz Corsair Vengeance RAM.
To ensure adequate power is available to all components, the testbed is powered by a Corsair HX1500i 1,500W 80 Plus Platinum power supply. All tests were run within Windows 11 Pro, which was installed on a 1TB Corsair MP600 Pro NVMe 4.0 SSD.
At the moment, Nvidia’s GeForce RTX 4080 is the main rival to the Radeon RX 7900 XTX. The cards differ in numerous ways, but both are new cutting-edge graphics cards that, as you’ll see below, frequently produce comparable performance numbers.
All of the cards in these charts have advantages and disadvantages, but it’s critical to keep price in mind when comparing them. The AMD Radeon RX 7900 XTX, in particular, has a price advantage over the $1,199 Nvidia RTX 4080, which helps to put the performance numbers into context.
It’s also interesting to compare the performance of the AMD Radeon RX 7900 XTX to that of other AMD cards, such as the AMD Radeon RX 7900 XT and the last-gen AMD Radeon RX 6800 XT. We won’t spend much time comparing these cards because the older AMD Radeon RX 6800 XT is simply outclassed by its successors. We will spend more time comparing the two Radeon RX 7900 cards in our upcoming review of the Radeon RX 7900 XT.
Benchmarks for Synthetics
Synthetic test results must always be interpreted with caution because they do not always accurately reflect real-world performance. Even if synthetic test results are best-case scenarios for each card in their respective test areas, driver performance and the inner workings of the card can result in significantly different game results.
Except in the Time Spy Extreme test, the Radeon RX 7900 XTX performs reasonably well in 3DMark and Furmark, but not well enough to catch up to its main competitor, the Nvidia GeForce RTX 4080. The situation is reversed in Unigine’s Superposition tests, with the RX 7900 XTX scoring higher in DirectX and OpenGL. Luxmark, which specifically tests GPGPU performance, places the RTX 4080 far ahead of the RX 7900 XTX.
Ray-Tracing, DLSS, and FSR Evaluation
AMD worked hard to improve ray-tracing performance on Radeon 7000-series cards, which is especially noticeable when compared to last-generation cards like the RX 6800 XT. In the AAA game F1 22, the Radeon RX 7900 XTX has a performance advantage of 74% to 100% over the Radeon RX 6800 XT. That is a significant improvement from generation to generation. It is worth noting, however, that the Radeon RX 7900 XTX is a more direct successor to the Radeon RX 6950 XT, which we did not test.
Though AMD has achieved significant generational performance gains, especially without raising prices, it’s worth noting that ray-tracing is a bit of a Radeon 7000-series Achilles’ heel. To say that cards like the Radeon RX 7900 XTX perform poorly when ray-tracing is used is an enormous overstatement and unjust, as the cards perform quite well in this regard. The issue for AMD here is that Nvidia’s new RTX 40-series cards simply outperform them.
The Nvidia GeForce RTX 4080 outperforms the Radeon RX 7900 XTX by 11% to 17% depending on the resolution in F1 22. When FSR and DLSS are enabled for their respective cards, the Radeon RX 7900 XTX gains a 7% to 10% performance advantage over the RTX 4080.
Things look even worse for the AMD Radeon RX 7900 XTX in Marvel’s Guardians of the Galaxy, but we have reason to believe that game or driver optimization, rather than the card, is to blame. We are unable to test the game with FSR on the Radeon RX 7900 XTX because the option is greyed out and the feature cannot be activated, implying that something isn’t quite right with this game. We anticipate that this will be resolved shortly after launch.
AAA Game Testing in the Modern Era
In games that don’t support ray-tracing, the Radeon RX 7900 XTX consistently outperforms the RTX 4080. In some games, there is effectively no difference in performance between the two cards, with any difference of 3% or less being considered significant. In none of these games, the RTX 4080 outperforms the Radeon RX 7900 XTX by more than that margin.
In a few select tests, the Radeon RX 7900 XTX manages performance leads ranging from 5% to 12%. The best of these is Far Cry 5 at 4K, where the Radeon RX 7900 XTX outperforms the competition by 12%.
Legacy game tests are always a bit of a mixed bag, with performance varying widely. We run these games to see if they will work on the latest cards rather than to see how fast they will run, though we are interested in both. This is another area where Nvidia has a slight advantage, as the RTX 4080 runs these games faster on average.
Thermals and power
We use a Kill-A-Watt monitor to measure the power consumption of our GPU test to get a rough idea of how power consumption varies from one card to the next. This is yet another area where Nvidia’s RTX 4080 outperforms, consuming significantly less power in our Furmark test and a moderate amount less in Guardians of the Galaxy.
This is another plus for Nvidia in terms of performance per watt, and temperatures may benefit as well. In terms of operating temperature, AMD’s Radeon RX 7900 XTX is cooler, but we hear its fans running loud and hard throughout all tests, whereas the GeForce RTX 4080 is noticeably quieter.
Nvidia designed its GeForce RTX 4080 Founders Edition cards with large physical dimensions and massive heatsinks to keep them cool without requiring the fans on AMD’s card to work as hard. However, this is another mixed bag, as the enormous size makes it difficult to fit into cases.
All that and a bag of Chiplets, in other words.
Taking everything into account, we can conclude that neither the AMD Radeon RX 7900 XTX nor the Nvidia GeForce RTX 4080 wins on performance alone. They are both highly competitive in terms of performance, and which will work best for you will most likely depend on the games you intend to play.
When pricing is taken into account, the AMD Radeon RX 7900 XTX becomes far more appealing. Why pay $200 more for the Nvidia GeForce RTX 4080 when you can get similar performance from the AMD Radeon RX 7900 XTX? You might as well get the card that gives you (a lot of) lunch money.
Nvidia’s biggest advantage is in games that use ray-tracing, but games without it are still far more common than those that do. Furthermore, the AMD Radeon RX 7900 XTX may still outperform in games with ray-tracing support and FSR, as we saw in F1 22.
Without a doubt, $999 is a lot of money to spend on a graphics card. But there’s no denying that AMD’s card offers better overall value and performance per dollar than Nvidia’s RTX 4080. As a result, most buyers looking for a card for peak 4K play or truly extreme esports frame rates at lower resolutions will choose the AMD card. As a result, it is our Editors’ Choice for high-end gaming graphics cards.
Source: PCMag.com
