AMD Radeon RX 6900 XT, review

This graphics card takes a very heavy load on its transistors. On the one hand, it should be consolidated as a more ambitious proposal than the Radeon RX 6800 XT, a solution that we were able to thoroughly test just a few days ago and that left us with a great taste in our mouths due to its interesting price / performance ratio. But his biggest challenge is not this. The most compelling challenge ahead is proving that you can measure yourself against the best that NVIDIA currently has . With the more powerful GeForce RTX 30 .

During the presentation of the Radeon RX 6000 family of graphics cards at the end of last October, Lisa Su, the CEO of AMD, assured that this graphics card rivals the NVIDIA GeForce RTX 3090 in performance. However, the official price of the Radeon RX 6900 XT is 550 euros lower than that of the more powerful GeForce RTX. We haven’t had a chance to review the RTX 3090 yet, but we have taken the pulse of the ambitious Radeon RX 6900 XT by pitting it against two graphics monsters: the GeForce RTX 3080 and the Radeon RX 6800 XT . We promise you a fierce duel.

AMD Radeon RX 6900 XT: Technical Specifications

The graphics processors of the Radeon RX 6000 family debut the RDNA 2 architecture . In fact, this is their main asset because it is what clearly distinguishes them from the GPUs of the Radeon RX 5000 family. In the next section of this analysis we will investigate the main characteristics of this architecture, but now it is good to know that the New chips are being manufactured, like their predecessors, by Taiwanese semiconductor maker TSMC using 7nm photolithography.

These GPUs have gone from the 10.3 billion transistors of the Radeon RX 5700 XT graphics processor to the 26.8 billion of the new Radeon RX 6900 XT and 6800 XT.

However, the complexity of the graphics processors of the Radeon RX 6000 family is significantly greater than that of their predecessors. And it is that these GPUs have gone from the 10 300 million transistors of the Radeon RX 5700 XT graphics processor to the 26 800 million of the new Radeon RX 6900 XT and 6800 XT.

Maintaining the same integration technology of the previous generation and multiplying by 2.6 the number of transistors has a direct impact on the space occupied by the GPU core, which in the new Radeon RX 6000 reaches a remarkable 519 mm 2 . In the section dedicated to our benchmark, we will see if the cooling system of the Radeon RX 6900 XT reference design manages to keep the GPU permanently below its maximum temperature threshold.

Radeon RX 6900 XT and 6800 XT graphics processors feature more compute units, more stream processors, and more ROP units (rendering units) than Radeon RX 5700 XT GPUs. But this increase in the brute force of the new GPUs was predictable. What was not easy to guess is that, in addition, they would be able to work at clock frequencies above 2 GHz .

Radeon RX 6900 XT and 6800 XT graphics processors feature more compute units, more stream processors, and more ROP units (rendering units) than Radeon RX 5700 GPUs

CPUs broke through this barrier a long time ago, but GPUs had gotten us used to running at significantly lower clock rates. They are still lower than those of microprocessors, that has not changed, but now the distance that separates them is a little less . Later we will review the impact that this increase in clock frequency has on the temperature of the GPU in high load.

One very cool feature that the Radeon RX 6900 XT shares with the 6800 XT and 6800 models is its VRAM memory. These three cards are supported by 16 GB GDDR6, while the GeForce RTX 3090 and RTX 3080, which are the NVIDIA models with which the RX 6900 XT is called to compete, have 24 GB GDDR6X and 10 GB GDDR6X respectively. As the generation progresses we will find out whether or not it is crucial to bet on 16GB of VRAM or more to play at 2160p in the next titles.

AMD RADEON RX 6900 XTAMD RADEON RX 6800 XTAMD RADEON RX 6800
ARCHITECTURERDNA 2RDNA 2RDNA 2
PHOTOLITHOGRAPHY7 nm7 nm7 nm
TRANSISTORS26.8 billion26.8 billion26.8 billion
CORE AREA519 mm2519 mm2519 mm2
CALCULATION UNITS807260
BEAM ACCELERATORS807260
STREAM PROCESSORS512046083840
GPU CLOCK FREQUENCYUp to 2015 MHzUp to 2015 MHzUp to 1815 MHz
MAXIMUM GPU FREQUENCY2250 MHz2250 MHz2105 MHz
MAXIMUM PERFORMANCE IN SIMPLE PRECISION23.04 TFLOPS20.74 TFLOPS16.17 TFLOPS
MAXIMUM PERFORMANCE IN MEDIUM PRECISION46.08 TFLOPS41.47 TFLOPS32.33 TFLOPS
MAXIMUM TEXTURE FILLING RATE720 GT / s648 GT / s505.2 GT / s
ROP UNITS12812896
MAXIMUM PIXEL FILL RATE288 GP / s288 GP / s202.1 GP / s
AMD INFINITY CACHE128 MB128 MB128 MB
INTEGRATED MEMORY16GB GDDR616GB GDDR616GB GDDR6
MEMORY BANDWIDTH512 GB / s512 GB / s512 GB / s
MEMORY INTERFACE256 bits256 bits256 bits
MAXIMUM CONSUMPTION300 watts300 watts250 watts
PRICE1099.90 euros669.99 euros599.99 euros

The strengths of the RDNA 2 architecture, explained

The technologies implemented by AMD in the Radeon RX 6900 XT graphics processor are identical to those of the Radeon RX 6800 XT GPU that we reviewed several weeks ago, so this section of the review is the same as the article we dedicated to this latest graphics card. If you have already read our analysis of this last solution and do not need to review what these innovations consist of, we suggest that you skip this section. Otherwise, here is the most interesting information.

The most ambitious goal AMD engineers set for themselves during the design phase of the RDNA 2 microarchitecture was to increase the energy efficiency of their new graphics processors and enable them to run at higher clock frequencies without the need for a more photolithographic process. advanced. And they seem to have succeeded. According to AMD itself, the GPUs of the Radeon RX 6000 family propose an increase in the performance / watt ratio of up to 54% compared to the RDNA architecture, a very notable improvement in which we will try to investigate in the section of this article dedicated to our test bench.

The first slide that is worth stopping at reflects some of the features of the new AMD graphics cards that users are interested in learning about. Communication with the CPU is solved by a 16-line PCI Express 4.0 link , and the memory interface, which, as we have seen, uses a 256-bit bus, manages to reach a maximum transfer speed of 16 Gbps (we will investigate later in the novel cache architecture of these graphics cards). One more note: the HDMI outputs of these Radeon RX 6000 implement the 2.1 standard .

The RDNA 2 microarchitecture manages, as we have seen, to significantly increase the performance / watt ratio that the first generation of RDNA offers us. To make this possible, AMD engineers have worked on several fronts, among which are the refinement of the microarchitecture itself, the redesign of the rendering channel, the optimization of the displacement of data packets and the increase in performance per clock cycle , a strategy in which the novel cache architecture that we are about to investigate takes on a crucial role.

In the next slide we can see what the new cache hierarchy implemented by AMD looks like in the GPUs of the Radeon RX 6000 family. In addition to using level 0, 1 and 2 caches, they are supported by a new cache with a capacity of 128 MB that acts as an intermediary between the level 2 cache and the VRAM memory integrated in the graphics card. According to AMD, the write strategy in the cache sublevels that its engineers have developed manages to minimize the movement of data packages, latency and consumption by intelligently reusing as much data as possible.