twice as fast and have up to 48 GB VRAM

The AMD RX 7900 XT and XTX are the first AMD graphics with RDNA 3 architecture. These were released on December 13, a week ago, and despite the fact that we have only been able to see two high-end models, there are already rumors about the future from AMD. Specifically, the rumors speak about the architecture RDNA 4 Y AMD GPUs that they will have until 129 TFLOPs FP32, 48GB by heart GDDR7 And till 3.5GHz frequency, among other novelties.

The latest AMD graphics cards have come to an end and now that we have been able to see reviews and comparisons with their rivals at NVIDIA, we can draw conclusions. If we take into account the power in rasterizedthe AMD RX 7900 XTX are even better than the RTX 4080 from NVIDIA, while the 7900 XT falls a bit below. However, none of these AMD graphics cards is capable of matching or surpassing NVIDIA’s flagship, the RTX 4090. This is totally logical, since this graphics card is significantly faster than the RTX 4080 and here the green company plays in a different league.

RDNA 4 GPU with Navi 41 will be much faster than RX 7900 XTX

RedGamingTech, a YouTuber leaking hardware information, has been in charge of giving us information about the architecture RDNA 4 and future GPU chips ship 41, ship 42 Y ship 43. This is a very unusual development, as most of the leaks in the GPU sector tend to be from NVIDIA. In fact, from AMD we can rarely find out anything and we usually arrive with quite a few doubts until the day of the official announcement. That said, even if the YouTuber has offered reliable information, it must be take it completely as a rumoreven more so considering that the RDNA 3 GPUs have only just been launched on the market.

So, first of all, RedGamingTech (we’ll simplify it as RGT), claims that the number of ALUs per CU of RDNA 4 is the same as RDNA 3. This implies a total of 128 ALU, a calculation that will help us to know the number of Stream Processors in the graphics. That said, we start with the chip that will correspond to the highest-end AMD RDNA 4, the ship 41.

With the Navi 41 chip, the leaker claims he will have up to 32 GB of GDDR7 memory and 4 MCD. In addition, it says that there will be another version with 24 or 48 GB GDDR7 and 6 MCD. Along with this, we talk about 144 cuwhich if we remember the proportion was 128 ALU and this would give us a whopping 18,432 Stream Processors, but if we count as AMD does it would be 64 Shaders per CU and therefore would be 9,216 Stream Processors. Also, the GPU will run at a frequency of 3.5GHz and will have a power of 129 TFLOPs FP32, almost nothing. To give us an idea, the RX 7900 XTX current has 61 TFLOPs Y 6,144 Stream Processors.

RDNA 4 Low-Mid Range Graphics Will Exceed 40 TFLOPs

Now that we have said the specifications of the Navi 41 chip that corresponds to the most powerful GPUs, we are going to move on to the upper-middle range. This will use the chip ship 42which continues to maintain the 3.5GHzalthough it will have 96 cuthe same as the RX 7900 XTX, thus giving a total of 12,288 cores or 6,144 Stream Processors if we calculate as AMD does with RDNA 3. With these data, we would talk about a theoretical performance of about 86 TFLOPsgiving on paper a 40% more performance with the same SP number.

The last of the chips would be the ship 43 and this would correspond to the low-mid range of AMD RDNA 4 GPUs. Although here we see a large cut in half with 48 cu at most, we will still have a clock frequency of 3.5GHz and for this, it will achieve 43 TFLOPs performance FP32. More than 40 TFLOPs for the low-mid range as a typical GPU RX6600 (8.9 TFLOPs) is great news, because we are talking about a lot of power (on paper) compared to past generations.

Now, regarding the overall RDNA 4 architecture improvements, RGT says that it’s quite likely that PCIe 5.0 will stick around and all GPUs will hit around 3.5GHz. Coupled with architecture improvements, this would give a improvement of performance per watt between 50 and 60%. Finally, AMD will not forget about Ray Tracing and therefore, the 3rd Generation of Infinity Cache together with the WMMA V2 system will allow double performance per CU. That is why, although we have only seen one 50% CU improvement with Navi 41, the performance would more than double if we take into account the TFLOPs.