AMD introduces new graphics card generation - RX 7900 XT(X) with two video engines
[10:52 Fri,4.November 2022 by Rudi Schmidts]
The new generation of AMD graphics cards is called RDNA 3. The first chip of this series, on which the new RX7900 graphics cards are based, is called Navi 31. For the first time, a GPU appears in the chiplet design, which was already significantly responsible for AMD's successes in the CPU area in recent years. Here, the functional units in the processor are distributed over several dies, which not only increases the yield and lowers the production costs. It also allows more functions to be implemented in one chip than with a monolithic design on a single large die.
Navi 31 already consists of no less than seven dies at the start. The central Graphics Compute Die (GCD) is manufactured at TSMC in the N5 process, around which there are six Memory Cache Dies (MCD) that can be manufactured in the cheaper N6 process. All in all, the seven chips together house around 58 billion transistors.
AMD Navi31 GPU
Additional cost advantages arise for AMD from the fact that not every graphics card has to be delivered with six functioning MCDs. Only the top model Radeon RX 7900 XTX relies on the full configuration, while the Radeon RX 7900 XT uses only five active MCDs.
The XTX version thus offers a 384-bit wide memory connection with 24GB GDDR6, resulting in a transfer rate of 960 GB/s. The XT version, on the other hand, connects 20GB over 320 bits at 800 GB/s.
Interesting for video editors and creators: In contrast to Nvidia's latest generation - which remains at DisplayPort 1.4 - the new AMD GPUs have DisplayPort 2.1 connections. This allows resolutions of up to 8K165p or 4K480p to be output. The output is also possible with up to 12 bits per colour channel.
As with Nvidia's new generation, the hardware codecs of the Media Engine are dual. For example, an AVC and an HEVC codec can work in parallel, which will soon be supported by Adobe's Premiere. De- and encoding can also be used in parallel for transcoding tasks. In addition, encoding is supposed to be "AI-enhanced" (?).
Even more exciting, however, would be a concrete statement from AMD on whether the decoders can finally handle 10-bit 4:2:2 formats, which are becoming increasingly important in video editing. So far, only Intel and Apple can do that.
Many had expected that AMD's new graphics cards would also support PCIe 5.0, because this is after all also offered by the new AM5 platform from their own company. But like Nvidia, AMD is (still?) sticking to PCIe 4.0 with the current GPU models. Also not unimportant: with a maximum of 330W for the top model, AMD's graphics cards consume noticeably less power than Nvidia's new models, which is why a 2x8-PIN power connector is also sufficient for AMD's cards. Upgraders will therefore probably not need new power supplies here.
AMD RX 7900 XTX
Last but not least, the prices are a real challenge: with 43.0 TFLOPS and a faster memory interface (800 vs. 736 GB/s) for 899 US dollars without taxes, the RX 7900 XT should already be at least on the level of an RTX 4080 with 16GB in terms of video editing performance. For just more, the RX 7900 XTX comes in at 56.5 TFLOPS, putting it solidly between an RTX4080 and RTX 4090. Both cards should be available from 13 December. The prices in euros are not yet known, but should end up around 1,200 and 1,300 euros due to the strong dollar and additional VAT.
;)
;)