Radeon RX 7950 XTX and XT - AMD accidentally reveals RTX 4090 competitor

[12:39 Wed,3.May 2023 by Rudi Schmidts]

AMD is currently doing well by far not as bad as Intel, but the bigger rival has become Nvidia anyway.

Hardly any other IT company is as highly valued and in focus as Nvidia at the moment, precisely because Nvidia is associated with the current AI developments like no other company. The reason for this is again the wide use of Nvidia chips in GPUs, which are used for most AI projects today.

If you look at the pure hardware performance of the current graphics cards, AMD has offered the better price-performance ratio in this area for generations, but the software stack is the main focus for AI applications. And here, Nvidia clearly dominates with its CUDA libraries, which are currently used in most AI projects.

However, AMD has also started a race to catch up in this area in the last few years with the so-called ROCm platform, which has been able to catch up more and more with typical CUDA applications in the past few months. Including automatic conversion tools from CUDA to ROCm.

And because Nvidia is currently paying for the CUDA advantage of its graphics cards with princely surcharges, many price-conscious users follow every ROCm innovation from AMD to see if this might not enable AI training on the home computer at a significantly lower price.

And so the community has been waiting with excitement for the new ROCm version 5.5 for days, with which current AMD GPUs should finally be able to be used with generative AIs without any problems. And of course for meaningful benchmarks, whether and if so how much slower the AMD GPUs will be in practice. Because

for AI, you will primarily need a lot of GPU memory in the future, and AMD currently offers it at a significantly lower price.

Now, first builds of the ROCm version 5.5 have appeared on Github for a short time, but they were quickly removed again and the reason for the withdrawal seems to be mainly in the enclosed information from AMD.

Because two top models appear in the

GPU compatibility list, about which there have always been rumors, but which have never been directly mentioned or confirmed by AMD: An AMD Radeon RX 7950 XT as well as an AMD Radeon RX 7950 XTX.

Will the AMD RX7950 calculate on the level of an Nvidia RTX 4090?

Both models seem to be very similar to

the current top models RX 7900 XT(X). Thus, 20 and 24 GB RAM with 800 and 960 GB/s connection, respectively. The only difference seems to be "only" a higher boost clock. However, according to first estimates, this should increase the FP32 computing performance of the card to about 81 TFLOPS - which is pretty much on par with the current Nvidia top model RTX 4090. By the way, this value is also relevant when used as a video editing accelerator under Premiere Pro or Resolve.

Should ROCm catch up with the relevant CUDA features, we could see a real alternative to Nvidia&s top model here for the first time in a long time - which might also cool down the prices in this segment a bit in the future.

However, the topic AI is so hot that the amount of memory will be the decisive argument in the medium term. So, the "cheapest" graphics cards with 48-64 GB of memory will be really exciting in the next generation.

more infos at bei www.reddit.com

deutsche Version dieser Seite: Radeon RX 7950 XTX und XT - AMD verrät versehentlich RTX 4090 Konkurrent