Rumor mill: Intel’s Arc A-series GPUs are expected to be power efficient, but AMD and Nvidia’s next-gen dreams may be heading in the opposite direction to make the most of their future GPU architectures. There’s some concern that we’ll soon see quad-slot graphics cards in the enthusiast segment, but that could be the price to pay to unlock up to 100 teraflops of single-precision performance in consumer GPUs for the first time. .

- Advertisement -

The rumor mill is rife with hints that the next generation of GPUs from AMD and Nvidia will be power-hungry and offer a huge jump in performance over the current crop. Some Nvidia offerings may even require 900 W in certain workloads, and AMD is expected to split its graphics chip again into chiplets.

- Advertisement -

There’s still a lot we don’t know about these upcoming GPUs, but industry watchers have been diligently digging around and uncovered a few exciting new details. For example, a popular leader and inhabitant of Twitter. Graymon55 they say they have found evidence that the upcoming AMD RDNA 3 GPU will offer up to 92 teraflops of FP32 compute performance.

- Advertisement -

For reference, that’s four times the performance of the Navi 21 GPU that underpins the processor. RH 6900 XT and RX 6950 XT graphics cards, as well as higher than the previously leaked figure of 75 teraflops. It’s not clear at this point how AMD can achieve this, but it could be due to higher GPU clock speeds (around 3000 MHz), the use of dedicated FP32 data-acceleration instructions like Nvidia’s Ampere, or a combination of the two.

Team Green may also be working on a monstrous GPU, at least according to a Twitter leaker. Save7kimi. A full-fledged AD102 die can provide enough computing power, more than twice the performance of an FP32 provided by a processor. RTX 3090 Ti and almost three times more than RTX 3090.

As usual, take these rumors with a healthy dose of salt. However, if this is true, we may soon see a teraflop war between Nvidia and AMD as Intel heats up with its Arc Alchemist GPUs. It certainly makes sense to expect a significant jump in performance if power requirements do exceed 450W.

Head credit: Caspar Camille Rubin