We’re heading for a flurry of high profile product launches from Nvidia, Intel and AMD. The Big Three are all releasing, or expected to release, new generations of their CPUs, GPUs, or both. Of all these, however, the topic we’ve probably heard the least about is AMD’s upcoming RDNA 3 architecture. This morning AMD released a blog post to give some color behind its claims of “over 50%” performance per watt uprising on RDNA 2. This may lend credence to a new rumor that these GPUs could reach clock speeds close to 4 GHz.
This rumor comes to us via an Eternal leaker on Twitter @9550pro. The bold claim of “Nearly 4GHz GPU” is only backed up by an accompanying photograph of Dr. Lisa Su on stage in front of a Radeon GPU image. Current RDNA 2 GPUs can sometimes be squeezed to clock around 3GHz, however 2.6GHz is more typical. Gaining almost a gigahertz more would be quite an achievement. So, is this leak just a breath of hot air?
Almost on cue, AMD released its blog post on the company’s commitment to performance per watt this morning. In it, AMD outlines the demand for new techniques such as ray tracing, variable rate shading, and advanced scaling technologies, while recognizing that these drive ever-increasing power consumption. high. AMD claims its Radeon 6000 series delivered both generational efficiency improvements and better frame rates per watt than competing NVIDIA solutions with few exceptions.
These are existing solutions, so what does that mean for the RDNA 3 “Almost 4 GHz” claim? AMD goes on to explain how it “continues [its] push for more efficient gaming with AMD RDNA 3 architecture.” The company says the move to a 5nm process combined with the chip packaging strategy helps achieve “>50% better estimated performance per watt” compared to the current RDNA 2 architecture.
AMD says these architectural improvements are complemented by enhancements to the adaptive power management system used in RDNA 2 and a new generation of Infinity Cache. Adaptive power management features adjust GPU power consumption based on current workload. This helps GPU components avoid consuming power unnecessarily. AMD Infinite Cache is located between the L3 cache and the GDDR6 memory which reduces the dependency on the latter. This improves bandwidth and further decreases power consumption.
Other leaks have already discussed RDNA 3 dies also having a reduced surface area. The Graphics Compute Die (GCD) part should be
about 350mm²— a shrinkage of around 375mm² for comparable portions of RDNA 2. Exact size comparisons are complicated by the new approach to chip packaging, but the conclusion is that we can expect much higher density in the new architecture. In addition to matrix shrinking, this is accomplished by removing old “architectural bloat”, including support for the old XGMI GPU-to-GPU interface, legacy geometry pipeline, and legacy scan converter.
Of course, when efficiencies are achieved, companies don’t just deliver the same performance with less power. Instead, these elevations are used to produce more performance with the same power envelope. With processors of any variety, this usually manifests as higher clock speeds, the addition of extra cores, or both. As such, it’s not out of place to imagine seeing RDNA 3 clock speeds north of 3.6GHz, especially if it’s short bursts and unsustained averages. It could also be that those near 4.0GHz clocks were only achieved with exotic cooling and extreme overclocking.
Clock speeds are nowhere near the full performance picture, either. GPUs are highly parallelized computing devices that cannot be properly evaluated with a single hardware specification. Such high clock speeds are certainly a promising metric, although the focus on efficiency could be an indicator that AMD thinks it could lose out to NVIDIA in terms of raw power. We will closely follow how these pile up against the Nvidia GeForce RTX 4000 Series and Intel Arc Alchemist cards in the weeks and months to come.