Senior Vice President of AMD, sam naffzigerwas interviewed in VentureBeat, a technology website located in San Francisco. In this he spoke of the GPUs of amd, efficiency, consumption, cache and chiplet design. In addition, they comment on their current position against their main competitor Nvidia and this compare both companies.
Sam Naffziger has been with AMD for 16 years and currently serves as Senior Vice President and Product Technology Architect. In his areas of expertise, he is dedicated to the energy efficiency department, in order to respond to the difficult conditions of the company. And it is that a few years ago AMD imposed an initiative of Efficiency energetic of 25x for 2020something that met APUs AMD Ryzen 3rd Gen with integrated Radeon Vega graphics (Renoir).
GPU TDP skyrockets, soon over 600W
It’s no mystery that the next generation of GPUs are going to consume a lot of power, although it’s surprising to see the increase from previous years. According to AMD, “energy consumption has skyrocketed, request exceed profitsThus, it confirms that the figures of 600W of PDT on front GPUs 2025although seeing the graph makes us fear that they won’t end up being 800W.
“The frequencies of our GPUs now come from over 2.5GHz, which means reaching levels never reached before. We systematically reviewed the design, we redesigned the low-level critical paths, the things that interfere with the high frequency, and we did it in an energy-efficient way.”
By removing the frequency, they confirm that the main increase in consumption comes from the increase in tension graphics. This one has a quadratic effect in the consumption energetic (high square). Therefore, they illustrate Nvidiawhich could reach those 2.5 GHz that AMD mentions but doing overclock and raise the voltage to 1.2 or 1.3 volts. In the meantime, AMD confirms reaching these high frequencies with more modest and therefore more efficient voltages.
“We looked carefully at Nvidia’s designs and what they did, and of course we looked to improve it,” Sam Naffziger told VentureBeat.
AMD hopes to double the 128 MB of cache, while future NVIDIA GPUs have 96 MB L2
AMD is very happy with the addition of Infinity Cache, as it was a stuck technology in processors. Bringing it to graphics cards manages to increase its efficiency and performance at high resolution, such as 1440p That is 4K. In fact, they set up 128 MB of Infinite Cache on current AMD architecture RDNA2something that surpasses its main competitor, NVIDIA.
Additionally, next-gen Ada Lovelace (RTX 40) GPUs are expected to arrive with up to 96 MB of hidden L2, a much larger increase during the 6 MB which has its current high end. According to AMD, they opted for a smallest bus width, but with a large cache, something that worked well. Additionally, the company expects to double and even triple these 128 MB of hidden on the next generation of AMD RDNA 3 GPUs.
They talk about chiplets and Intel, saying Raja Koduri is a visionary
When asked about chiplets and the rest of the competition, Sam mentions that they have an advantage over NVIDIA, as he hasn’t jumped on the bandwagon of chiplets. Over time, they assure that it will eventually do it, while its rival Intel has already achieved it with Ponte Vecchio. Thus, AMD’s VP celebrates Intel’s decision to use chiplets in its Ponte Vecchio GPU, calling it “the model child of chiplet extremes.”
“Raja Koduri is a visionary. It paints a fantastic and compelling picture of the future of gaming and the features needed to take the gaming experience to the next level. He’s very good at it. As for the practical implementation of silicon, his background is in the software realm.”
“Without a doubt, helped amd to improve our software suite and its features. I worked closely with Raja, but didn’t join the graphics group until he left. took a sabbatical Yes went to intel. So when it comes to performance per watt, it wasn’t really Raja’s footprint. But some of the software dimensions and so on, said AMD’s Sam Naffziger.”