NVIDIA, AMD and Intel aim for maximum power at CES 2025

There was no doubt that NVIDIA's RTX 5000 GPUs would be one of the biggest stories at CES 2025, and I figured Intel and AMD would be unveiling some new hardware of their own. But what I didn't expect was that each of these companies would have their foot on the gas differently when it comes to the power of their chips. After all, we've spent the last few years highlighting AI processors for PCs that focus on efficiency rather than pure performance.

While NVIDIA RTX 5000 GPUs seem to deliver the performance leap we've come to expect over 2022-era cards, AMD is also reimagining the mobile workstation experience with its Ryzen AI Max chipscombining powerful graphics and a huge amount of built-in memory. Intel is not sitting still either – it is finally bringing Arrow Lake into the arena of high-performance and gaming devices with its Core Ultra 200HX chipswhich can reach 24 cores and a frequency of 5.5 GHz.

I'm not just talking about power in a pure performance sense. The $1,999 NVIDIA RTX 5090 requires a 1000W power supply to run and draws up to 575W. Meanwhile, Ryzen AI Max chips can consume up to 120W. Intel Core Ultra 200HX chips have power up to 120 W. Obviously, neither of these devices are intended for those concerned about their energy bills or potential laptop battery life.

So what do you get for all this energy consumption? AMD says the RTX 5090 will deliver roughly twice the performance of its previous flagship, the $1,499 RTX 4090. Cyberpunk 2077 In the full ray-traced demo, the 4090 hovered around 108fps, while the 5090 topped out at 240fps. This frame count is a bit controversial, however, since the RTX 5090's DLSS 4 AI scaling generates three frames for every frame rendered in native mode. Most people may find the end result smoother, but some gamers may question the integrity of the so-called false frames.

It's these same AI-generated frames that allow NVIDIA to claim that the $549 RTX 5070 can be as powerful as the 4090. That may be true when it comes to pure frames per second count, but it's definitely not Same for rasterized images. performance without DLSS 4.

AMD's Ryzen AI Max chips don't aim for the same graphical heights as NVIDIA's new GPUs, but they're still notable for the sheer amount of hardware they contain. The top-of-the-line Ryzen AI Max+ 395 processor features 16 Zen 5 CPU cores, 50 TOPS of AI performance, and 40 RDNA 3.5 GPU compute units. According to AMD, it should be on par with the 14-core Apple M4 Pro chip (and even faster in Vray tests), and in 3D rendering it is 2.6 times faster than the Intel Core Ultra 9 288V.

IN interview with AMD CVP and Product CTO Joe MacriHe told Engadget that the success of Apple Silicon is the main reason for the existence of Ryzen AI Max. “Apple has shown that consumers don't care what's inside the box,” he said. Macri later noted: “I always knew, as we built APUs, and I always pushed for this big APU that I could build, a system that was smaller, faster, and could deliver much higher performance at the same level.” ” power.”

AMD also briefly introduced RDNA 4 graphics at CES, although for now the company is clearly targeting the mid-range rather than the NVIDIA RTX 5090. Notably, AMD will be introducing a new AI-based upscaling technology in its RDNA 4 GPUs – FidelityFX Super Resolution 4 (FSR 4). This should finally give AMD the ability to directly compete with NVIDA's DLSS, which has looked better than earlier versions of FSR for years. The first RDNA 4 cards, the Radeon RX 9070 and 9070 XT, will arrive sometime in the first quarter.

Intel's presence at CES 2024 was more subdued than its competitors, but Intel loyalists will likely appreciate the new Core Ultra 200HX chips. Even though they reduce NPU performance compared to the latest AI PC hardware (by 12 TOPS instead of 48 TOPS), the Core Ultra 9 285HX looks like a 24-core beast. It will be interesting to see how it competes against AMD's Ryzen AI 300 hardware, although it likely won't stand a chance against the Ryzen AI Max when paired with a discrete GPU.