It's been a packed week in hardware. Let's jump straight in.
AMD Is Done Being the "GPU Alternative"
Lisa Su's Advancing AI keynote this week wasn't just another product launch — it was a repositioning. AMD is no longer content being the cheaper Nvidia alternative. The message was clear: they want to be a full-stack AI infrastructure vendor.
The centerpiece is the MI450 paired with the Helios rack system. AMD claims 10-15% more performance at the same rack power compared to Nvidia's equivalent, and "up to 30% more tokens per dollar." Those are bold numbers, and honestly, I've heard similar claims before from AMD that didn't fully materialize in real-world deployments. But this time feels different — they named OpenAI as a deep partner, with joint engineering teams already running GPT-class workloads on Helios. That's not a paper launch.
The partner lineup was genuinely impressive. Su mentioned demand from "the largest AI labs to hyperscalers," and Wells Fargo just doubled down on AMD stock after the Anthropic deal — AMD is up over 170% year-to-date. The market is buying the story.
Still, a few things nag at me. CUDA isn't going anywhere. AMD's ROCm software stack has improved, sure, but it's not a drop-in replacement, and enterprise buyers know this. Promising "more tokens per dollar" is one thing; getting actual deployment at scale across diverse workloads is another. We'll see how this plays out over the next two quarters.
Nvidia's Vera Chip: A Quiet but Serious Move
While AMD was making noise in San Francisco, Nvidia dropped something that didn't get enough attention — the Vera chip, built around the Olympus core. This is Nvidia's first custom CPU core for the datacenter, and their first custom core design since the Denver and Carmel efforts from the Tegra era, which was... almost a decade ago.
From my perspective, this is Nvidia quietly telling the world they're not just a GPU company anymore. Vera isn't meant to replace x86 overnight — it's a strategic play. By designing their own CPU core, Nvidia controls the entire compute stack in a way neither AMD nor Intel can match. Tight integration between CPU, GPU, and networking on the same platform? That's a serious moat.
The real question is whether the Olympus core can actually compete with AMD's EPYC or Intel's Xeon on general-purpose workloads. Nvidia's previous custom CPU efforts were... let's say mixed. Denver and Carmel never really took off. But the datacenter is a different beast, and Nvidia has way more resources now. Keep this in mind — if Vera delivers, it reshapes the competitive landscape.
The Chinese GPU That Actually Plays Games
This one surprised me. A YouTuber built a gaming PC with zero Intel, AMD, or Nvidia components — using a Zhaoxin KX7000 CPU and a Lisuan LX7G100 GPU — and it actually ran modern games at playable frame rates.
The numbers: Forza Horizon 5 at 62.5 FPS, Red Dead Redemption 2 at 39 FPS, Cyberpunk 2077 at 32.8 FPS, and Shadow of the Tomb Raider at 62 FPS in DX12. That's roughly RTX 3060 territory, despite the Lisuan being advertised as RTX 4060-class. The LX7G100 packs 6,144 cores, 12 GB of GDDR6, and a 225W TDP with WHQL certification.
Now, before anyone gets too excited — there are real caveats. DX11 titles still struggle badly due to immature drivers, and the Zhaoxin KX7000 CPU was a clear bottleneck in CPU-bound scenarios. You're not going to build a competitive gaming rig around this any time soon. But the fact that a fully domestic Chinese x86 system can run Cyberpunk at over 30 FPS is not nothing. A year ago, the same setup with a Moore Threads MTT S80 couldn't even get close.
This is still very early. But the trajectory is worth watching, especially for anyone concerned about supply chain diversification.
Quick Bits
- VCI Global launched VGAIN Cloud in Malaysia, a "post-quantum-ready" sovereign AI platform deploying Nvidia Blackwell-class GPUs and Intel-powered CPU systems. Architecture targets up to 1,000 GPUs. Another sign that AI cloud infrastructure buildout is accelerating outside the usual US-China axis.
Alright, that's the week. Three very different stories, but they all point in the same direction — the hardware landscape is shifting faster than most people realize. AMD is building systems, not just chips. Nvidia is building CPUs. And China is slowly but surely building competitive alternatives.
None of these stories are resolved yet. That's what makes this space interesting.
Speaking of specs and configurations — if you are comparing hardware options and need a practical starting point, Manual Assistant has straightforward reference material. No sign-ups, just the essentials.
Top comments (0)