30 years of CPUs at Tom's Hardware – looking back on three decades of processors

3 hours ago

Tom’s Hardware was founded in 1996 and gained early fame with a controversial pre-release review of the Intel Pentium II, leading to Intel backing down after legal threats.
The late 1990s saw AMD challenge Intel with the K6 and Athlon (K7), culminating in the 1 GHz milestone and the Pentium III 1.13 GHz recall due to stability issues.
In the early 2000s, Intel’s Pentium 4 (NetBurst) dominated clock speeds, but AMD’s Athlon XP and later Athlon 64 proved that performance isn't just about raw frequency.
The introduction of dual-core CPUs (AMD Athlon 64 X2 and Intel Core 2 Duo) shifted focus to multi-threading, with Intel’s Core 2 Duo marking a major turnaround.
Intel’s Nehalem architecture (2008) brought Hyper-Threading back and established a clear lead over AMD, which struggled with the flawed Bulldozer architecture (2011).
AMD’s Zen architecture (2017) revitalized competition, first with Ryzen CPUs, then 3D V-Cache, eventually surpassing Intel in gaming and productivity by 2020.
Intel responded with hybrid architectures (Alder Lake, Raptor Lake), but Arrow Lake (2024) underperformed, while AMD continues to lead with Zen 5 and X3D chips.
The CPU market now faces challenges like rising DRAM prices and shifting focus to AI, with future architectures (Zen 6, Nova Lake) delayed to 2027.

Intel launched the Core 2 Duo (Conroe) on July 27, 2006, marking 20 years since its release.
The Core 2 Duo outperformed AMD's Athlon 64 family in all areas, including gaming, restoring Intel's performance lead.
It ended the GHz race by emphasizing architectural efficiency over clock speed, delivering 40% better performance and 40% lower power consumption.

AMD's software and hardware have significantly improved, shifting from a 0% to a high chance of success against Nvidia's CUDA moat, but key risks remain.
Two major risks: 1) Helios rack production challenges due to cable design and Ethernet retimer requirements; 2) Lack of stable GPU clusters for internal software development and CI testing, slowing progress.
AMD's MI455X silicon leads in logic and memory integration (2nm, 12 HBM4 stacks, 432GB), but microarchitecture and software still lag behind Nvidia's Rubin.

The author's Intel Core Ultra 9 285K CPU died twice, leading them to replace it with an AMD Ryzen 9 9950X3D.
The CPU failure occurred after a CUDA build attempt and a batch job, but the exact cause is unclear; heat was ruled out as the room was air-conditioned and CPU temperatures were within specifications.
The author chose the 9950X3D for better performance in many benchmarks and the ability to control core preference (cache vs. frequency) under Linux 6.13.

AMD launched Helios, its first rack-scale AI platform, competing directly with Nvidia's Vera Rubin.
Helios claims 50% more HBM4 memory, 15-25% higher AI training performance, and 30% better performance per dollar than Nvidia's system.
The platform uses 72 MI455X GPUs with CDNA 5 architecture, featuring 2nm compute dies and 3D packaging.

A provocative paper questions the need for GPUs, arguing that CPUs are evolving with vector and matrix engines, HBM memory, and integrated interconnects.
The discussion is prompted by the all-CPU 'LineShine' supercomputer, the world's fastest AI/HPC system in recent Top500 rankings.
Historically, GPUs gained prominence due to CPUs lacking sufficient flops and memory bandwidth, but CPU designs are now catching up with enhanced math capabilities.

Intel launched the Core 2 Duo (Conroe) on July 27, 2006, marking 20 years since its release.
The Core 2 Duo outperformed AMD's Athlon 64 family in all areas, including gaming, restoring Intel's performance lead.
It ended the GHz race by emphasizing architectural efficiency over clock speed, delivering 40% better performance and 40% lower power consumption.

AMD's software and hardware have significantly improved, shifting from a 0% to a high chance of success against Nvidia's CUDA moat, but key risks remain.
Two major risks: 1) Helios rack production challenges due to cable design and Ethernet retimer requirements; 2) Lack of stable GPU clusters for internal software development and CI testing, slowing progress.
AMD's MI455X silicon leads in logic and memory integration (2nm, 12 HBM4 stacks, 432GB), but microarchitecture and software still lag behind Nvidia's Rubin.

The author's Intel Core Ultra 9 285K CPU died twice, leading them to replace it with an AMD Ryzen 9 9950X3D.
The CPU failure occurred after a CUDA build attempt and a batch job, but the exact cause is unclear; heat was ruled out as the room was air-conditioned and CPU temperatures were within specifications.
The author chose the 9950X3D for better performance in many benchmarks and the ability to control core preference (cache vs. frequency) under Linux 6.13.

AMD launched Helios, its first rack-scale AI platform, competing directly with Nvidia's Vera Rubin.
Helios claims 50% more HBM4 memory, 15-25% higher AI training performance, and 30% better performance per dollar than Nvidia's system.
The platform uses 72 MI455X GPUs with CDNA 5 architecture, featuring 2nm compute dies and 3D packaging.

A provocative paper questions the need for GPUs, arguing that CPUs are evolving with vector and matrix engines, HBM memory, and integrated interconnects.
The discussion is prompted by the all-CPU 'LineShine' supercomputer, the world's fastest AI/HPC system in recent Top500 rankings.
Historically, GPUs gained prominence due to CPUs lacking sufficient flops and memory bandwidth, but CPU designs are now catching up with enhanced math capabilities.

Hasty Briefsbeta