Nehalem is a 10-year-old who uses Intel products.
Nehalem is a 10-year-old who uses Intel products.
Intel's desktop CPUs from November 2008 onward use the Nehalem microarchitecture. Even the Core i9 9900K relies on this older design. Since the launch of Sandy Bridge, all consumer Intel CPUs can run at 5GHz with a liquid cooler. Over the years, Intel has steadily raised base and boost clock speeds to give the impression of better performance, though in reality most desktops already meet that speed with proper cooling. By 2018, further enhancements on Nehalem (Coffee Lake) proved difficult. In late 2019, even the Coffee Lake revision still couldn't see meaningful upgrades. It appears that beyond a certain point, only shrinking the process node brings noticeable gains. Labeling each new generation as an "architecture" can be misleading since Intel has been evolving the same core since 2008. I recommend moving away from Nehalem microarchitectures and investing in a fresh design. The relatively new Ryzen architecture is efficient, still offers room for improvement, and already outperforms Intel in servers with much higher performance per cycle. What are your thoughts on this? I'm happy to adjust if needed.
I owned a Nehalem board around 2010 or 2011 (Xeon W3520 + Asus Rampage ROG 2 Gene mtax). That motherboard was extremely hot—I experienced some screen issues. The overclocking worked well up to 4.4GHz. After selling that X58 rig, I switched to an AM3+ platform. It turned out to be the worst decision I ever made.
There are minor IPC enhancements since SandyBridge. They actually boost yields and sometimes improve node performance. The architecture remains consistent, similar to GCN's use in AMD's 7800 series. Depends on the specific technology and results. Second and third generation chips can reach 5GHz but tend to run hot at high voltages. It seems they're making progress without major overhauls. Ryzen shows some FX traces, so improvements are likely limited. Focus has been on 10nm, with Sunny Cove probably receiving more attention.
I purchased a used 990fx-ud7 at a low price since it was version 1.0 without LLC. I tested it on a Phenom X6 1035t at 4.2ghz, then acquired an FX8320E in 2014 and ran it at 4.7ghz. The 990FX-UD7 failed by late 2016, so I upgraded to an Asus 970 Aura Pro, which now serves as my secondary system.
They've continuously enhanced Nehalem since its release. Although there have been advancements, Sandy Bridge still supports 5GHz. They move at a leisurely pace—10 years and counting—and the future roadmap doesn't even reference Sunny Cove... As long as the FX traces don’t slow performance or efficiency, it doesn’t matter. It makes sense, but realistically, infinite microarchitecture improvements aren’t possible.
You seemed to think there hasn't been any major shift since Sandy Bridge. That was the point I was making. There actually have been updates. The 5GHz Skylake offers a noticeable improvement over the 5GHz Sandy Bridge version. Ice Lake represents a fresh take on "macroarchitecture." AMD has been employing GCN since 2012 and continues to do so. For CPUs with fewer cores, the Sandy Bridge lineup remains quite strong. Eventually, you'll have to balance trade-offs—some things improve while others decline.
You made a great observation. It's a valuable insight, though it may not last long.
for low core counts that could last a long time, ignoring changes in nodes. Ringbus is really tough to beat. just add an edit: I meant to say there have been iterative updates. Often with AVX or small cache tweaks, IPC rises in specific tasks. I don’t have the precise improvements between each generation (I’m too young to remember those details), but I understand the general jump from Sandybridge to Skylake—about 3 to 11% better IPC, depending on what you’re doing.