Cinebench overclock i5-4670K breaks 12-year multicore record

The chip hit 6,301 MHz during the validated run, or roughly 6.3GHz. Out of the box, the i5-4670K runs at a 3.4GHz base clock and boosts to 3.8GHz, so Luumi ran it well beyond its rated top speed.

Cinebench R11.5 is an older version of Maxon’s rendering test. It asks the processor to draw a detailed 3D scene and gives a score based on how quickly it finishes. The 10.8 points came from the multicore test, which uses every core at once.

Luumi submitted the result on October 8, 2026. It now sits first among 187 entries in HWBot’s i5-4670K category and earned 70 hardware points on the site’s ranking system.

The race for first place was close. The entry right behind Luumi belongs to an overclocker called Strat, who scored 10.76 points at 6,249 MHz. Third is Moose83, at 10.74 points and 6,231 MHz.

That means Luumi’s chip ran roughly 50 MHz faster than the runner-up and turned it into a lead of 0.04 points. In extreme overclocking, gaps like that often come down to how stable the cooling stays during a single run.

How long the old score held depends on who you ask. The first reports said it had stood for 12 years. One HWBot member, congratulating Luumi in the submission comments, called it a 14-year-old record. Either way, nobody had beaten it for a very long time.

Nothing in this build is new. Luumi used an ASRock Z97 OC Formula motherboard, a board designed for extreme overclocking with a 12-phase power design and a dedicated LN2 mode switch. That switch helps a chip boot when it is extremely cold, a moment when some processors simply refuse to start. ASRock ended service for the Z97 version in 2021, so working boards are getting hard to find.

The memory was just 4GB of XPG DDR3, tuned to 2,708 MHz with tight 8-12-7-28 timings. That sounds tiny today, but this benchmark cares about fast memory response, not capacity.

• Power came from a Corsair AXi 1600W supply.
• Storage was a Corsair Force GT SSD.
• The test ran on Windows 7 64-bit Pro.

The real star is the cooling. Liquid nitrogen boils at around minus 196°C. Overclockers pour it into a metal pot that sits on the processor in place of a normal cooler, then fight to keep that pot at a steady temperature for the whole test. Too warm and the chip loses its clock speed. Too cold and it can crash. The area around the socket also has to be insulated, because frost and condensation can kill a board.

The headline number needs context. This is the best score for one specific processor, the i5-4670K, not the best Cinebench R11.5 result for every four-core chip. HWBot’s overall 4-core top score is 24.71 points, set on much newer hardware.

Even other Haswell chips have done better. At CES 2014, an extreme overclocking event pushed a Core i7-4770K to 6.4GHz and scored 14.03 points. The i7 had Hyper-Threading, which lets each core run two tasks at once, giving it eight threads. The i5-4670K has four cores and four threads, so it is judged only against itself.

Modern chips leave both far behind. A reporter who ran the same old test on a current laptop with 12 cores and 24 threads got about four times Luumi’s score, even with the laptop set to a cool and quiet mode.

For most people, the takeaway is not to grab a flask of liquid nitrogen. It is that hardware ages more gracefully than the upgrade cycle suggests. The i5-4670K was an unlocked chip, built on Intel’s 22nm process with 6MB of cache, an 84W power rating and HD 4600 graphics. In its day, you needed an i7 to get eight threads.

HWBot keeps a separate leaderboard for every processor model, which is why a retired chip can still make the homepage. One community member said it was refreshing to see a featured result that did not involve an X3D processor or an RTX 5090 graphics card. That structure rewards skill over spending: picking a strong chip, tuning old memory and managing the cold.

There is also a practical lesson for anyone who likes old software. Cinebench R11.5 still installs and runs on Windows 11 without trouble, while some older benchmarks, such as early 3DMark tests from the DirectX 10 era and before, have problems on modern PCs that make their results invalid for HWBot rankings.