Overclocker Cranks An Arc iGPU To 4250MHz Setting A GPU Frequency Record
To put this in perspective, 4250 MHz is more than double the iGPU's stock 2 GHz clock. From the perspective of "overclocking", reaching that speed took the usual bag of overclocking tricks: massive cooling, brutal overvolting, and a willingness to push the silicon as far as it would go. In this case the efforts were anything but "usual", though—the team used liquid nitrogen cooling, a 1.7V core voltage, and temperatures as low as -170°C. It also took the right hardware, including ASUS' overclocker-friendly ROG Maximus Z890 Apex motherboard and assistance from ROG's own Shamino, a legend in the LN2 scene.
The record was first hit in a private session ahead of Computex and later reproduced live on stage at the G.SKILL OC World Record event. Validation came via GPU-Z, showing an 85X graphics ratio. If that sounds off, it's because (as Skatterbencher explains in his blog) Intel's integrated GPU halves the input base clock before applying the multiplier: 100÷2 is 50, and then ×85 gets you 4250 MHz.
What's more interesting than the number, though, is what happened as they closed in on 4 GHz. Benchmark scores—using Novabench, Furmark, 3DMark Speed Way, and GPUPI 1B—started to level off, even as clocks kept climbing. Turns out, iGPU clocks alone weren't enough. Digging deeper, the team discovered that increasing the reference clock from 100 MHz to 110 MHz restored performance scaling, improving GPUPI 1B times by about 10% at the same frequency. That suggests the graphics die-to-die (D2D) interface may be acting as a hidden bottleneck once frequencies get high enough.
The author is a well-documented fan of integrated graphics gaming, so of course we have to talk about the benchmarks. XOC never results in configurations stable enough for complete benchmark passes, though. To get some real performance numbers, Skatterbencher had to scale things back to "only" 3900 MHz. Paired with DDR5 memory at 8600 MT/s, the chip held steady through all tests and the results are awesome: a geomean improvement of over 75% and some results climbing over 90%, with the largest improvement being a 99% uplift in Novabench GPU.
Of course, this isn't about practical gains for your next gaming session, but it does show how far Intel's integrated graphics have come—and how much juice is hiding under the hood when you throw all thermal sanity out the window. It's not something you'll replicate in your living room, but as a tech demo? Our hats are off for Skatterbencher and team.