The End of High-Performance Overclocking May Be Nigh

The End of High-Performance Overclocking May Be Nigh

  • 7/28/2019
  • 150 views

A little over seven years ago, I wrote a story called “ Physics, Ivy Bridge, and the Slow Death of Overclocking.” The argument is made, in essence, was that the realities of process node scaling were steadily going to worsen and overclocking headroom would continue to decline. A recent update from Silicon Lottery makes the same argument, in starker terms.

Silicon Lottery is a website that sells binned CPUs at specific frequencies and voltages, for both AMD and Intel products. Think of it as a one-stop-shop for overclockers who prefer to pay for a CPU at known-good rates rather than a chip they’ve still got to take a chance on themselves. The company has released a price list for its upcoming Ryzen CPUs, which are currently listed as out-of-stock. Yes, some of these speed grades are listed for cheaper than the actual base Ryzen 7/Ryzen 9 CPUs. No, I don’t have an explanation for that.

The overall clocks are low, as we expected. There was a profound mismatch between AMD’s clock expectations for Matisse and what the enthusiast community predicted would be possible. AMD told us at E3 that it didn’t expect to gain clock frequency on these chips at all, and that the boost it saw on its 7nm parts was an unexpected win. AMD used those frequency gains for its components and appeared to have binned aggressively. The all-core boost frequencies Silicon Lottery feels like it can commercialise, at least in the short-term, are relatively low. Demand for the binned higher rates dried up early with Ryzen 1, 2, and Threadripper, and a recent post on Reddit(found via THG) by Silicon Lottery indicates the company doesn’t necessarily expect a different outcome this time around.

SL notes that demand for Ryzen chips has been growing and that it will carry them and bin them as long as it can profitably do so, even if the clock headroom is low. The site claims to test aggressively before shipping CPUs, which leads them to advertise lower clocks than some achieve with weaker tests. I expect an overclocked to CPU to be as stable as its non-overclocked part before I’d ever considered using it. SL also notes that there’s a silicon difference between the 3700X and 3800X, with the latter hitting speeds roughly 100MHz faster than the former. Using AVX2 raises temperatures compared to not (the small FFT test in the latest version of Prime95 that we used for power consumption testing in our Ryzen 7 review reflected this).

SL then writes:

AMD has done a fantastic job here overall, and we’re very aware this is the start to the end of our company in general. As both AMD and Intel optimize their binning process more and more, overclocking will not be possible as CPUs will boost themselves on their own to the highest clocks possible.

Where’s the Headroom Going?

Two things are happening here. First, the total amount of clock headroom has scarcely budged in years. Ten years ago, a good overclock put you in 3.8–4.2GHz territory. Six years ago, a good overclock might be in the 4.6–4.7GHz range. Today, modern top-end chips target 4.6–5GHz for their all-core boost frequencies and leave very little room for overclockers to go anywhere else.

These days, winning the silicon lottery doesn’t mean finding a chip that can hit 6GHz, so much as it means finding a CPU that can hold something close to its single-core frequency as an all-core boost frequency. But the entire point of the AVFS implementation that AMD uses is that the CPU already automatically finds and “uses” that additional headroom dynamically, actively, to provide extra performance.

Or, to put it differently: Manufacturers used to have the luxury of leaving overclocking headroom in their products because they didn’t need that headroom to sell you a meaningful upgrade. Now, they do — and they’re tapping what little headroom remains for themselves.

Does this mean the complete end of overclocking? Not necessarily. AMD and Intel could both continue to offer deliberate midrange SKUs with unlocked multipliers and mid-tier prices. Lower base clocks mean these chips would still overclock well, providing enthusiasts with parts they want to purchase for economic overclocks. Properly positioned, these chips fill a need without sabotaging higher product tiers.

But there have always been two groups of overclockers, generally speaking. There’s a community of budget buyers, who overclock with a modest investment in cooling and hardware because it’s a better way to improve performance than just buying the more expensive chip, and a group of high-end buyers who jump for upper-tier device and then OC it to improve it more. The second group tends to spend more than the first, and that’s the group that’s most likely to cease existing altogether. Once water-cooling is no longer enough, there’s only one plausible jump left — single-stage freon. And frankly, the work it takes to insulate a motherboard and safely operate a sub-zero cooler (not to mention the cost and noise) are rarely going to be perceived as worth investing in.

Related Articles