Intel’s Rocket Lake is expected to deliver a 20 percent IPC improvement, but new rumors suggest the top-end desktop version of the chip will still draw a lot of power at peak load. The CPU reportedly runs up to 98C with a sustained power draw of ~250W.
According to reports from Chiphell, the Core i9-11900KF was tested with an entry-level 360mm cooler while running AIDA64’s FPU stress test. One of the reasons you should take these rumors with a grain of salt is that AIDA64’s FPU stress test is specifically designed to hammer the chip. This is great if your goal is to stress test equipment and isolate errors. It’s less great if your goal is to measure normal workload power consumption. AIDA64’s FPU test also contains AVX-512 instructions. If it’s using that code path for the Core i9-11900KF while the CPU is clocked at 4.8GHz, it’s going to generate a lot of heat.
Allowing the CPU to run as hot as 98C is less than ideal from an enthusiast’s perspective, but it also may demonstrate Intel is comfortable with allowing its silicon to run at this temperature. Years ago I had a conversation with an Intel engineer, who told me the company expected its then-current Nehalem and Westmere CPUs to run perfectly at 95C in 24/7 operation, and that the chips should be able to handle these kinds of temperatures for up to a decade, at minimum. This does not mean enthusiasts should chuck out their expensive coolers and let their chips bake under a $15 heatsink+fan, but the fact that the CPU can run up to 98C isn’t automatically a problem.
It’s not surprising (assuming this rumor is even accurate) to see the 11900KF running this hot. As we’ve recently discussed, motherboard OEMs often program enthusiast boards with system defaults that run the CPU at full boost clock for a much longer period of time than Intel specifies. Intel is aware of this and its own guidelines are recommendations, not requirements. Typically we would expect a CPU running AVX-512 code to down-clock itself. The fact that this hasn’t happened could mean Rocket Lake holds its clocks in AVX-512 workloads where other Intel CPUs step down. It could also indicate that the motherboard is configured not to drop the multiplier during AVX-512 when executing AVX-512 code. The fact that this chip hit 98C in an unknown configuration doesn’t mean this will be normal behavior for every CPU.
Intel has chosen to push Rocket Lake as far as it can while remaining backward compatible with Comet Lake, which includes defined TDP ranges. In this way, it’s not much different than AMD’s decision to keep the Ryzen 5000 series’s maximum TDPs identical to their Ryzen 3000 counterparts. Intel can still claim a power efficiency gain, so long as it completes workloads in less time than 10th Gen chips within the same power envelope.
Rocket Lake is expected to deliver strong performance gains in single-threaded workloads and a somewhat smaller uplift in multi-threaded workloads. Comet Lake topped out with 10 cores while Rocket Lake only offers eight, but the IPC gains are expected to mostly or entirely compensate. AVX-512 support may even give the eight-core Core i9-11900K a few wins over the 10-core Comet Lake, though this will depend on benchmark support.
We’ll know more when Rocket Lake launches at the end of March.