Leaked Benchmarks Paint Conflicting Picture of Intel’s Rocket Lake
There’s been a flurry of leaks around Intel’s upcoming Rocket Lake desktop CPUs, and it points in opposite directions as far as what to expect from the CPU.
Most recently, there are some SiSoft Sandra results via THG that show an 8C/16T Rocket Lake chip running below the performance of a Ryzen 3 3300X with a claimed 1.8GHz base clock and a 4.4GHz boost clock. Last week, data from Ashes of the Singularity and Geekbench implied that the new CPU was an approximate match for the Ryzen 7 5800X. Both can’t be true, but the higher, better figures are more likely to be accurate. The more recent results may reflect an early CPU, lower-than-reported clocks, a low-watt TDP part, or some other attribute that would impact the relative comparison.
Rocket Lake uses Intel’s Cypress Cove architecture on a 14nm process. Cypress Cove is basically Intel’s Sunny Cove (Ice Lake) CPU core, backported to 14nm. Sunny Cove, if you recall, was a genuine IPC uplift for Intel, but it traded back almost as much frequency as it gained in IPC. As a result, CPU performance in many applications was flat on a core-for-core or clock-for-clock basis.
The problems that forced lower clocks on Intel’s 10nm node with Ice Lake, however, don’t exist on the older 14nm node — and that, in turn, suggests that Intel should be able to deliver a net CPU performance improvement with this shift. At the same time, higher IPC typically requires more power consumption at a given frequency, and Intel cut the total number of CPU cores on this chip to eight, down from 10. We should also expect to see signs of higher per-core power consumption.
This image, from WCCFTech, summarizes some GeekBench 5 performance results from last week. Unlike the Sandra results, this data looks exactly like what we’d predict.
The Core i9-10900K is 17 percent slower in single-core testing, implying that Rocket Lake keeps all of Comet Lake’s clock at 14nm and fully benefits from its IPC uplift. On paper, the Core i9-10900K’s 10 cores should be 25 percent faster than the unnamed RKL CPU’s eight, but in practice, it’s only about 1.12x faster. As we theorized, higher IPC from the RKL CPU helps to compensate for the reduced number of cores.
The Core i7-10700K’s relative performance is also interesting, especially if you consider them compared with, say, AMD’s improvements from the 3950X to the 5950X. In AMD’s case, the 5950X gains more performance against the 3950X in 1T mode than it does at full load, partly because both CPUs bump up against the power limits baked into all AM4 motherboards. The fact that the RKL CPU can beat the Core i7-10700K by 22 percent in 1T but only by 9 percent in multi-threading suggests that the clock gap between the two cores is larger in one mode than the other.
These results also fit with what we know about Intel’s product plans and strategies these past few years. The company has been emphasizing single-threaded burst performance and short-duration workloads, and leaning on categories like gaming, which it still won more often than it lost until the launch of AMD’s Ryzen 5000 series a few months ago.
Incidentally, this data also suggests rough parity between RKL and the Ryzen 5000 series on a core for core basis. AMD’s desktop platform would still offer more total cores than Intel, allowing for higher maximum performance in a single socket, but early data suggests Intel will still be in a much better position when the new chips debut at the end of March. With core-per-core parity, Intel could even take an overall price/performance leadership position in certain match-ups, should it choose to price to do so.
As always, take all rumors, leaks, and equivalencies with a grain of salt.