Intel’s 10nm Ice Lake-SP Xeon CPU offers alleged 100% core-for-core performance boost
The first benchmarks have leaked from Intel’s upcoming Whitley Lake 10nm server CPUs, revealing potential 100% performance improvements core-for-core when compared to Intel’s 14nm chipsets.
Intel’s new Xeon processors are expected to ship later this year. Designed around the 10th Generation Ice Lake architecture, the Ice Lake-SP Xeon server chips will be poised to head-to-head against AMD’s upcoming Zen 3-power EPYC Milan server processors.
Ice Lake-SP Xeno processors will utilise Intel’s Whitley chipset (succeeding Purley) and will require Socket P+ in 2S configuration. The first benchmarks to come from Ice Lake SP are taken from early engineering samples and so may differ from the final release version but, most importantly, at double the speed this is absolutely blowing Intel’s current server CPUs out of the water.
The leak in question comes courtesy of a Geekbench benchmark showcasing the first Whitley processor scoring 3,427 single-core score and a 27,926 multi-core score. This is a fairly low-end 12C/24T server processor which, when compared to previous-gen Purley, is basically double the core-for-core speed. Intel’s Cascade Lake Xeon W-3235 12C/24T CPU, for example, scores just 1,084 single-core score and 11,987 multi-core score.
Being a server chip, the unannounced Ice Lake-SP Xeon has low clock speeds and therefore fairly poor single-core performance. When it comes to multi-core, however, it picks up serious points and is indicative of strong gains from Intel heading into its 10nm lineup.
However, there is some confusion to be found, due to Geekbench’s reading of Intel’s setup, over whether we’re looking at 12 or 24 Cores. WCCFTech, who discovered this original benchmark, seem to be pretty sure this is just a result of a mistake by Geekbench, although we’re not so sure. Considering this is an engineering sample, a performance uplift of upwards of 100% seems a little ludicrous. Still, the single-core performance does evidently bear fruit of this, so perhaps it’s possible.
As it currently stands, Intel only has low-power 10nm laptop chips to its name, so this move into server products would present a big uptick in Intel’s 10nm performance, if it is to believed.
What do you think then, has Intel squeezed out a hugely impressive performance boost from adopting 10nm tech? Or do you think there are some benchmark hijinks at play? Let us know!