notebookcheck made an article about
This one???
Astounding M1X multi-core performance estimates have the Apple Silicon playing in the same ballpark as the Intel Core i9-10900K and Ryzen 7 5800X
Estimated multi-core performance results for the expected M1X processor have the upcoming Apple Silicon taking on power-hungry desktop rivals such as the Intel Core i9-10900K and AMD Ryzen 7 5800X. Speculated Geekbench 5 and Cinebench R23 scores for the Apple M1X chip are based on it offering...
www.notebookcheck.net
From article.
"... result would actually position the Apple M1X just -4.40% behind the 12-core Ryzen 9 5900X. Miani stresses that he does not believe the Apple Silicon will be ready to take on such a Ryzen 9 competitor just yet ..."
The article has an estimated score for the M1X as being GBv5 Multicore of 13586. The GBv5 Multicore score for the Xeon W-3245 is around 14261
Intel Xeon W-3245 - Benchmark, Test and Specs
Intel Xeon W-3245 - Benchmark, Geekbench 5, Cinebench R20, Cinebench R23, Cinebench R15 and FP32 iGPU (GFLOPS) benchmark results
www.cpu-monkey.com
So the estimate is approximately -5% off of the W-3245 mark. ( and probably even further behind a W-3335 (16 core Ice Lake) ). Given that the latest 12 core Ryzen is ahead also, that is has a good chance of being right.
The 12 core ( W-3235 ) is 12996 (+5% to 'M1X") . So the 8 and 12 core would probably get beat. However, this "16 core beat" meme is likely an overreach. ( and a substantially even bigger over reach for the W-3333 at higher clock , higher IPC , higher bandwidth , and more optimized SIMD foundation. ). I suspect there is a bit of "telephone game" here where '12' is being mutated into '16' as the story is passed around. Or there are folks throwing in a "2x" mulitplier as their SWAG guess multiplier; which this article doesn't do.
The leaks so far is that Apple is putting more "weight" on bumping up GPU cores than CPU cores. There is a window where if not asked to walk and chew gum at the same time ( just do CPU and almost zero GPU work) that it will be in the same ballpark. But on tasks where the CPU cores are going to have to share with bandwidth hungry GPU/NPU/etc cores, it probably won't pan out as well.
This "1.77" estimate also presumes four E cores which decent chance not going to get. Apple swapping two Ps for two Es wouldn't be a full "addition" to the mix . Get to 8 P but just 2 E; so only 2 "full" adds. [ Again Apple having higher priority on taking out dGPUs and allocating more die space to non CPU cores. This showed up in A15 . Additional GPU core and elsewhere in the SoC, but not tons of movement on the P cores. ] There isn't a good foundation for the 2x ( or bigger) multiplier SWAG with Apple in part doing a "swap".
P.S. Apple more than doubling up on the GPU add rate while keeping the memory channel add rate still at 2x , puts the 3070 performance claims to doubt. Once Apple gets into the zone where adding GPU cores faster than bandwidth then the performance estimates are get more "hand wavy". More cache can help in some context but won't in others. Once highly oversubscribe the memory bus performance tends to go sublinear ( less slope of '1' ).
Last edited: