3 ms·
Looks like it’s on par[1] with a top of line Core i9-13900KS but at a fraction of the wattage. I’d never buy one, but I’d love to see Apple do a Mac Pro build
by runjake 3y ago
Looks like it’s on par[1] with a top of line Core i9-13900KS but at a fraction of the wattage.
I’d never buy one, but I’d love to see Apple do a Mac Pro build at a watt/heat equivalent Apple Silicon build.
1. https://browser.geekbench.com/processor-benchmarks https://browser.geekbench.com/processor-benchmarks
- duskwuff 3y agoAnd OpenCL ("GeekBench 6 Compute") scores comparable with an RTX 2070: https://browser.geekbench.com/opencl-benchmarks https://browser.geekbench.com/opencl-benchmarks
- ex3ndr 3y agoThat's not bad actually because bottleneck in training is in RAM and we can have 128gb...
- thangngoc89 3y agoI believe that Apple Silicone is still too slow for training. The big RAM is really useful for interference of big models though
- blovescoffee 3y agoIf you’re training a model that needs 128gb, the GPU speed is going to be a “bottleneck”
- skavi 3y agodoing a few sketchy conversions puts Metal compute performance closer in capability to an RTX 4070. M3 Max is ~170% M1 Max per some graphic Apple released. That puts it at ~95% the performance of a 6800 XT on the Metal charts. RTX 4070 is ~95% the performance of the 6800 XT on the Vulkan charts.
- Sparkyte 3y agoIntel hasn't even been on point for a while.
- tehnub 3y agoDoes this mean it's actually as fast in real-world scenarios? I'm not that familiar with Geekbench
- runeks 3y agoDirect comparison: https://browser.geekbench.com/v6/cpu/compare/3443230?baseline=3442661 https://browser.geekbench.com/v6/cpu/compare/3443230?baselin...