4 ms·
It's underappreciated how much thermal capacity varies from machine to machine, and over time. It makes benchmarking hard. Clock speeds, because they're crysta
by tlb 8y ago
It's underappreciated how much thermal capacity varies from machine to machine, and over time. It makes benchmarking hard.
Clock speeds, because they're crystal-controlled, don't vary by more than several PPM between machines. Thermal capacity varies by ±10% easily.
The thermal resistance depends on a bunch of mechanical factors that are hard to control in manufacturing, like how well the thermal grease fills the tiny pores in the metal of the heat sinks, and how well the air flows through the ventilation ducts. And these usually get worse over time as bubbles form in the grease and dust sticks to the fan blades. And of course it depends on room temperature and pressure.
It means that benchmarks will have to be done under more tightly controlled conditions. The automotive world has standards for this, where horsepower measurements are adjusted by a correction factor that depends on temperature and pressure.
- jdietrich 8y agoThose are all reasons why you should engineer in a thermal margin. If some proportion of machines are throttling out of the box, then most machines will start throttling as the thermal management system degrades. You're supposed to account for the ageing of the thermal material and clogging of the airflow path and degradation of the fan bearings. You're supposed to leave in some headroom for warm rooms, high altitude and partially blocked vents. It's yet another symptom of Apple's preoccupation with thickness. The 2015 MacBook Pro was 18mm thick. The new MacBook Pro is 15mm thick. After all the furore over USB-C and unreliable keyboards and the loss of MagSafe, was that 3mm of z-height a worthwhile tradeoff?
- tlb 8y agoI don't think there's room for margin any more. Margin means fixing a power limit slightly lower than what the worst machine can sustain. That probably means leaving 10-20% performance on the table for average machines.
- stefan_ 8y agoNo, you don't get to blame this on the processor. Modern processors are perfectly capable of operating exactly at the limit of the cooling system they are provided. No worries there. There is no "setpoint" you need to set; the processor ramps up, then sustains. This is purely a mechanical engineering issue. They haven't designed it properly.
- tlb 8y agoChips vary considerably in the amount of power they use at a given frequency, as a result of silicon process variations. Consider table 5 in http://charm.cs.illinois.edu/newPapers/16-08/paper.pdf http://charm.cs.illinois.edu/newPapers/16-08/paper.pdf. So even with identical cooling systems, some chips will outperform others.
- EpicEng 8y agoYet that's all rather irrelevant im context considering that apple controls every aspect of their machines.
- yoz-y 8y agoBenchmarks such as geek bench are an aggregate of benchmarks done in many different conditions.
- glup 8y agoI understand this variability in thermal efficiency in who-knows-who-made-this-but-it-sure-was-cheap OEM ecosystems, but shouldn't Apple of all companies have the engineering/sourcing/testing resources to get this down to +-1%?
- noir_lord 8y agoI mentioned the other day on here that my T470P doesn't throttle under load and someone who didn't own one told me it did based on an online review. I mean I only own one and have watched it's clock speed while under load from a chess engine..what do I know.
- microdrum 8y agoThermal capacity varies by 10% between different examples of the same Apple product?
- jagger27 8y agoThat's the first I've heard of such a claim. What's the difference? Thermal paste?
- falcolas 8y agoCrystal controlled clock speeds are thermally dependent too (upwards of 40ppm) - they're just compensated for via dedicated circuits to bring that down to the 2-3ppm range. Which raises the question - are there 'el-cheepo' motherboards out there which don't use thermally compensated oscillators?
- redial 8y agoAll those problems exist in smartphones and maybe manifest even more as you literally hold the heat dissipator in your hands while using it, yet Apple manages just fine. Thermal engineering is hard, but doable, most of all to the richest technology company in the world.
- Const-me 8y agoThat i9 CPU dissipates 45 watts of heat, on average: https://ark.intel.com/products/134903/Intel-Core-i9-8950HK-Processor-12M-Cache-up-to-4_80-GHz https://ark.intel.com/products/134903/Intel-Core-i9-8950HK-P... I don’t think it’s possible to design an ultrathin laptop around that CPU. That’s just too much heat. Many desktop CPUs are close to that, e.g. AMD’s Ryzen 5 2400G can be configured to be 45W too.
- EpicEng 8y ago>I don’t think it’s possible to design an ultrathin laptop around that CPU. Maybe not, but I don't care about "ultra thin" for a professional machine. I want performance. Apple keeps making sacrifices in the name of dimensions and weight. They seem to be fine selling "pro" machines to people who use it as little more than a web browser. They make a ton of money and I'm sure they know what they're doing, but they lost me as a customer.
- Const-me 8y agoThey never had me as their customer, I use Windows, but the trend is the same over the whole market. A thick laptop from 5 years ago, with a 35W CPU and cooling system that can work under heavy load for long time: https://support.hp.com/lv-en/document/c03331546 https://support.hp.com/lv-en/document/c03331546 Same lineup, 5 years later: https://store.hp.com/us/en/pdp/hp-probook-640-g4-notebook-pc https://store.hp.com/us/en/pdp/hp-probook-640-g4-notebook-pc 5 generations difference, but because the TDP is now just 15W, the performance is very close: https://www.cpubenchmark.net/cpu.php?cpu=Intel+Core+i7-3612QM+%40+2.10GHz https://www.cpubenchmark.net/cpu.php?cpu=Intel+Core+i7-3612Q... https://www.cpubenchmark.net/cpu.php?cpu=Intel+Core+i7-8550U+%40+1.80GHz https://www.cpubenchmark.net/cpu.php?cpu=Intel+Core+i7-8550U...