Geekbench vs Real-World performance

[lockerc18]

Geekbench is the system benchmark that people salivate over. But I wonder how effective it is in representing real-world system performance for the kinds of workloads that people actually use.

Geekbench measures things like CPU performance, memory access, and GPU performance. But, while these are important measurements, aren't there other tests which represent actual system usage with real world applications? Of course there are, by the bazillion. Every application vendor has their own suite of them. But which one is the one which is the best at representing what the real world user experience is for a new device like the 10.5 iPad? That would necessarily include things like transactions and testing of other capabilities which would extend beyond just the local platform. Networking and database come to mind.

I know system performance measurement is often black magic. I've been involved in various testing efforts using things like TPC-D. That's a few levels above the iPad. But I'd rather to have a measurement of normal usages that typical users would do, than just a very granular benchmark that measures specific system component performance.

The point is that while the CPU/GPU are important to system performance, there's a lot more to the user experience than just those, and it would be useful to know how this experience changes from one device version to the next.

 

[TheRealAlex]

Geekbench is like car horsepower which is stated on paper 250hp but in reality with crankcase loss and when you finally put down power at the wheels it's more like 220hp.

That being said, since there nothing being held back from the A9X Pro iPads I'd say the only measurable difference is what I call First World Problems.

Youtubers will race to do the first side by side loading tests showing that the 10.5" Pro loads things 0.5 seconds faster than 12.9" Pro from 2016

Or look how much faster CNN.com loaded 0.2 seconds.

 

[burgman]

Fanless chips have to throttle quicker than fan models. I have read that GeekBench and other test suites only run for 15 to 20 seconds so in real world we need to see more testing like in intensive gameplay. I agree it's more about marketing and ubergeek obsession at this point. For me I want to see USB bus speed on the 12.9. If iPhone 8 switches to USB 3 that would be a real disappointment.

 

[rui no onna]

For gaming, GFXBench is a good benchmark. In particular, look at the on-screen performance test.

I think more often than not, iOS stutters/lag owe more to graphics and storage (random small block read/write) rather than general processing performance. My personal "usability" ranking of iOS devices look awfully similar to the rankings on GFXBench Manhattan 3.0 On-screen.

Internet stuff, mostly networking components and ISP.

If you want an extensive review, there's AnandTech which is excellent but often come out several months after release. They do extensive testing of storage including small 4K read/write. Also, they seem to test for sustained sequential read/write and not just quick bursts (which can result in super high numbers due to caching).

http://www.anandtech.com/show/10685/the-iphone-7-and-iphone-7-plus-review/4

http://www.anandtech.com/show/10286/the-97-ipad-pro-review/2

 

[gobikerider]

The way I look at Geek bench these days is how much overhead does a product head have at launch because I know in reality real world usage isn't going to be insanely different unless your say exporting a video, or maybe stitching photos in Affinity. I also like its compute score as its a simple way to gauge the horsepower of a gpu which usually is indicative of real world performance being the way gpu are used in application.

 

[rui no onna]

The way I look at Geek bench these days is how much overhead does a product head have at launch because I know in reality real world usage isn't going to be insanely different unless your say exporting a video, or maybe stitching photos in Affinity. I also like its compute score as its a simple way to gauge the horsepower of a gpu which usually is indicative of real world performance being the way gpu are used in application.

Nah. Compute scores don't mean much for gaming. Those are just raw calculations. Plenty of graphics benchmarks available that more closely approximate real world performance.

The iPhone 5s is chugging along quite nicely while the Air has very noticeable lag despite both running A7 chips. Why? iPhone 5s is only 1136*640 or 727,040 pixels. iPad Air is 2048*1536 or 3,145,728 pixels. Resolution matters a lot.

 

[gobikerider]

Nah. Compute scores don't mean much for gaming. Those are just raw calculations. Plenty of graphics benchmarks available that more closely approximate real world performance.

The iPhone 5s is chugging along quite nicely while the Air has very noticeable lag despite both running A7 chips. Why? iPhone 5s is only 1136*640 or 727,040 pixels. iPad Air is 2048*1536 or 3,145,728 pixels. Resolution matters a lot.

I know this I meant indicative of performance outside of games, like photo editing, and iMovie rendering type things. I have one simple game on my iPad so yah I'm the last person focused on games.