What better data than what the manufacturers claim. Apple claims 11 TOPS, Qualcomm claims 26. As simple as that. They both claimed the most easiest operations per second. But things go hand in hand. ALU Vs ALU, not ALU vs FPOU
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Snapdragon 888 @ 26TOPS vs 865 @ 16 TOPS vs A14 Bionic @ 11 TOPS
- Thread starter Next Stop
- Start date
- Sort by reaction score
- Status
What type of operation is Apple claiming, what type Qualcomm? Please provide detail. TOPS alone doesn't mean anything. How wide is Apple's decode block, how wide Qualcomm? It's like carrying water in buckets. If you have to move 100 gallons of water from A to B with two buckets, only one bucket is 1 gallon, the other is 5 gallons. How often do you have to run from A to B?
Man, it's OPERATIONS. I don't think Apple would have missed claiming bigger numbers if it was possible. It's the most easiest operations for both of chips.
It seems like you can't live without chasing me everywhere I post something. Relax, I'm not coming to the iPhone forum, I'll stick to the alternative one. And I think your exhaustive knowledge would better help those with issues in the iPhone section.
You still do not understand how this works. There's no such thing as "easiest" operations. There is a huge difference in datatypes that affects how many operations you can run and how many instructions you can execute.
Simple example. You have one second to transport as much water as possible from A to B. Qualcomm employee #888 runs 26 times from A to B (seems like that guy is really fast) with a 1 gallon bucket and transports 26 gallons of water in one second. Qualcomm employee #865 runs 16 times from A to B with a 1 gallon bucket and manages 16 gallons in one second. Apple employee #14 runs 11 times from A to B (what a lazy guy...) with a 5 gallon bucket and manages... oops, 55 gallons of water, despite he's so "slow".
Simple example. You have one second to transport as much water as possible from A to B. Qualcomm employee #888 runs 26 times from A to B (seems like that guy is really fast) with a 1 gallon bucket and transports 26 gallons of water in one second. Qualcomm employee #865 runs 16 times from A to B with a 1 gallon bucket and manages 16 gallons in one second. Apple employee #14 runs 11 times from A to B (what a lazy guy...) with a 5 gallon bucket and manages... oops, 55 gallons of water, despite he's so "slow".
Might be worth starting to read here: https://www.kdnuggets.com/2020/05/tops-just-hype-dark-ai-silicon-disguise.html
108/12 = 9 so a 108mp sensor is obviously 9 times better than a 12mp sensor.
On a serious note, seems like my previous post got deleted for "name-calling", so here it is again but slightly modified:
OP is comparing TOPS from the Hexagon 780 with TOPS from the NE.
But the questions we have to ask are: What operations? Floating point? Integer? What happens in one operation? Do the Hexagon 780 and NE even do the same stuff?
He doesn't know, and neither do we to be honest.
What we do know however is that you can't compare TOPS from company to company or product to product.
Interesting reads:
Are Tera Operations Per Second (TOPS) Just hype? Or Dark AI Silicon in Disguise? - KDnuggets
This article explains why TOPS isn’t as accurate a gauge as many people think, and discusses other criteria that should be considered when evaluating a solution to a real application.
As AI chips improve, is TOPS the best way to measure their power?
There's more to understanding the performance of AI-capable chips than just a single summary number, even if that's what chipmakers are marketing to users.
Not all TOPs are created equal
Deep Learning processor companies often highlight their products’ blazing-fast speeds in terms of metrics such as Tera Operations per…
Arm chips doesn't transport water. They handle operations, microcode. Risc code. Otherwise it will have been cisc.
Man, always, but always when you say floating point and count it in Terra, it will be Terra Flops or TFLOPS. We are talking about Terra Ops or TOPS.108/12 = 9 so a 108mp sensor is obviously 9 times better than a 12mp sensor.
On a serious note, seems like my previous post got deleted for "name-calling", so here it is again but slightly modified:
OP is comparing TOPS from the Hexagon 780 with TOPS from the NE.
But the questions we have to ask are: What operations? Floating point? Integer? What happens in one operation? Do the Hexagon 780 and NE even do the same stuff?
He doesn't know, and neither do we to be honest.
What we do know however is that you can't compare TOPS from company to company or product to product.
Interesting reads:
Are Tera Operations Per Second (TOPS) Just hype? Or Dark AI Silicon in Disguise? - KDnuggets
This article explains why TOPS isn’t as accurate a gauge as many people think, and discusses other criteria that should be considered when evaluating a solution to a real application.www.kdnuggets.com
As AI chips improve, is TOPS the best way to measure their power?
There's more to understanding the performance of AI-capable chips than just a single summary number, even if that's what chipmakers are marketing to users.venturebeat.com
Not all TOPs are created equal
Deep Learning processor companies often highlight their products’ blazing-fast speeds in terms of metrics such as Tera Operations per…medium.com
Uhm, no. Also, how wide are they? 4-wide, 8-wide? How much data? Get it? 1 gallon, 5 gallon... You need to start to understand how this works when making such claims.
Uhm, no. I've pointed that out to your old account already. You keep mixing up things, because you don't understand the basics. I'm going to ask again, when was the last time you've seen an inference network work with integers?
Lesson 2 (from someone who's actually teaching this stuff on Bachelor/Master university level 
): https://www.quora.com/Does-NVIDIA-TOPS-mean-TFLOPS
): https://www.quora.com/Does-NVIDIA-TOPS-mean-TFLOPS
Just pure speculation with baseless claim without any pure benchmark. You can’t compare future Snapdragon 888 releasing next year possible in Xiaomi mi11 against Apple current A14 when you don’t have any idea of what A15 will bring. No test. No benchmark. No comparison. Baseless claim based on TOPS. No real-world speed test. Hope you’ll justify your claim in 2021 when they are released.
Here's a more realistic analogy to your water transportation :
Subwoofer 1 (Qualcomm) 26 Watt peak Power - 16 Watt RMS
Subwoofer 2 (Apple) 11 Watt peak power - 8 Watt RMS.
Simple as that. Apple was also talking about basic integer operations. Why don't you ask them what do detail they're operations? You're asking me about their's claims? ?
Btw, you shot yourself in the foot here. That's probably the worst you can do. With matrices used in neural networks, you want operations on tensors. Look up what Nvidia did ages ago and see how much more performant they are.
That's 2 moths after the iPhone releases. ? Not next millennium. No, they're to lazy and greedy.
Man, you're the winner here. Apple pushes wider operations than Qualcomm because you say so. Happy now? It's not like that who's heavier 17 tons of steel or 17 tons of feathers. It's like you, because you know that Apple is using wider operations.Lesson 2 (from someone who's actually teaching this stuff on Bachelor/Master university level
Nope, not as simple as that. Except for constantly showing off that you don't understand math and computer science, you also don't understand electronics. When making power claims about subwoofers or speakers in general, I'm asking again, what sensitivity do subwoofers 1 and 2 have? The actual power is irrelevant, you have to specify sensitivity. That's why serious manufacturers always specify sensitivity on subs/speakers, see here for an example: https://www.wisdomaudio.com/products/sts/
If subwoofer 2 has a much higher sensitivity (a larger bucket so to say), then it can play much louder than subwoofer 1 with a lower sensitivity. There's more involved of course, but you really have to understand the simple stuff to understand the more advanced things. I'll give you a link to an SPL calculators for speakers, so you can play around. Going with the actual math/physics behind it would be wasted, as you obviously don't even try to understand things.
Here's the link: https://myhometheater.homestead.com/splcalculator.html
Because I say so? How about everyone says so? https://www.anandtech.com/show/16226/apple-silicon-m1-a14-deep-dive/2
Apple announced A14 which released in new iPhone in October.
Snapdragon announced 888 yesterday which will release next year.
Snapdragon announce their next gen ending of the year while Apple mid of the year. Different release cycle and it should be compared within the year it’s released. Regardless, still speculation on an unreleased product.
- Status