Maybe you explain us, Mr. Professor, how about the memory bandwidth of the CPU, we all know the bottleneck is there. Or how many 8 wide alus are there. 2 8 wide alus isn't the same thing as 4 4 wide alus? What was the memory bandwidth?
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
In general, yes. But even then it doesn't mean much without specifying the metric "TOPS". This metric has to be identical, then one can make direct comparisons. This is best done with identical benchmarks running on both systems. Geekbench does that. For the conspiracy fanatics, this document describes what Geekbench is doing exactly: https://www.geekbench.com/doc/geekbench5-cpu-workloads.pdf
You can easily download each implementation individually and compile on your own, that way you won't have to use Geekbench and still get the same result.
We don't know the generation of score in geekbench because it's not open source. Maybe you should compile an app for both platforms. But hey, compile it to measure what? Only ALU, only fpu engines? Only ai engine? Overall? What scores will you give to ALU, what to fpu, what to ai engines? You see that ALU does an better score on Apple, you give 85% of the ALU engine to the total score of the benchmark? Is that fair?
What about the memory bandwidth? Also, are we talking SoC or CPU? This whole thread is about compute performance, now we're bringing in memory bandwith which is something completely different for larger data chunks and will introduce additional dependencies, the OS itself, so we're not talking chip level anymore now?
Also, what the heck are you talking about with 8 wide ALUs and 4 wide ALUs? Have you actually read the article I linked to? I'm going to quote:
I'm simply going to assume you do not understand this. I can't explain this in a single sentence. Books are written about it, lectures are taught. Here's a starter: https://www.robots.ox.ac.uk/~dwm/Courses/2CO_2014/2CO-N2.pdf
You'll probably need additional sources to fully understand it.
Because they have optimized (changed algorithms/frameworks) for 64-bit instead of supporting algorithms for 32-bit systems. You can see this by looking at CPU workload for version 4 and version 5 and following to the actual implementations used:
https://www.geekbench.com/doc/geekbench5-cpu-workloads.pdf
https://www.geekbench.com/doc/geekbench4-cpu-workloads.pdf
https://www.macworld.com/article/34...ased-with-all-new-tests-modes-and-scores.html
Regardless of what's being discussed, talking like is very telling. And so the ignore list grows.
Mr. Professor, you seem to be stuck at some utopia love for cpu architecture. We're looking at an overall performance, not how much work can an cpu do n an ALU or FPU. Did it go to your head that maybe 1 strong ALU can't beat 3 weaker ALUS? Did you know what AMD did to rezolve this? Many alus, many cpu cores etc. Leave behind your utopic love and look at the overall picture. Add up all ALU engines, all FPU engines, all cpu cores, pipeline, instructions per clock, clock speed add them all make the total and look at the final score. Then teach us.
It doesn't work like this as many have pointed out to you on this account and your old one that is now deactivated. You're absolutely not making any sense at all in what you say. Adding up things, does not work. That's not how CPUs and SoCs work at all. In order to understand this, you need a very basic understanding of the technology. Start with the links that have been given to you and work your way up from there. So please, read and understand the numerous sources.
Once you have done that, you will (hopefully) have gained the knowledge that 5 TOPS from manufacturer A are not necessarily the same as 5 TOPS from manufacturer B and C and D and so on (again, you have been given links for this). Only if the metric for TOPS is identical across manufacturers (again, it is not, you have been given several links for this), you can actually make comparisons. The best you get for this are benchmarking suits, to run on both devices. For example Geekbench for out of the box benchmarks or MLPerf and numerous others for those who compile themselves and run them. But others and myself have been through this step by step before with your old account in older discussions, so you sound a little like a broken record, repeating false statements over and over again. Just because you create new accounts, doesn't mean what you say actually becomes true.
So you're telling us that the overall soc performance doesn't work by adding things like number of engines ALU, fpu, etc, core count, clock speed. Then how it works, Mr Broken Record? It works like in geekbench which is not open source and we don't know how they generate the final score, what percentage of that score goes to Aly, what to fpu, what to ai and so on? Heck we don't even know if they take advantage of the full instruction set or being optimized for the Qualcomm architecture. That's how it goes Mr profesor? 😂 Go and teach your iPhone students, not us.
I would suggest you learn the terminology first, there’s so much wrong with what you say it’s unbelievable. What the hell is an ”engines ALU“ btw? Again links have been given. And actually, we know exactly what Geekbench does, but again, links and so on.
Thanks for letting me know what to teach to who. Big bummer though on the iPhone teaching. I do most of my teaching with Intel x86, Nvidia GPUs, embedded systems and ARM. The former two on single CPU/GPU systems as well as clusters all running Linux.
Engines: ALU, fpu, ai etc.
You do realize that my stubbornness to respond you on each reply is to see how far you can go with your obsession?
Move on, snapdragon SMOKES the Bionic. Even iphone forum section is full of people who says theirs A13 or a14 can't match a 865 in heavy games like genshin impact, call of duty etc. This geekbench thing is so smoked man.
Thanks for proving my point, you don’t understand it. Why you have to prove it over and over again is beyond me though. Some people like education, gaining knowledge, others apparently don’t.
Wouldn’t call it “stubbornness” at all, but it starts with “stu“ as well.
Im not obsessed, but thanks. When people without any knowledge show up and make false claims to spread propaganda I speak up. That’s what hopefully makes the world a better place. See what happened in the 1930 in Germany when people showed up making idiotic claims and not enough people stood up. See what happened four years ago in US when a moron showed up to make false claims as fact and not enough people stood up. Luckily this time around enough people did stand up and voted him out. Or in other words, when people show up without knowledge, making false claims, even if it’s just on technology, hopefully enough people will stand up to correct them. I have a very strong believe in science and education, that’s what drives mankind further, spreading nonsense isn’t.
- Status