Disqus - Latest Comments for garysims

Re: TSMC’s 1.4nm Chips Named A14 Could See Daylight in 2027 After Company Introduces 2nm Chips

garysims — Fri, 15 Dec 2023 00:01:06 -0000

It isn't a chip, it is a process node technology. You should remove this article it is completely misleading and wrong. A14 means 14 Angstrom, rather than N2 which is 2nm. In other words 1.4nm.

Re: Which is better 1 core @3Ghz or 2 cores @1.5 GHz?

garysims — Tue, 21 Jan 2020 06:11:39 -0000

So you would agree that if developers do a better job at creating multi-threaded applications then lowering the frequency and increasing the core count could benefit from lower power consumption while still offering performance possibilities?

Re: Which is better 1 core @3Ghz or 2 cores @1.5 GHz?

garysims — Tue, 21 Jan 2020 05:33:09 -0000

I understand the distinction you are making between add/lower and add/keep. But the point, I am clearly failing to make, is this. You said (I repeat) that multi cores is "only valid for a very limited set of tasks. Most of them not relevant to mobile phones." This means that in the A12X single-core performance remains the same, but so does overall general performance since those extra two cores are only good for "for a very limited set of tasks." In other words the A12X brings nothing new and just maintains the single-core performance. Since you are saying that multi-core performance is for only very small subset of tasks. This means the A12X is a marketing trick in your view. Hence my first comment, "So you don't think Apple should have made the A12X?"

Re: Which is better 1 core @3Ghz or 2 cores @1.5 GHz?

garysims — Tue, 21 Jan 2020 01:14:58 -0000

In your first comment you wrote, "My point is that this is only valid for a very limited set of tasks. Most of them not relevant to mobile phones." If that is true, then Apple adding two more cores to the A12X is just marketing, and only has value for a limited set of tasks. It pushes up the price but achieves nothing. According to your first comment.

Re: Which is better 1 core @3Ghz or 2 cores @1.5 GHz?

garysims — Mon, 20 Jan 2020 09:50:39 -0000

I was referring to "The reason Apple has their design of one or two bigger faster cores"

Re: Which is better 1 core @3Ghz or 2 cores @1.5 GHz?

garysims — Mon, 20 Jan 2020 07:49:12 -0000

So you don't think Apple should have made the A12X?

Re: Which is better 1 core @3Ghz or 2 cores @1.5 GHz?

garysims — Sat, 18 Jan 2020 14:55:38 -0000

I wrote it in pure C using pthreads, not Python or any variations of Python. The test program does not use or need any locks.

Re: Finally! Snapdragon 865’s CPU Beats the Apple A13

garysims — Tue, 17 Dec 2019 06:24:52 -0000

And then what about A14 vs SD875...

Re: Finally! Snapdragon 865’s CPU Beats the Apple A13

garysims — Mon, 16 Dec 2019 13:17:54 -0000

Nobody trust Speed Test G? Really? Why do you trust Geekbench more than Speed Test G.

Re: Finally! Snapdragon 865’s CPU Beats the Apple A13

garysims — Mon, 16 Dec 2019 10:51:02 -0000

What magically happens when you go from Dec 31 to Jan 1?

Re: Snapdragon 855 performance and benchmarking

garysims — Thu, 17 Jan 2019 05:03:18 -0000

I used the official scores from the Geekbench website.

Re: How does Apple’s new A12X processor compare to other 7nm mobile SoCs?

garysims — Wed, 14 Nov 2018 10:03:25 -0000

Ah, I see your misunderstanding. No it isn't that 90% of 1 test is the same across platforms and then 10% is different. If that was so, then what you described would happen. No. Of the whole collection of tests, as separate individual units, most are exactly the same, line by line, with exactly thee same code.However one test, one standalone test, not part of a library or subsystem, but one test, is written in assembly and tuned for the processor.

Re: How does Apple’s new A12X processor compare to other 7nm mobile SoCs?

garysims — Wed, 14 Nov 2018 02:52:28 -0000

No the kicker isn't in the "bulk of the tests" at all. What I mean is that for all of the test but one then Geekbench uses C, for the other one it uses assembly language tuned for the processors, specifically for the memory test. Please explain how you "could code the same program twice with variations that do not produce significant performance differences on a microscopic level, but do produce significant differences on a macroscopic level."

Re: How does Apple’s new A12X processor compare to other 7nm mobile SoCs?

garysims — Tue, 13 Nov 2018 23:43:43 -0000

Andrei's deep dive will certainly be lower level than my comparison, but in fairness we are aiming at different audiences. However I think it is a bit harsh to call my video "trash."

Re: How does Apple’s new A12X processor compare to other 7nm mobile SoCs?

garysims — Tue, 13 Nov 2018 23:42:01 -0000

Lou, I was intrigued by your dual boot comment. Do I went and looked for YouTube videos showing this and the very first one I watched didn't show "massively different numbers." I might look into this more but for the moment I don't see that.

Re: How does Apple’s new A12X processor compare to other 7nm mobile SoCs?

garysims — Tue, 13 Nov 2018 23:29:44 -0000

Bst01, Geekbench is not coded differently across iOS and Android. It uses the same C code for the bulk of the tests. I have checked this with the Geekbench engineers.

Re: How does Apple’s new A12X processor compare to other 7nm mobile SoCs?

garysims — Tue, 13 Nov 2018 15:08:05 -0000

I reject your premise that the burden of proof is on me. If this was an article about benchmarks in which I failed to provide a logical argument then yes, you could say the burden of proof is on me. However you have made a comment saying the tests are not valid but you haven't said why, you just listed a few things like RAM. But I will humor you.

Your initial comment was that the devices run different OSes. True. However you say that means that the tests are invalid, which isn't true. When I ask why you said, "Kernel, screen resolution, RAM and RAM spec, storage and storage spec."

So the kernel is the first item. The kernel performs several tasks relevant to our tests: Tasking scheduling, RAM management and IO. We can assume that the first is negligible as there is only one high priority tasking running, the benchmark itself. The exception is during the multi-core tests. The question is does one OS provide a significant performance boost over the other due to its scheduling of threads. No, it doesn't. For proof of this we can look further afield to Linux and macOS in general. Will there be a difference, sure, but not enough to bias the results. As for RAM management and IO, I will get to those in a moment.

Next screen resolution. This of course is a factor for GPU tests. For Geekbench it is completely irrelevant. For AnTuTu it is indeed a factor, however AnTuTu has also altered its loads to include off screen rendering. Also, the screen resolutions of the devices under test are similar, not the same, but similar. So, overall nothing here, please move along.

Up next on your list was RAM. Well RAM isn’t part of the OS and it is actually part of the subsystem that we want to test. The differences in RAM performance are important to us, hence running the benchmarks in the first place. If you mean RAM management then that is a part of the kernel as I mentioned earlier. If the tests are doing lots of dynamic memory requests, asking for RAM from the OS and then returning it, then you would be correct. This would exercise the OS much more than the hardware. However, that is not what the tests do. I have spoken with the people at Geekbench about this (but not AnTuTu) and the general idea is that RAM requests are handled out-of-bandwidth, say in an initialization phase which is either unmeasured or has little impact on the overall timing results.

Now we come to storage. It is indeed true that Geekbench contains IO related tests and it is also true that these do have a heavier OS reliance than the other tests. However two things to note. 1) Is that the tests are the same on both operating systems (i.e. using SQLite) and they aren’t specific aimed directly at the native filesystem. 2) While I agree that a bias dependent on the performance of the filesystem will exist, the question is will that biases be more than the performance characteristics of the hardware. Tough question. I will go with no, because the same tests run on different generations of the same platform (say the iPhone 7S compared to the iPhone 8 compared to the X) show the same relative performance increase as observed across platforms. In other words, if the processor under test shows performance gain X on iOS compared to iOS, but a very different relative performance gain on iOS to Android then something is wrong. However, we don’t see that.

One more things to mention.The tests from Geekbench and AnTuTu are written in C, so there is no comparison of the Java Runtime library with Obj-C/Swift etc.

Also, I don’t fully understand your comment, “If the differences are negligible, why have a different app at all? It's all the same right?” What exactly are you referring to here? The differences in Android and iOS meaning they need different apps? If so, then the point is that you don’t need different apps (substantially) when you use a cross platform language like C. A couple of years ago I wrote a benchmark app (in C) which ran on Android and iOS with only the wrapper being different for each OS. The results confirmed the benchmark data from apps like Geekbench. My test app, did no memory allocation and no IO. It was as close to bare metal as could be. As I said, the results confirmed other third party results.

Finally, cross platform benchmarking is used all across the industry as a way to test hardware while minimizing the impact of the OS. Well known examples including tests like SPECint.

Re: How does Apple’s new A12X processor compare to other 7nm mobile SoCs?

garysims — Tue, 13 Nov 2018 13:58:45 -0000

Well Spencer, I have a degree in Computer Science and I was a professional software engineer for 10 years including writing system software, embedded software and doing performance profiling. So just saying to me, "it displays a huge lack of technical understanding" isn't really going to convince me. If you truly think that these tests are invalid then I suggest you tell me how much time you think these tests are spending in say the kernel during a computationally heavy task like those performed by Geekbench. Please tell me how much influence you think the different operating systems have on the tests. Please extrapolate.

Re: How does Apple’s new A12X processor compare to other 7nm mobile SoCs?

garysims — Tue, 13 Nov 2018 13:45:34 -0000

Well, if we are dealing with CPU intensive tasks then the affect of the OS is minimal. During the computational tasks call to the OS are at a minimum. So I don't agree that they are clickbait or invalid.

Re: How does Apple’s new A12X processor compare to other 7nm mobile SoCs?

garysims — Tue, 13 Nov 2018 13:10:11 -0000

Please let me know what other details would have been useful?

Re: How does Apple’s new A12X processor compare to other 7nm mobile SoCs?

garysims — Tue, 13 Nov 2018 13:09:48 -0000

Why can't you?

Re: Intel hires Tesla and Apple veteran for major chip position

garysims — Fri, 27 Apr 2018 07:54:12 -0000

He did, the article says, "bringing 20 years of experience with ARM and x86-based architectures for multiple platforms." Keller is highly skilled and to be admired, but the bigger and more important question is not about Keller's qualifications, but what Intel plan to do with him.

Re: DGiT Academy: Pre-Launch and Giveaway!

garysims — Tue, 17 Apr 2018 15:02:43 -0000

Hi, I have sent you an email to the address associated with your account. When you have a moment please reply and we can sort things out. Thx.

Re: Qualcomm launches its new AI Engine brand, works on existing Snapdragon processors

garysims — Fri, 23 Feb 2018 04:42:25 -0000

Why is that strange? Qualcomm is showing that any of its processors with the Vector eXtensions can be used for advanced machine learning. It works with the 845 but also with older processors and importantly with the 660. Qualcomm's processors don't just curl up and die just because they aren't used in the current flagship smartphones.

Re: ASUS Windows 10 on ARM device benchmarks are terrible

garysims — Sat, 25 Nov 2017 03:39:53 -0000

If it was emulated (which I also suspect it was) then this article is seriously lacking. In fact there is an argument that the scores reported are excellent considering the emulation occurring.