Thanks for the info Gary, I made added the code to a godbolt instance here: godbolt.org/z/d77ovd and added clang 10.0 for posterity, all with the same flags (I'm not a windows expert so there may be more optimization options).

Seems as if GCC replaces printf calls will putchar in places. The rest of the assembly is identical and not particularly slower or faster in any way. The difference in performance is an artefact specific to this very code and doesn't even concern the compiler. It's a difference in the runtime library and its use (printf vs putchar in terms of optimisation) and not a general issue with MSVC at all.

@@totalermist Sorry but I completely disagree for several reasons. 1. The code hardly does any output. You can actually see where the code takes a long time between solutions 5 and 6, for example. 2. All 3 versions of the subleq function do equal amounts of output but some are faster, again proving that the output is not a factor. 3. If you remove all the output, except for printing the times (which itself isn't timed so it doesn't matter if it is slow) then the performance remains the same. Sorry.

@@GaryExplains Gary you are correct - the issue is something else entirely. It's so obvious that I'm ashamed to say it took me a while to see. You are comparing apples to oranges in this test! *GCC* is a *C* compiler and renaming the subleq-program to .cpp doesn't change the fact that it's still C code! Use this *C++* version pastebin.com/ZA7PsECk You'll see that even with MSVC the CPP version is actually *faster* than the pointer "optimisation" and not slower (at least on my machine) than the g++ and even the gcc version. MSVC is a bad C compiler - that's no secret. But if you use it to compile *actual* C++ code and not C code (which the gcc compiler is *much* better at optimising - it's used to compile the Linux kernel after all!), it does an excellent job. Btw, clang generates the same output as msvc when you throw the C program at it and is also slower than the gcc code... Again - if you use proper C++ instead, the optimisation engines kick in and make the code as fast as it should be. So shame on me for not immediately realising that and shame on you for throwing C code at a C++ compiler and then wondering why it fails to perform compared to a C compiler...

​@@totalermist Dude, all you have done is create a fourth variant of the function, this time a C++ version that is quicker than the original. Congratulations. I have about 20 of those here. Micheal Barber (another viewer) has also submitted one here: ukposts.infoUgxWgvnB5MmJnSOknDp4AaABCQ?lc=UgxVMvqvmdh9Zc9bJjd4AaABAg.9CQboIaz9Qq9CQzDW1EYme All very interesting, but not the point of the exercise. I want to know why is the first one slow and why is subleq_g3 not. MSVC is a C and C++ compiler. It can handle the C stuff just fine. Renaming my test files to just .c and using clang, cl, and gcc all yield the same results. BTW, I am not comparing apples to oranges, GCC is also the generic name for the GNU Compiler Collection and it says very clearly at the top of the code to use g++ (but I also mention you can compiler it with gcc). But as I said, using the .c extension and just gcc, clang or cl doesn't alter the results at all. So, since you seem a bit blinked about the whole C/C++ thing, here is a special version of the question, just for you. Using pure C code (with .c files) etc, why is MSVC slower than gcc and clang? Why is subleq_g3 quicker using MSVC when it shouldn't be. Is there a workaround that doesn't involve hand optimizing the code for each and every function when clang and gcc don't need such mollycoddling.

Rotfl

Lmao ! Get this to the top rfn

Yes, I tried restarting my computer. Linux was faster.

Thanks, Ben, that helped... restarted into Linux and what do you know? Instant speed boost.

🤣

Yes, so am I. That will be in the video about using Speed Test G PC under macOS, as I said towards the end of the video.

Gary Explains I missed it. Sorry about that! Thanks for replying. Excited to see the results!

Can you even download gcc in macOS? Btw i like eating macos.

For comparing compilers with smaller programs or code snippets, you can always use godbolt.org for assembly and quick-bench.com actual benchmarking.

watch "what is a creel?" video comparing the compilers, clang performed the optimizations even better than gcc (for those examples at least)

The dark side is so not the jedi way.

MSVC is an exceptionally good compiler, as is LLVM.

@@fireclown68 yea these linux goons need to stop saying their feature less compilers are great. Its not about what runs an 8 queens fast.

@@fireclown68 lol did you even watched the video ?

@@LoyalSlime As a user you want the fastest program. Who cares about features when your program will be this much slower?

😁

*vs compiler

Yup, there used to be a mingw build of Blender that pretty much negated the speed difference between linux and windows builds. Unfortunately it wasn't maintained.

Windows is slow to use in general user experience

Thanks to the wonderful clickbaity title, that's what you get. Talking about compilers on the video title wouldn't get many views

@@pipyakas But it will make people comment in your video.

a pre-requisite question is wether he knows assembly well enough to do justice to that. Does Dave ?

It usually is though because it's more lightweight

Been using it since the same time too. DJGPP was my gateway drug thanks to id Software using it.

Linux can't handle high memory usage, and freezes on nearly full memory usage, never happens on windows. Also, battery performance in linux is very poor on nvidia optimus laptops.

@@JazibOfficial that is just generally false. there is something wrong with your setup if that is happening.

@@aktw1234 Linux eats up the battery as it does not fully support hybrid graphics(Nvidia optimus), so if your laptop has a dedicated graphics card, it will almost always use it to draw views on the screen, which eats up a lot of battery. Hence, poor battery performance.

عمر؟

Appreciate it!

Glad it was helpful!

I found the opposite true for rust lang.msvc toolchain was faster than gcc toolchain for the same prime generator program

Those are interesting results... but for my test suite it needs to be automated using my own code. However the code tries to be as "real world" as possible, so there is a ray tracing test in there, etc.

@@GaryExplains You can do headless renders from the command line with blender AFAIK. So that might be an easy and useful addition. You can do similar things with Inkscape and GIMP.

True, but that isn't the aim of Speed Test G PC.

In my testing using an insider version of Windows and some ARCH Linux blender gives way better results using OptixAi accelerated GPU API setting for cycles render gives better results in photo render, and animations, however while enabling bloom in eevee Linux is faster due to no GPU acceleration and just pure software render, since it doesn't need to acces gpu performance

@@GaryExplains blender supports Ray traced reflections through Nvidia studio capable PCs only available for RTX platforms on Windows 10, though u can still change code for Linux version to get the Optix Ai gpu api on Linux at a massive performance loss

No, I don't think so. Native code produced for Windows using the native tools is demonstrably slower than native code produced for Linux using the native tools. So, Linux is faster than Windows.

Yeah, this phenomenon has even got a name: "clickbait".

@Misha Amashukeli Hmmm... While I respect that maybe your view, I wholeheartedly disagree. There is a fine line between a catchy, enticing headline and clickbait. Where you draw that line is matter of debate.

@@GaryExplains Thanks for your response. Comparing two OS systems (to me) implies comparing their kernel: thread scheduler, efficiency of system calls, resource management etc, not User space applications. Microsoft justifiably deserves criticism for poor developer tools and resulting poorly performing programs built on top of those tools. But the OS itself I think is a separate entity. Recently I had a pure CPU-bound program that run around 40% slower on Linux than windows (pytest parallel), even after taking steps like closing all background apps etc. Should we really conclude something as bold as "Linux is slower than windows" from that?

Funny that you include "kernel: thread scheduler, efficiency of system calls, resource management etc" but not the tool provided by the OS maker to write your programs. Odd.

Several.

No, absolutely not. Each compiler generates different machine code.

also.So is not website ?

@Xrafter LOL

@@xrafter Now fixed?

@@makisekurisu4674 Yes.

The code has been released, see pinned comment. Please try whatever optimization options you want.

valgrind is your friend...

Use debug symbols, use GDB and Valgrind, and that should be enough!

@@GauthamKrishna9991 I was in 1st year of college back then, with very few experience with debugging. now its like 4 years ago

Yes, the WSL speeds are very similar to the native Linux speeds.

docs.microsoft.com/en-us/dotnet/framework/net-native/

That only applies to C#

Is that an important factor?

I would guess Windows because of the GPU support.

Glad it was helpful!

@@GaryExplains Which MSVC compiler did you use, the free version or the paid version that has optimizations for the specific CPU instruction set?

@@seriykot1 Can you give me a link to where Microsoft describes the "pro" level optimizations for the paid versions?

@@GaryExplains Sorry for the confusion, it's the Intel compiler that is paid for. Looking forward to the rest of your Speed Test G PC series!!

not surprising if you're a seasoned programmer.

Because the whole point is that I want to test the new Macs, which of course will be ARM based not x86.

Windows is compiled with Visual C++, so the resulting binaries won't run as fast as a (hypothetical) GCC-compiled build. Since you're running the OS binaries natively, the compiler plays a key role when it comes to performance. To that end, I wholeheartedly disagree that the video title is incorrect, but Gary probably should've explained the relationship between the compiler and OS performance better...

The WSL times were very similar to the Linux times.

@@GaryExplains strange, I ran some Nestjs test on WSL2 and Ubuntu 20.04 and the performance of Ubuntu was far better! (on the same I5 10th Gen Laptop) Outcome: I quick to Windows and I'm an Linux user since then...

@@franciscosusana2292 Does Nestjs test server load performance or CPU performance?

@@GaryExplains when I run the test I see on htop the CPU bars increasing abruptly and sometimes I have to wait for the tests to finish to use the laptop again. And by load do you mean in terms of RAM?

Windows as an ecosystem using the OS and the development tools from the OS maker results is slower code.

Im a followers of your channel, and like the way you explain things, i confess that this idea is good, but the result is not accoplished since on your title you says that Windows is slower, but you demonstrarem that the same Application compiled by the same compiler runs equally um Windows and Linux. Not all Windows applications are developed using visual studio, so the Windows ecosystem has more than VS. Unless you want intentionally to transmit a biased message, the result is not very well acconpished because you demonstrate that gnu compiler procures the same result on Windows and Linux. By Logic your vídeo title should be more like "Visual Studio compiler produces slower binaries than gcc". I think the idea is good as well as the approach, i feel your conclusions looks biased. Just trying to be imparcial

Not all Windows applications are developed using VS, sure, but a lot are, the majority I would say. My conclusion isn't biased at all, I clearly show what is happening and my conclusion is based on the facts. Please don't confuse the 5 words in the title, with the detailed content I give over 9 minutes of explanation. Don't judge a book by its cover, etc, etc.

Well, its just a suggestion, if the title were a question it would fits 100% with the content, in my opinion as it is a statement, i feel it works more like a clickbate, its just my opinion, dont take it personally. Also as a suggestion taking a real world case, as for example compiling nginx with gcc on both operating systems, serving only static files, how does it performs? how many requests per second? how much resources are used? how operating system interferes/helps? Thanks for your work, it is great and very appropriated. Hope to continue seeing your explanations

I have tried titles with question marks before and they don't work, it seems that many people can't actually read properly and don't even notice the question mark. Then people still accuse me of click-bait because they say that anything can be put as a question but still be click-bait.

😁

Yes here: ukposts.info/have/v-deo/bptka4KHaoqmsas.html

@@GaryExplains Wow, super fast reply appreciated. And a good job showing the cross-platform abilities of .Net Core. However, given the speed comparison of compilers shown here, I was wondering how the .Net Core cross-platform performance compared. It was fairly difficult comparing performance from the Hello World response times. 🤣🤣🤣 Just kidding, keep up the good work. 👍

As I said in the video there is more to investigate.

What do you mean by "streamlined"?

Gary Explains kinda like mainstream. Like how PCmark has apps. I could honestly see CPU manufacturers like AMD & Intel using your benchmarks to measure efficiency or something as time goes on

Sorry, I am not being awkward, I promise, but what do you mean by mainstream? Do you mean release it to the public?

Gary Explains yeah i may have chosen the wrong word my apologies. But yeah a mobile app or something

Thanks for clarifying. So on the mobile side I have Speed Test G and a whole channel for testing devices: ukposts.info/the/8Cd4TlU1d5mPNGKXibCWqw.html and there is an FAQ which covers thing like the availability to the public garyexplains.com/faq. On the PC side, Speed Test G PC is very new, like the paint is still drying on it, so the future is yet to be decided.

I just double checked and the binaries are native, no CLR in sight!

@@GaryExplains thanks for checking!

Windows 10 and Visual Studio 2019.

@@GaryExplains ok, are you using any insider specific version or using the standard windows?

Zaent YT who cares, Windows sucks anyway

Standard everything.

I thought the same, wich Windows version and if the correct commands to enable multicore usage on AMDs R5 to achieve a better performance.

Cristian David Home Acosta that’s right. My company provided windows virus scanner will hold up anything npm by as much as 40% because of the hundreds of thousands of files touched and generated.

Why do you think that multi-threading, multi-processing are not well presented?

Also why is not possible to compare using my suite? I haven't explained hardly any details about how it works. What assumptions are you making?

Are the flags not part of the compiler?

If only I said in the video that I checked things like that... wait, I did! 🤦‍♂️

I released the code. Please see the pinned comment. Please play with the code a let me know. Several people have tried and the conclusion has been the same.