'BandInte' - Bandwidth & Integer prowess of your machine, GUI

[Sanmayce]

Here comes a CPU benchmark generating one picture with stats, screenshot-ready, hee-hee. Thus, every random machine can be evaluated.

Many times I need a tool reporting the transferring speed of uncached RAM and a metric for CPU's ability (across all cores) in supersimple integer calculations (inhere, factorizing numbers):

Code: (Select All)

    for (i = 2; i <= n; i++) {
        while (n % i == 0) {
            n /= i;
        }
    }

Cannot resist the simplicity and throwing this two-level loop to all available threads. Since it doesn't stress caches and uncached RAM, it serves as the closest equivalent to getting the RAW and CUMULATIVE power of the CPU in "MHz" i.e. frequency department. Meaning, the resultant number represents the ability of CPU to ... loop.

The used number is 4*4096 numbers for factorization, my slowest laptop (4 threads) calculates them in ~4000 seconds whereas my fastest (8 threads) in ~2000 seconds, it might seem unnecessarily big but it is not since a formidable Threadripper has 128 threads, the benchmark has to cover it.

My laptop 'Djudjeto':

   

My laptop 'Dzvertcheto':

   

So, I wrote two console tools in C doing that, utilizing all the availble cores, they are invoked by QB64PE GUI and using the superb @OldMoses MagScope, the resultant lines are easily viewable - on any monitor (including 1366x768 modes).

All the sourcecodes are included into the attached package, even though it is fully portable, in reality it is effective/useful only in Linux environment, simply latest GCC and CLANG generated too far away from one another code. CLANG, being 2x faster, didn't dig what causes this huge gap...

Oh, since the BANDWIDTH reporter sums all the QWORDS within the memory block it is some ~2GB/s below the value reported by the AIDA's Memory Read, once I asked one of the authors of their benchmarks few things, he said that AIDA uses hand-written highly optimized Assembler. I chose different path, using C summing all the QWORDS with all the threads within AVX2 vectors, this makes it less synthetic and more reliable.

The benchmark uses 4GB and 6GB with Celeron and AVX2 capable CPUs, respectively. So, 8GB RAM are needed.

Glad that one picture can tell thousand words.

[Pete]

I'll +2 that, even though I'm running Windows... on a computer so slow that the planet might not be around before the "d" in my running "Hello World" app gets printed to the screen. Still...

Pete

[Sanmayce]

There is no looking down on machinettes on my part, no matter how slow they are.
In contrary, my favorite computer, EVER, is the above Thinkpad for ... education.

The problem, currently with Windows, is that the included GCC executable is 2x slower than the Linux CLANG counterpart! Couldn't provide the CLANG Windows binary because I lack the omp.h header which allows me not to have multi-threaded EXEs.

[Sanmayce]

Had to fix the stupid handling of mouse, so r2 is here.
Also, allowed the mouse-only (or trackpad-only) counterpart of Spacebar+Mouse.
Initially, didn't prefer holding LeftMouseButton and moving the mouse as a way to navigate/scroll the image because 2 problematic cases:
- not having mouse;
- some trackpads are not ergonomic enough (unable to handle well pressing-and-moving or even being unresponsive).
As always, the more the merrier, now either mouseless or ... mouseful one can easily zoom/scroll.

An old laptop (with 'Ivy Bridge') results:

   

It shows that 'Whiskey Lake' is a bit faster than 5 generations back. I have another Integer benchmark (which in contrast to the current Factorizer, stresses caches in schoolway multiplication digit by digit) which confirms that 8th gen is not significantly faster than 3rd, which in its turn confirms that FREQUENCY rules in simple scalar tasks.

Oh, my brother gave it a try on his (AMD Zen 3) laptop:

   

AMD site says that 7530U has 384KB L1 cache, but the actual size is 192KB instr and 192KB data, hm, never believe web-pages, run hardware readers.

The Factorization was much faster than I expected?!
These 3x boost is huge, it means that compression (a very common task) will be much faster also.

The Roster, so far:

Code: (Select All)

+-------------+-----------------------------+-------------+---------------------------+
| Submision   | Machine CPU/RAM             | Memory Read | Factorizations-Per-Second |
+-------------+-----------------------------+-------------+---------------------------+
| Sanmayce    | ThinkPad L14 Gen 4          |  37.67 GB/s |                 22.13 FPS |
|             | Ryzen 5 PRO 7530U Zen 3     |             |                           |
|             | DDR4 ? MT/s                 |             |                           |
+-------------+-----------------------------+-------------+---------------------------+
| Sanmayce    | ThinkPad L490               |  30.11 GB/s |                  7.54 FPS |
|             | i7-8565U 'Whiskey Lake'     |             |                           |
|             | DDR4 2400 MT/s              |             |                           |
+-------------+-----------------------------+-------------+---------------------------+
| Sanmayce    | HP EliteBook 8470P          |  21.04 GB/s |                  7.12 FPS |
|             | i7-3630QM 'Ivy Bridge'      |             |                           |
|             | DDR3 ? MT/s                 |             |                           |
+-------------+-----------------------------+-------------+---------------------------+
| Sanmayce    | ThinkPad 11e 5th gen        |  11.65 GB/s |                  3.90 FPS |
|             | Celeron N4100 'Gemini Lake' |             |                           |
|             | DDR4 2400 MT/s              |             |                           |
+-------------+-----------------------------+-------------+---------------------------+