AMD Ryzen 5 1400 Linux Benchmarks: 27-Way CPU Comparison On Ubuntu

[Michael_S]

As I buy quite a lot of DVDs and BDs and systematically archive them with x264, it's quite frequent that I keep a PC running on all cores for hours on end - as such stability (and heat) is paramount.
I'm looking at the Ryzen 1700 with quite some interest, but I'll probably wait for Zen 2 as my current rig is still valid (Core i5 [email protected] GHz with 16 Gb of 2400MHz DDR3) and with the prices of RAM and GPU being what they are these days, building a new rig would be expensive.

Heh, I start my rips and reencodes at bed time and move the finished file into the appropriate place at the end of work the next day. So I'm happy running the same process with two rigs that your 4670K would blow to hell in all benchmarks.

[Apopas]

I ended up purchasing the Ryzen 5 1400 for an upcoming Linux HTPC build to be documented in a future Phoronix article in conjunction with Gigabyte and checking out their mini-ITX AB350N-Gaming WiFi motherboard.

Looking forward to it.
Thank you Michael!

[JustRob]

Such a wonderful basement, must be tough using the Projector to enjoy a Movie with all the fans whizzing.

[JustRob]

Michael - On Page 1 you say you are using GCC 6.3 but 7.1 is the newest Release version (and 8.0 is available).

I did read your Article "The Impact Of GCC Zen Compiler Tuning On AMD Ryzen Performance": http://www.phoronix.com/scan.php?pag...&num=1 .

On Page 2 of this Article it says you used "--with-tune=generic" but shouldn't "-O3 -march=znver1" be added to the Command Line for Zen, as suggested in your other Article.

Also newer gcc is likely better tweaked than the older versions, the unreleased 8.0 may or may not be 'better' depending upon the "Bug of the Day" that unreleased versions are subject to.

I'm OK with the other CPUs getting their own 'marching orders'; without optimization gcc is likely pulling in the 'let's get it working' Code as opposed to the 'Guru' Code.

I don't like bringing this up, and I realize that Intel has excellent AVX, but I wonder about the results on Page 4 where AMD failed miserably on the "Himeno" where only "-O3" was used. I wonder if other Benchmarks that are less 'AVX intensive' might also be affected.

Thanks for all the work that went into this Article. I don't want to sound unappreciative on my 2nd Post.

YT,
Rob

[duby229]

Michael - On Page 1 you say you are using GCC 6.3 but 7.1 is the newest Release version (and 8.0 is available).

I did read your Article "The Impact Of GCC Zen Compiler Tuning On AMD Ryzen Performance": http://www.phoronix.com/scan.php?pag...&num=1 .

On Page 2 of this Article it says you used "--with-tune=generic" but shouldn't "-O3 -march=znver1" be added to the Command Line for Zen, as suggested in your other Article.

Also newer gcc is likely better tweaked than the older versions, the unreleased 8.0 may or may not be 'better' depending upon the "Bug of the Day" that unreleased versions are subject to.

I'm OK with the other CPUs getting their own 'marching orders'; without optimization gcc is likely pulling in the 'let's get it working' Code as opposed to the 'Guru' Code.

I don't like bringing this up, and I realize that Intel has excellent AVX, but I wonder about the results on Page 4 where AMD failed miserably on the "Himeno" where only "-O3" was used. I wonder if other Benchmarks that are less 'AVX intensive' might also be affected.

Thanks for all the work that went into this Article. I don't want to sound unappreciative on my 2nd Post.

YT,
Rob

Well, I wouldn't expect current Zen architectures to compete well with AVX code, AMD's implementation is only half as wide as Intel's and is based on an older version of that ISA.

[JustRob]

duby229 - No disagreement that Intel is wider but with sustained AVX it loses Turbo to maintain the TDP, Ryzen does not.

I'll have move time to look at this next week but the best advice I was able to come up with, for an older GCC, is that the HPC Advisory Council advises for Himeno to use:

“-O3 -ffast-math -ftree-vectorize -ftree-loop-linear -funroll-loops”

That advice was prior to "-march=znver1" being available, it should be added to the above Command Line. There may be even newer and more optimal Options that could be added.

My point was that "-O3" was not enough for either CPU, and that AMD might have came out the worse for it, particularly since the Compiler almost certainly has less mature Code and Optimizations for Ryzen.

[mitch074]

Heh, I start my rips and reencodes at bed time and move the finished file into the appropriate place at the end of work the next day. So I'm happy running the same process with two rigs that your 4670K would blow to hell in all benchmarks.

So do I, but since I use "slow" x264 presets, a BD encode takes at best real time and at worst twice that. I usually buy several disks at once (bargain bins, buy 3 get 1 free etc.) and I'd rather not spend the week encoding them. If I used "fast" presets, I'd probably be able to encode them as I rip them, but "slow" gets me on average 30% smaller files.