14-Way Intel/AMD Benchmarks On Ubuntu 17.04 + Linux 4.10

[7oby]

Hi Michael,

Please consider updating the OpenSSL Test from version 1.0.1 to 1.0.2:

https://openbenchmarking.org/test/pts/openssl

Version 1.0.2 contains many Intel hand optimized assembler routines, available in recent CPU generations. See Ivy-Bridge, Haswell, Broardwell optimizations in RSA-4096 (that's the one you test):

https://software.intel.com/en-us/articles/improving-openssl-performance

version 1.0.2 is quite common these days: e.g. Ubuntu 16.04 LTS features this already and makes use of it when Setting up webservers with apache or the like. I know results differ in this case and make comparing results difficult. The alternative would be a dedicated article comparing improvements beween CPU generations and software versions.

the above optimizations use plenty of recent CPU extensions such as AES-NI, AVX(2), MULX, ADCX, ADOX, RORX, and RDSEED. It would be interesting to see how these perform on New AMD CPUs since they for sure have different Clock-Cycles per instruction characteristics. Maybe Even extend the test to a mix of RSA, AES and elliptic curve tests (just Those that are Most commonly used by current browers).

[Michael]

Hi Michael,

Please consider updating the OpenSSL Test from version 1.0.1 to 1.0.2:

https://openbenchmarking.org/test/pts/openssl

Version 1.0.2 contains many Intel hand optimized assembler routines, available in recent CPU generations. See Ivy-Bridge, Haswell, Broardwell optimizations in RSA-4096 (that's the one you test):

https://software.intel.com/en-us/articles/improving-openssl-performance

version 1.0.2 is quite common these days: e.g. Ubuntu 16.04 LTS features this already and makes use of it when Setting up webservers with apache or the like. I know results differ in this case and make comparing results difficult. The alternative would be a dedicated article comparing improvements beween CPU generations and software versions.

the above optimizations use plenty of recent CPU extensions such as AES-NI, AVX(2), MULX, ADCX, ADOX, RORX, and RDSEED. It would be interesting to see how these perform on New AMD CPUs since they for sure have different Clock-Cycles per instruction characteristics. Maybe Even extend the test to a mix of RSA, AES and elliptic curve tests (just Those that are Most commonly used by current browers).

Hey 7oby! I hadn't realized the OpenSSL release had introduced those improvements, so yes, will update the test profile -- probably to 1.1 now.

[kaseki]

"....here are benchmarks from 14 of the systems."

Was there supposed to be a link to the results?

[Michael]

"....here are benchmarks from 14 of the systems."

Was there supposed to be a link to the results?

Do you not see them in the article?

[VikingGe]

Old least matter, but these both sucking 95W power for no reason... even 3 year old 25W Athlon 5350 has similar perf with these

There are a lot of things that the 5350 is really *really* horrible at, though. Compiling code is unbearably slow with that thing, and anything relying on memory bandwidth is held back by the ridiculously inefficient memory controller, which is probably the worst one AMD has ever produced in terms of actual bandwith / maximum theoretical bandwidth and latency (even worse than that of the Phenom). The X3 won't always beat it, but when it does, it will be substantial.

As for the Phenom, just double the cores, add a lot of clock speed, and those things are actually still quite usable and competitive to the FX lineup. The ones used in the article weren't exactly high-end models., but still fun to see lose every single test by a massive amount.

Now for reasons I don't understand I cannot compile kernel 4.9 on my desktop no matter what I do (it just fails at the linking stage) so I can't run the timed kernel compilation test. Which is a shame because that one was always quite interesting.

Edit: I'm running the tests from the article just now on my 5350, and it has yet to win a single test against these old X3 chips. Scored 700 seconds in timed kernel compilation.... it's a good chip for what it is, a low-budget, low-power chip well suited for my home server. It is not a good desktop chip.

[kaseki]

Do you not see them in the article?

I saw only a single page that includes a list of CPUs. The only link was to the Pharonix Test Suite.

However, I had failed to notice that my intent of allowing all scripts in NoScript for your site was somehow incomplete. Fixing that revealed your graphs. Sorry for the false alarm.

kas

[dungeon]

There are a lot of things that the 5350 is really *really* horrible at, though. Compiling code is unbearably slow with that thing, and anything relying on memory bandwidth is held back by the ridiculously inefficient memory controller, which is probably the worst one AMD has ever produced in terms of actual bandwith / maximum theoretical bandwidth and latency (even worse than that of the Phenom). The X3 won't always beat it, but when it does, it will be substantial.

As for the Phenom, just double the cores, add a lot of clock speed, and those things are actually still quite usable and competitive to the FX lineup. The ones used in the article weren't exactly high-end models., but still fun to see lose every single test by a massive amount.

Now for reasons I don't understand I cannot compile kernel 4.9 on my desktop no matter what I do (it just fails at the linking stage) so I can't run the timed kernel compilation test. Which is a shame because that one was always quite interesting.

Edit: I'm running the tests from the article just now on my 5350, and it has yet to win a single test against these old X3 chips. Scored 700 seconds in timed kernel compilation.... it's a good chip for what it is, a low-budget, low-power chip well suited for my home server. It is not a good desktop chip.

I didn't said you will beat these, just that it would be quite similar to these 25Wattage top low power AM1 Athlons... i don't think that would be far from X3 no and these yours 700 secs sounds somewhat too much to me, didn't test in a while 5350 so maybe you are right if something is borked curently... Well, i might try it now also on that machine also, have Debian 8 there hopefully test-suite will work there

Or maybe just... bolded one text indicate how you are doing or something wrong happens for you What distro is that?

[dungeon]

And here are my results on Athlon 5350 from that page to compare:

http://www.phoronix.com/scan.php?pag...inux-feb&num=5

smallpt = 251 seconds > here it beat both X3 Athlon/Phenom!!!

c-ray = 72 seconds > a bit slower, but quite there

kernel compile = 570 seconds > slower yes, but not quite far from X3 Athlon's 480 secs... and why you have 700 secs there i dunno

....
Yeah and this is on WD HDD probably that eats some seconds here and there also

Again this is just 25W APU not just CPU, you can be sure these 95W X3 CPUs sucking 5-6 times!!! more power for similar results really

Michael openssl test does not download , maybe you are updating it to 1.1 or something?

[dungeon]

it's a good chip for what it is, a low-budget, low-power chip well suited for my home server. It is not a good desktop chip.

It is low power Desktop chip, quite usable for that if you don't compare it perfromance class and power sucking products of today or better at the time that was released... it is only not for gaming on crappy APIs as single is slow in comparison to performance or even mainstream products

How something is good or not for gaming on crappy API and more CPU bound drivers, you can found by looking as let say on "libjpeg-turbo tjbench"

http://www.phoronix.com/scan.php?pag...inux-feb&num=3

Singlethreaded test and scale ideal, just like CPU bound gaming lets run that on 5350...

57 M/sec, slow single - so not for gaming

Thing is once low power Ryzen product appear that would replace all previous Bulldozer performance/mainstream based products or perform somewhere in between these... similar perf again but way less watts used, blah. blah...

And final point is in about 2-3 years Michael should drop also these current non Ryzen CPUs as those will just suck power again

[VikingGe]

kernel compile = 570 seconds > slower yes, but not quite far from X3 Athlon's 480 secs... and why you have 700 secs there i dunno

Just checked my results and noticed that I had CFLAGS set to -O3 -march=native -fomit-frame-pointer for some reason. Not sure if it affects the kernel compilation test at all, or if it's just due to platform/compiler differences or whatever. I'm getting the same C-Ray numbers as you do, though.

But my point stands, it's slow, and often (again, not always) my 19W Kaveri notebook does a lot better, and totally blows it out of the water in single-threaded tasks when the 3.3 GHz turbo mode kicks in, which the Athlon simply doesn't have.

Or maybe just... bolded one text indicate how you are doing or something wrong happens for you What distro is that?

Arch on all my machines. Now my kernel 4.9 build issues aren't limited to the PTS test, I also cannot build the kernel from the ABS or any patched kernel from the AUR. 4.10 works fine though, I really have no ideal what's going on there.