PINE64

Full Version: Performance issue/question
You're currently viewing a stripped down version of our content. View the full version with proper formatting.
Greetings!

I am curious about something: as I know so far, RockPro64 should be more powerful than Odroid-XU4.

The overall experience makes it seems so (installing, browsing, response time, etc)

Checked hardinfo's benchmark, and it confirmed it.

Cryptohash benchmark says RockPro64 is 226.12 and Odroid-XU4 is 198.49 - another evidence.

However, I tried it by mining a cryptocoin aswell, and I have troubling results. With Odroid, I was capable to do it with 32khash, but with RockPro64 it was only 27khash.

Can someone explain why RockPro64's hashrate is lower than Odroid-XU4's?

Is it something to do with the kernel (which one is the most recent?)? Or is it because the clock rate or the number of threads? (Though I have to add that this mining program is armhf, but I managed to make it work by adding the armhf architecture)
mining is a heavily cpu centric application. So eight cores in exynos5422 against six cores in rk3399 is one point for the odroid.

You should give the rk3399 a better chance by using arm64 with a gcc >=7, optimizing for CortexA53,CortexA57

Next thing depends on the hash you are brute forcing. If it is a hash, that uses standard crypto methods like sha2 or aes you can gain a lot from the cpu-features of the rk3399 - personally I was thrilled while benchmarking the rk3328 against a 4th generation i5 on sha256. I got ca. 80% of the performance at 10% of the power costs. So choose your altcoin wisely.

However, I don't think you can compete with farms of apics to get a hash in time or get more out of mining than you pay for the power, but it is fun - cause you can learn a lot about hash-algorithms, hardware extensions and the "nature" of blockchain-methods itself.
Yes, I was a little bit suspicious about the number of cores/clock rate.

The truth is, the difference is not so big, so I'm semi-fine with this.

Especially, because mining is not that important for me, I usually like it as a sort of benchmark and if I don't use some of my SBCs, or the task I give to them is light on resources/CPU, it is nice to mine some (usually new/fresh) altcoin for fun.

However, I'm curious about something. As I checked, A57s are capable to work on 2 GHz, yet they are operating at 1.8 GHz. Is it possible to make it work on 2 GHz?
(10-23-2018, 11:08 AM)bobdaroz Wrote: [ -> ]However, I'm curious about something. As I checked, A57s are capable to work on 2 GHz, yet they are operating at 1.8 GHz. Is it possible to make it work on 2 GHz?

A57s - wrong forum Wink 

My A72s run fine up to 2.1GHz at 1.25V - I compile my own 4.19 kernel after minor edits to the device tree definition. Unlikely the Ayufan releases will boost clock speed too soon as those releases aimed at stability.

PS - sbc-bench includes cpuminer in benchmark mode. Broadly on my RockPro64 I got 9.6kH/s at 1.8/1.4GHz, 10.5kH/s at 2.0/1.5GHz and 10.8kH/s at 2.1/1.5GHz.
(10-23-2018, 03:49 PM)dukla2000 Wrote: [ -> ]
(10-23-2018, 11:08 AM)bobdaroz Wrote: [ -> ]However, I'm curious about something. As I checked, A57s are capable to work on 2 GHz, yet they are operating at 1.8 GHz. Is it possible to make it work on 2 GHz?

A57s - wrong forum Wink 

My A72s run fine up to 2.1GHz at 1.25V - I compile my own 4.19 kernel after minor edits to the device tree definition. Unlikely the Ayufan releases will boost clock speed too soon as those releases aimed at stability.

PS - sbc-bench includes cpuminer in benchmark mode. Broadly on my RockPro64 I got 9.6kH/s at 1.8/1.4GHz, 10.5kH/s at 2.0/1.5GHz and 10.8kH/s at 2.1/1.5GHz.

I check sbc-bench, thanks.

About the clock rate... Is there some way to adjust it - without compiling a kernel?
(10-25-2018, 10:09 AM)bobdaroze Wrote: [ -> ]About the clock rate... Is there some way to adjust it - without compiling a kernel?

I am far from an expert on this, but think the answer is

In the short term no. In the medium term yes - needs someone to setup overlays for the device tree, I think.

It is not that hard to compile a kernel these days, hell, even I can do it!
If you do give it a go, yell & I will share how I edit for overclocks.
(10-25-2018, 04:59 PM)dukla2000 Wrote: [ -> ]If you do give it a go, yell & I will share how I edit for overclocks.

Well, it is always good to learn something new, so I think I will try to do it.
(10-26-2018, 05:15 AM)bobdaroze Wrote: [ -> ]
(10-25-2018, 04:59 PM)dukla2000 Wrote: [ -> ]If you do give it a go, yell & I will share how I edit for overclocks.

Well, it is always good to learn something new, so I think I will try to do it.

Sent you a PM with I think the relevant bits: if it works I will add to my kernel build post and put in the tutorial section.
(10-18-2018, 08:43 AM)bobdaroze Wrote: [ -> ]Greetings!

I am curious about something: as I know so far, RockPro64 should be more powerful than Odroid-XU4.

The overall experience makes it seems so (installing, browsing, response time, etc)

Checked hardinfo's benchmark, and it confirmed it.

Cryptohash benchmark says RockPro64 is 226.12 and Odroid-XU4 is 198.49 - another evidence.

However, I tried it by mining a cryptocoin aswell, and I have troubling results. With Odroid, I was capable to do it with 32khash, but with RockPro64 it was only 27khash.

Can someone explain why RockPro64's hashrate is lower than Odroid-XU4's?

Is it something to do with the kernel (which one is the most recent?)? Or is it because the clock rate or the number of threads? (Though I have to add that this mining program is armhf, but I managed to make it work by adding the armhf architecture)

I'm no expert but during ATS development, I use a miner to test temps at full throttle..

It depends on the coin you mine..
it depends on the algo
But above all, it depends on optimization of the software...you wont go there only with a simple compilation..

To give you an example, I started with 25Kh/s, then 28, then 30, and now between 35-40 Kh/s - {big,litle}:{1.8,1.4}Ghz
but depends on the algorithm..

To give you an idea I hava a AMD APU 8700, 4 cores( or 2 cores, and 4 threads..maybe the better way to put it...kaveri.. )
At 3.9Ghz, it doesn on same algo close to 50Khs@65Watts of power..

It all depends. but I don't care too much in it, its only for testing purposes..and its a very good burn test!!