 |
August 01, 2017, 09:45:14 AM |
|
So, a quick search and looking through more recent forum posts shows there is not much data regarding the newer AMD Ryzen CPU for mining with XMG. So I thought I would run some tests for those interested in the community, and in the process I found some interesting results.
Testing was done with a r5 1600 on an Asrock Taichi x370 board, not sure on the BIOS revision, I'm more than happy to check if anyone is interested. I used m-minerd for the testing. I felt that some users may be interested in the performance differences with different -e settings, so I've included a number of different settings. I was interested to see if RAM speed played as large of a role in XMG as in other tasks, as well as the impact of an overclock, so tests have been run at 3.4/2133, 3.4/3200, 3.9/2133, 3.9/3200. This was run on my horribly messy installation of windows, simply because I didnt have a good working linux boot when I had time for the tests. The numbers are my perceived "average" of the reported hashrate inside of the miner. Not the most scientific of methods, but again this was a quick test.
here's the results:
3.9/3200CAS15
6t 25% - 4.5/26.8
6t 50% - 9.9/59.5
6t 75% - 19.3/116.5
6t 90% - 19.3/112.5
6t 95% - 19.5/117.1
6t 100% - 19.5/117.1
---
12t 25% - 3.0/36.5
12t 50% - 4.8/57.3
12t 75% - 9.8/117.3
12t 90% - 9.9/118.3
12t 95% - 9.9/118.3
12t 100% - 12.5/151.4
----------------------
3.4/2133CAS15
6t 25% - 3.0/18.0
6t 50% - 8.5/52.5
6t 75% - 9.0/54.3
6t 90% - 17.0/101.8
6t 95% - 8.5/52.4
6t 100% - 16.8/99.1
---
12t 25% - 2.3/27.7
12t 50% - 4.8/57.9
12t 75% - 9.6/114.1
12t 90% - 9.6/116.2
12t 95% - 11.0/132.23
12t 100% - 11.0/131.6
---------------------
3.4/3200CAS15
12t 25% - 3.0/36.4
12t 50% - 4.9/58.3
12t 75% - 9.7/116.9
12t 90% - 9.8/117.8
12t 95% - 11.3/136.0
12t 100% - 11.1/132.6
----------------------
3.9/2133CAS15
12t 25% - 3.0/36.7
12t 50% - 4.7/56.8
12t 75% - 9.7/117.3
12t 90% - 9.9/119.1
12t 95% - 12.5/151.0
12t 100% - 12.6/151.6
What do I make of it?
Well, for starters, there seems to be a large drop off in performance gains after 75% in all situations, which to me seems to suggest a lack of optimizations for the new architecture or newer CPU in general. Also, it appears to me that clock speed is king, at least for now on Ryzen CPU. This could be due to the lack of optimization again, but that's all over my head!
Something else I noticed that may explain some of the strange performance numbers is for m-minerd, my reported CPU usage often exceeded or under performed the -e setting, ie. -e 75, usage at 20%. This could be a reporting issue, this could be a m-minerd issue, this could be a user error or installation issue. There are a lot of variables, which I am working to resolve. I'd like to retest on a clean installation of linux/windows, as well as try the other miners (spexx and wolf are the only other two I am familiar with). I would also like to do more testing with SMT disabled and compare these results to the 6 core tests with SMT enabled, if for no other reason to see how zen is utilized by the miner as well as under the current windows scheduler, etc.
Overall, however, I feel these results are more or less valid, and should work as a good representation of the Ryzen series chips, from this we can estimate performance of other models given the scaling of the zen architecture (assuming e-100, perfect scaling, similar performance to current testing method):
4c/4t - 77.8
4c/8t - 100.9
8c/16t - 201.8
12c/24t - 302.8
16c/32t - 403.7!!!
Overall very interesting!! Really nice performance, although I feel some optimizations can almost certainly be made, but again I am unsure as much of that goes right over my head!
If anyone would like me to run specific tests or gather other information please let me know! I'd be more than happy to!
|
