Monero Mining Performance

[Patrick]

Added AMD Ryzen 7 1700 results.

 

[Patrick]

Added AMD Ryzen 7 1700X Monero mining results.

 

[Marsh]

Is it me?
I think there are cheaper option to mine Monero other than buying a new AMD Ryzen.
Why AMD Ryzen H/s is lower than some Intel v3 CPU ? Lack of L3 cache?

 

[Spartus]

Awww, the ryzen results are disappointing. Ryzen was beating 6900k in some of the SHA256 and AES synthetics I saw.

 

[Patrick]

Is it me?
I think there are cheaper option to mine Monero other than buying a new AMD Ryzen.
Why AMD Ryzen H/s is lower than some Intel v3 CPU ? Lack of L3 cache?

Yea the issue is L3 cache size. 8MB cache/ 2MB problem size = 4 threads of 16 active. E5's are top dog for monero CPU mining.

 

[Klee]

Just an update, since I stopped using minergate and started to use pool.minexmr.com my accepted shares have been at %100.

Yes mining XMR the L3 cache size is the most important thing. Use a cpu with the most L3 that you can get/find/afford.

I lucked out because the E5-2667 V3 ES cpu's I have have 35mb each so I can run all 32 threads.

My E5-2670 V1 retail cpu's have have the same all core boost and number of threads, the difference is 20mb vs 35mb and being two generations older. The E5-2667 V3 have about %60 better performance.

Ubuntu 16.04 and Wolfs GPU miner has been %100 stable and has not crashed not once.

I have been wanting to make a detailed guide on running it on Ubuntu 16.04 with the AMDPRO driver but its running so good now, i really need to pull the current ssd's out and put an extra 240 gig ssd in and install Ubuntu on that after I set my bios to legacy only so I can mod the bios on the video cards to see if I can improve the performance without having to worry that the UEFI will choke on the modded video card bios.

 

[Klee]

Just a thought has anyone tried disabling hyper threading on a cpu where you have too many cores vs cache to use them all?

I think it would hurt the numbers i'm getting on this build but I wonder if it would help on my older build with the 2670's, its disassembled and the motherboard, and cpu's and ram will probably go on ebay this weekend so I can't test it out.

 

[Patrick]

Added AMD Ryzen 1800X Monero mining benchmarks to the first post. Modest improvement over the 1700X.

 

[Spartus]

Yea the issue is L3 cache size. 8MB cache/ 2MB problem size = 4 threads of 16 active. E5's are top dog for monero CPU mining.

Uh, if you were only using 4 threads that's your problem. and why it performs at half what i expected...

Ryzen has 4 MB L2 and 16MB l3. it is 8MB L3 per cluster complex and there are 2 clusters. Not 8MB total.

You should definitely retest at 8 and 10 threads. and what the hell test some others too because you never know...

 

[Spartus]

Uh, if you were only using 4 threads that's your problem. and why it performs at half what i expected...

Ryzen has 4 MB L2 and 16MB l3. it is 8MB L3 per cluster complex and there are 2 clusters. Not 8MB total.

You should definitely retest at 8 and 10 threads. and what the hell test some others too because you never know...

Also, as an aside, there is a slight downside to this layout for single threaded tasks because no single thread can access more than 8LB of L3 cache.

Also, there is a big penalty to context switches, so try core locking as well. You might need to turn off SMT though if the core topology is not being locked properly and some cores are idle...

 

[Patrick]

@Spartus you are totally right (and I was tired.)

Ryzen is not displaying the correct value via lscpu:

Architecture:          x86_64
CPU op-mode(s):        32-bit, 64-bit
Byte Order:            Little Endian
CPU(s):                16
On-line CPU(s) list:   0-15
Thread(s) per core:    2
Core(s) per socket:    8
Socket(s):             1
NUMA node(s):          1
Vendor ID:             AuthenticAMD
CPU family:            23
Model:                 1
Stepping:              1
CPU MHz:               3600.000
BogoMIPS:              7180.28
Virtualization:        AMD-V
L1d cache:             32K
L1i cache:             64K
L2 cache:              512K
L3 cache:              8192K
NUMA node0 CPU(s):     0-15

Which was keeping the thread count too low. Updated the Ryzen 7 1800X numbers and now getting over 500H/s.

 

[Spartus]

Awesome Patrick, that is what I expected. Better even, beating 6900k @ 4ghz OC...

Did you try 8 threads, 10, or both? I suspect 10 will be best due to the cache design of ryzen. Regardless, > 500 fulfills my expectations completely

@Spartus you are totally right (and I was tired.)

Ryzen is not displaying the correct value via lscpu:

Architecture:          x86_64
CPU op-mode(s):        32-bit, 64-bit
Byte Order:            Little Endian
CPU(s):                16
On-line CPU(s) list:   0-15
Thread(s) per core:    2
Core(s) per socket:    8
Socket(s):             1
NUMA node(s):          1
Vendor ID:             AuthenticAMD
CPU family:            23
Model:                 1
Stepping:              1
CPU MHz:               3600.000
BogoMIPS:              7180.28
Virtualization:        AMD-V
L1d cache:             32K
L1i cache:             64K
L2 cache:              512K
L3 cache:              8192K
NUMA node0 CPU(s):     0-15

Which was keeping the thread count too low. Updated the Ryzen 7 1800X numbers and now getting over 500H/s.

 

[Patrick]

I did not get as good of results with 10 and 16 was extremely poor.

Also, 133W while mining with 32GB (2x 16GB DIMMs) on the 1800X

 

[Spartus]

I did not get as good of results with 10 and 16 was extremely poor.

Also, 133W while mining with 32GB (2x 16GB DIMMs) on the 1800X

Nice, a solid 4H/W-second. I think that easily beats my R9-290s for efficiency.

 

[Patrick]

Nice, a solid 4H/W-second. I think that easily beats my R9-290s for efficiency.

True. The dual Xeon E5 setups are way ahead though. I think I have some of the 2x 12 core machines running at just under 900H/s at 160W each or so.

 

[Klee]

Wondering about Naples..........32 cores/$$....... mmmmmm

 

[Klee]

Looks like Monero is climbing, over $15.00 again.

 

[Marsh]

Good news,
I am thinking reducing numbers of worker since our Electric company just raised $$ this month.

For now, I am burning 26KW per day now to produce ~ 6KH/s.

 

[Spartus]

Good news,
I am thinking reducing numbers of worker since our Electric company just raised $$ this month.

For now, I am burning 26KW per day now to produce ~ 6KH/s.

assuming you mean 26 kWh, that is pretty damn good. nearly 6H/Ws. That should be profitable until well over 40c /kWh.

 

[compuwizz]

I'm doing 1069 H/s with dual Xeon E5-2640v4 with 25 threads using xmr-stak-cpu on ubuntu. The idrac reports 188watt. I've got 3 nodes. I've got a couple more spare CPU cores on some vmware hosts that could probably chip in some spare cycles too.

I'm contemplating solo mining but I'm not sure how to find out if it would be beneficial or not.