AMD FX very bad hashrate on CryptonightV8

[brunoalexbr]

Please provide as much as possible information to reproduce the issue.

Basic information

• Type of the CPU. AMD FX 8 threads, 8MB cache
• Type of the GPU (if you try to miner with the GPU).

Compile issues

• Which OS do you use?

  add **all** commands you used and the **full** compile output here

  run `cmake -LA .` in the build folder and add the output here

Issue with the execution

• Do you compiled the miner by our own?

  run `./xmr-stak --version-long` and add the output here

AMD OpenCl issue

  run `clinfo` and add the output here

Stability issue

• Is the CPU or GPU overclocked?
• Is the Main memory of the CPU or GPU undervolted?

At Cryptonight v7 hashrate at 3500MHz 387H/s, with same config of thread Cryptonight V7 make 270H/s. With 8 threads i can make 333H/s but with more consumition. In AMD FX processor best config is 5 threads in folow squeme:

{ "low_power_mode" : true, "no_prefetch" : false, "affine_to_cpu" : 0 },
{ "low_power_mode" : true, "no_prefetch" : false, "affine_to_cpu" : 1 },
{ "low_power_mode" : false, "no_prefetch" : false, "affine_to_cpu" : 3 },
{ "low_power_mode" : false, "no_prefetch" : false, "affine_to_cpu" : 5 },
{ "low_power_mode" : false, "no_prefetch" : false, "affine_to_cpu" : 7 },

Some implematation impact bad on performance. Old configs dont have asm parameter and this implementation cause this issue.

[Spudz76]

no, CN2v2 """caused""" the issue; it has more tough math in it

Hammer or whatever series were bad at integer sqrt/div math, but they were sweet at floating point (algo does not use)

and the ASM only runs on newer AMD asm:auto should be selecting normal good old C code version (aka asm:off aka what the old code always did) and however your compiler chooses to do its ASM from there... you may see better results with Clang vs GCC (you never said what OS though)

Likely the new normal hashrate for FX series, it's about on par with other devices "v7" to "v8" loss.

[brunoalexbr]

The loses in Fx is strong, almost 30% under the same setting. Video cards are not heavy losses. Only on the Radeons R9 285, the RX 500 series is ok.

[brunoalexbr]

if i use 8 threads, the hashrate is same under same config on v7, but consumition is higher. Before i use 5 threads with two in low power, and hashrate is higher than 8 therads.

[Zenitur]

You need to set all cores "low_power_mode" = "false"

[brunoalexbr]

i try this, but consumition is higher than best config in v7 and low hashrate. FX 8 with 3500MHz V7 5 threads make 385H/s with 118w, now V8 with 8 threads make 330H/s and 175w. Isso very bad h/w relation.

[psychocrypt]

you can not compare the power consumtion of cn-v7 and cn-v8. The algorithm has changed a lot.

[Zenitur]

With cn-v7, I'd used 7 threads of 8. It shows more hashes. With cn-v8 I'm using 8 threads of 8 (with asm algo) P.S. nvidia OpenCL stopped to work with 9800GT (but works on Fermi)