[LOCKED] cpuminer-opt v3.12.3, open source optimized multi-algo CPU miner

[maxime506]

Hi there,

I love your miner for YescryptR16 algo. I have some interesting observations.

On my i7-2600, if I use 8 threads I only have like 40h/s per thread. But if I drop down to two I have 90h/s instead per thread

On my miners which mostly feature dual core CPU, I have 100-150h/thread.

I wonder if Hyper-threading hurts the hashrate per thread, or the algorithm doesn't scale well on multi-core CPU. The CPU usage on 8 threads are like 97% on average. SO as the one in dual-core CPU.

Did I miss something I need to optimize, or that's it? I'd love to upgrade my miners to quad-core or even i7 (Sandy bridge CPUs are cheap these days), but I want it to be worth of my investment.


Thanks

[1714982643]

1714982643

[fanisk21]

Hi there,

I love your miner for YescryptR16 algo. I have some interesting observations.

On my i7-2600, if I use 8 threads I only have like 40h/s per thread. But if I drop down to two I have 90h/s instead per thread

On my miners which mostly feature dual core CPU, I have 100-150h/thread.

I wonder if Hyper-threading hurts the hashrate per thread, or the algorithm doesn't scale well on multi-core CPU. The CPU usage on 8 threads are like 97% on average. SO as the one in dual-core CPU.

Did I miss something I need to optimize, or that's it? I'd love to upgrade my miners to quad-core or even i7 (Sandy bridge CPUs are cheap these days), but I want it to be worth of my investment.


Thanks

That it's because the L3 cache of the processors i think that is the reason why that happens. More threads dont always mean best perfomance. 

[stas260385]

Argon work, but slow. Beastpoolminer +10%
Cpu-opt 3.8.72

There is no AVX targetted code anywhere in cpuminer-opt so there must be some better compiler
generated optimizations possible on Sandybridge vs Westmere. Most HW improvements on Sandybridge
should be transparent to the SW.

The only thing I found was a new instruction encoding format in AVX. Some (many?) of the
legacy SSE instructions have new encoding for AVX and they are possibly faster on a Sandybridge
CPU than the legacy SSE encoding of the same instructions. The new encoding would only be used if compiling
for AVX which would result in different binaries between Westmere SSE4.2 and Sandybridge AVX.

This is somewhat speculative but I can't think of any other reason the AVX compiler switch makes all
the difference.

It looks like I can't kill off the AVX build after all. It will return in th enext release, however, there will still
be no reporting of AVX algo features. AVX may affect some algos or all algos or may affect different algos
differently. They are the ghost optimizations. They exist, their effects can be felt, but they can't be seen.

Hey, Joblo!
How about to add the argon2i algo for Arionum coin: https://bitcointalk.org/index.php?topic=2710248.0

[joblo]

cpuminer-opt-3.8.8

https://github.com/JayDDee/cpuminer-opt/releases/tag/v3.8.8

Added cryptonightv7 for Monero.

[alexzerg111]

cpuminer-opt-3.8.8

https://github.com/JayDDee/cpuminer-opt/releases/tag/v3.8.8

Added cryptonightv7 for Monero.

Big thank's! And for processors with instructions AES and AVX which miner under the monero_v7 will be better? cpuminer-avx.exe or cpuminer-aes-sse42.exe? or for monero_v7 are equivalent?

[joblo]

Big thank's! And for processors with instructions AES and AVX which miner under the monero_v7 will be better? cpuminer-avx.exe or cpuminer-aes-sse42.exe? or for monero_v7 are equivalent?

How 'bout you tell me? As previously explained there are no code differences between the two builds but
some users have reported a significant performance difference with some algos.

[FlowLive]

Joblo I got a question. (or anyone that could)

I just changed my ram from 3200 (Only able to run 2666Mhz, it is dual rank ram) 16-18-18 to 3200 14-14-14 Samsung B-Die to be able to run full speed of 3200 on Ryzen...

I mine YescryptR32 or R16. On R32 I get speeds of 585H/s with the 2666 CL16 vs 450H/s with the 3200 CL14...
I read it could be that this algo has memory requirements... but can't get any more info than this...

Do you know?

[4ward]

Joblo I got a question. (or anyone that could)

I just changed my ram from 3200 (Only able to run 2666Mhz, it is dual rank ram) 16-18-18 to 3200 14-14-14 Samsung B-Die to be able to run full speed of 3200 on Ryzen...

I mine YescryptR32 or R16. On R32 I get speeds of 585H/s with the 2666 CL16 vs 450H/s with the 3200 CL14...
I read it could be that this algo has memory requirements... but can't get any more info than this...

Do you know?

If I understand correctly, you went from dual channel memory to single.
Math would say that the throughput is now (3200 / 2666 / 2) ~= 60% of the original, so the more memory dependent the algo, the more speed you loose

You can also take a look at the CPU load/temps/power consumption, if it is lower that it was before, this means that you dont have enough memory bandwidth to feed cpu with enough work

Yescrypt algo is bandwidth hard, since it performs memory lookups

If interested, see more in the whitepaper https://password-hashing.net/submissions/specs/yescrypt-v2.pdf

[FlowLive]

Joblo I got a question. (or anyone that could)

I just changed my ram from 3200 (Only able to run 2666Mhz, it is dual rank ram) 16-18-18 to 3200 14-14-14 Samsung B-Die to be able to run full speed of 3200 on Ryzen...

I mine YescryptR32 or R16. On R32 I get speeds of 585H/s with the 2666 CL16 vs 450H/s with the 3200 CL14...
I read it could be that this algo has memory requirements... but can't get any more info than this...

Do you know?

If I understand correctly, you went from dual channel memory to single.
Math would say that the throughput is now (3200 / 2666 / 2) ~= 60% of the original, so the more memory dependent the algo, the more speed you loose

You can also take a look at the CPU load/temps/power consumption, if it is lower that it was before, this means that you dont have enough memory bandwidth to feed cpu with enough work

Yescrypt algo is bandwidth hard, since it performs memory lookups

If interested, see more in the whitepaper https://password-hashing.net/submissions/specs/yescrypt-v2.pdf

No it's still dual channel, running @ 1.35V like it should, I'll take a look at this paper thanks!

[joblo]

No it's still dual channel, running @ 1.35V like it should, I'll take a look at this paper thanks!

Make sure your Ryzen has the latest BIOS, there have been issues with higher speed memory.

[FlowLive]

No it's still dual channel, running @ 1.35V like it should, I'll take a look at this paper thanks!

Make sure your Ryzen has the latest BIOS, there have been issues with higher speed memory.

Yes of course it is up to date... But still this is B-Die ram 2x8GB, even on older bios it works fine.

The question is, why or how is CPU hashing less on YescryptR16 / R32 ?

[FFI2013]

I'm mining DYN on a fx-8350 and fx-8320 for some reason after mining a little while cpuminer shows my temps 0 on the fx-8320 it starts up showing the temps I tried restarting a few times but always goes back to 0. The fx-8350 shows my temps all the time

[joblo]

I'm mining DYN on a fx-8350 and fx-8320 for some reason after mining a little while cpuminer shows my temps 0 on the fx-8320 it starts up showing the temps I tried restarting a few times but always goes back to 0. The fx-8350 shows my temps all the time

cpuminer just reads the temperature from the /sys file system, zero means read failed.

[joblo]

I'm mining DYN on a fx-8350 and fx-8320 for some reason after mining a little while cpuminer shows my temps 0 on the fx-8320 it starts up showing the temps I tried restarting a few times but always goes back to 0. The fx-8350 shows my temps all the time

You might have better luck in a memory overclocking forum.

[ULF]

      Which CPU Miner version best, most stable for BTX Bitcore?
      I am using Nvidia GTX 1060 and GTX 1080ti rigs.
      Let me know, Please

[FFI2013]

I'm mining DYN on a fx-8350 and fx-8320 for some reason after mining a little while cpuminer shows my temps 0 on the fx-8320 it starts up showing the temps I tried restarting a few times but always goes back to 0. The fx-8350 shows my temps all the time

cpuminer just reads the temperature from the /sys file system, zero means read failed.

Thanks for help now at least I no where to start I no my cpu never gets hot I have good cooling and its on my gpu miner so its open air also it just drives me crazy every time I look at it and see 0

[ljglug]

CPU: Genuine Intel(R) CPU 0000 @ 2.10GHz.
SW built on Apr  1 2018 with GCC 5.4.0.
CPU features: SSE2 AES SSE4.2 AVX AVX2.
SW features: SSE2 AES SSE4.2 AVX AVX2.
Algo features: SSE2 SHA.
Start mining with SSE2.

My CPU is playing SSE2. Instruction set mining?
But my CPU supports AVX2

[joblo]

My CPU is playing SSE2. Instruction set mining?
But my CPU supports AVX2

The algo doesn't.

[GEFE5T]

Which command to change the intensity?

[ICBitcoinpeeple]

Do you think you will add cryptonight-heavy to the miner? Several coins have HFed into that Algo