SILENTARMY v5: Zcash miner, 115 sol/s on R9 Nano, 70 sol/s on GTX 1070

[krnlx]

why not using vector store...damn it, I few more weeks and I'll get in rails with OpenCL.

vector store address must be aligned by 16 bytes, it is not possible in any round because of different offsets

[zawawa]

I have been tweaking disassembled GCN codes of SA's kernels, and there seems to be quite a bit of room for performance enhancements, especially by optimizing global memory access by reordering flat_store_dword and s_waitcnt in ht_store(). @eXtremal, how are your next batch of optimizations coming along? If they are almost ready, I will wait for them. Otherwise, I will optimize the OpenCL kernel myself and then tweak the GCN code.

xor_and_store and ht_store must be rewrited, and joined to one function.

unaligned 32 bits reads in xor_and_store -> join in 64bit in half_aligned_long -> 64bit xor in xor_and_store -> on 2,4,6,8 round 256bit shift on xi0xi1xi2xi3 in xor_and_store -> 256bit shift again in ht_store -> split in 32bit, and write in ht_store

must be rewrited to:

unaligned 32 bits reads  - > 32 bit xor -> 256bit shift -> 32 or 64 bit, or vector store
or
64 bits reads -> 64 bit xor -> 64 bit 256bit shift -> 64bit or vector store
or
64 and 32 bit reads -> 64 and 32 bit xor -> mixed 256bit shift -> 64bit or 32bit or vector store

depend on round

Excellent suggestions! Let me get to them ASAP.

[eXtremal]

Speedup +10-15% for NVIdia only:
http://coinsforall.io/distr/nvidia/input.cl
http://coinsforall.io/distr/nvidia/param.h

Sorry, but can't work more than 1 hour a day on miner now.

For other developers, you need:
- Decrease NR_ROWS_LOG to 13 or 12. ht_store function works much faster with low NR_ROWS values and when you decrease NR_ROWS, you also decrease total slots amount, because you can use less values for OVERHEAD constant.
- Optimize equihash round for big NR_SLOTS values. I begin do it in last NVidia release, but need much more work..

[zawawa]

Speedup +10-15% for NVIdia only:
http://coinsforall.io/distr/nvidia/input.cl
http://coinsforall.io/distr/nvidia/param.h

Sorry, but can't work more than 1 hour a day on miner now.

No prob, thank you for your contributions!

[laik2]

Speedup +10-15% for NVIdia only:
http://coinsforall.io/distr/nvidia/input.cl
http://coinsforall.io/distr/nvidia/param.h

Sorry, but can't work more than 1 hour a day on miner now.

No prob, thank you for your contributions!

Did you make any progress with AMD?

[eXtremal]

Did you make any progress with AMD?

Last release don't working on AMD, but if I found a reason, it will be same +10-15% on AMD cards. For more speedup, see my previous post.

[zawawa]

Did you make any progress with AMD?

Last release don't working on AMD, but if I found a reason, it will be same +10-15% on AMD cards. For more speedup, see my previous post.

Your last release is working on RX 480 with these modifications. Thanks a bunch!

Code:

// Number of rows and slots is affected by this. 20 offers the best performance
// but occasionally misses ~1% of solutions.
#ifdef cl_nv_pragma_unroll // NVIDIA
#define NR_ROWS_LOG                     16
#else
#define NR_ROWS_LOG                     18
#endif

// Setting this to 1 might make SILENTARMY faster, see TROUBLESHOOTING.md
#define OPTIM_SIMPLIFY_ROUND		1

// Number of collision items to track, per thread
#ifdef cl_nv_pragma_unroll // NVIDIA
#define THREADS_PER_ROW 32
#define ROWS_PER_WORKGROUP (64/THREADS_PER_ROW)
#define LDS_COLL_SIZE (NR_SLOTS * 24 * (64 / THREADS_PER_ROW))
#else
#define THREADS_PER_ROW 8
#define ROWS_PER_WORKGROUP (64/THREADS_PER_ROW)
#define LDS_COLL_SIZE (NR_SLOTS * 8 * (64 / THREADS_PER_ROW))
#endif

[zawawa]

I am currently getting 100-114 sol/s with RX 480. This is very nice...

[laik2]

I am currently getting 100-114 sol/s with RX 480. This is very nice...

102/103 here with 2080 OC.
Which card do you have?

[zawawa]

I pushed recent changes to my repo, including my Win32 multithreading mod:
https://github.com/zawawawa/silentarmy
New Windows binaries will be available shortly.

[zawawa]

I am currently getting 100-114 sol/s with RX 480. This is very nice...

102/103 here with 2080 OC.
Which card do you have?

XFX Black Edition with a modded BIOS.

[zawawa]

I am currently getting 100-114 sol/s with RX 480. This is very nice...

102/103 here with 2080 OC.
Which card do you have?

Oh, my numbers are with 4 threads per GPU, too. Multithreading seems to be working well so far.

[TIKCrazy]

I pushed recent changes to my repo, including my Win32 multithreading mod:
https://github.com/zawawawa/silentarmy
New Windows binaries will be available shortly.

check also this https://github.com/krnlx/silentarmy-nvmod
krlnx working on silentarmy too

[ioglnx]

@extermal: Nice optimizations I was close to hit the 500sol/s with 2 GTX1080 and 2 GTX1070 :-D
Thanks

[Amph]

I pushed recent changes to my repo, including my Win32 multithreading mod:
https://github.com/zawawawa/silentarmy
New Windows binaries will be available shortly.

is this version adding the extremal addition codes or any improved hashrate besides fixing the known bugs?

[zawawa]

I pushed recent changes to my repo, including my Win32 multithreading mod:
https://github.com/zawawawa/silentarmy
New Windows binaries will be available shortly.

is this version adding the extremal addition codes or any improved hashrate besides fixing the known bugs?

Yes. I haven't uploaded binaries yet, though. I just got new ideas for optimization. Please wait.

[disman]

138 sol - already not interested ... ((

Profit below the plinth.

200+ sol on 1070... one might think


I can see there remained some sportsmen altruists)))

[Venon]

138 sol - already not interested ... ((

Profit below the plinth.

200+ sol on 1070... one might think


I can see there remained some sportsmen altruists)))

That is the reason why this thread is quite now. If you pay $0.25/kWh, it is not profitable to mine ZEC.

[laik2]

138 sol - already not interested ... ((

Profit below the plinth.

200+ sol on 1070... one might think


I can see there remained some sportsmen altruists)))

That is the reason why this thread is quite now. If you pay $0.25/kWh, it is not profitable to mine ZEC.

Yep...switched to ethereum...more profitable right now

[laik2]

Is someone working on any improvements for AMD, expecially newer RX cards.