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

[krnlx]

Really it is a shame, that there is no opensource miner for nvidia, only with devfee one, which makes 53-54 s/s on one 1070, and low cpu usage.

As far as I know there is no NVIDIA closed-source miner with devfee that makes 53-54 S/s (hopefully you are not referring to this scam https://github.com/zcminer-dev/zcminer).

So for now Tromp-based miners are the only one available for NVIDIA (for example, https://github.com/nicehash/nheqminer).

Please, take a look. It is NOT scam(or it is scam as Claymore's fee miner). It works well only with us flypool server.

6x palit super jetstream 1070, Driver Version: 367.27
I use this driver version, because overclocking works on it.
I had to use libcudart from cuda-8-rc2, because the one from cuda8-release not working with  367.27 driver

Code:

Tue Nov  8 16:21:12 2016       
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 367.27                 Driver Version: 367.27                    |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|===============================+======================+======================|
|   0  GeForce GTX 1070    On   | 0000:01:00.0      On |                  N/A |
|100%   51C    P2   139W / 195W |   8004MiB /  8113MiB |     99%      Default |
+-------------------------------+----------------------+----------------------+
|   1  GeForce GTX 1070    On   | 0000:03:00.0     Off |                  N/A |
|100%   49C    P2   148W / 195W |   7997MiB /  8113MiB |    100%      Default |
+-------------------------------+----------------------+----------------------+
|   2  GeForce GTX 1070    On   | 0000:04:00.0     Off |                  N/A |
|100%   50C    P2   133W / 195W |   7997MiB /  8113MiB |    100%      Default |
+-------------------------------+----------------------+----------------------+
|   3  GeForce GTX 1070    On   | 0000:05:00.0     Off |                  N/A |
|100%   50C    P2    99W / 195W |   7997MiB /  8113MiB |    100%      Default |
+-------------------------------+----------------------+----------------------+
|   4  GeForce GTX 1070    On   | 0000:06:00.0     Off |                  N/A |
|100%   43C    P2    93W / 195W |   7997MiB /  8113MiB |    100%      Default |
+-------------------------------+----------------------+----------------------+
|   5  GeForce GTX 1070    On   | 0000:07:00.0     Off |                  N/A |
|100%   36C    P2    98W / 195W |   7995MiB /  8113MiB |     99%      Default |
+-------------------------------+----------------------+----------------------+
                                                                               
+-----------------------------------------------------------------------------+
| Processes:                                                       GPU Memory |
|  GPU       PID  Type  Process name                               Usage      |
|=============================================================================|
|    0      1721    G   /usr/lib/xorg/Xorg                              13MiB |
|    0     24759    C   ./zcminer_pascal_cuda-8.0                     7987MiB |
|    1      1721    G   /usr/lib/xorg/Xorg                               6MiB |
|    1     24759    C   ./zcminer_pascal_cuda-8.0                     7989MiB |
|    2      1721    G   /usr/lib/xorg/Xorg                               6MiB |
|    2     24759    C   ./zcminer_pascal_cuda-8.0                     7989MiB |
|    3      1721    G   /usr/lib/xorg/Xorg                               6MiB |
|    3     24759    C   ./zcminer_pascal_cuda-8.0                     7989MiB |
|    4      1721    G   /usr/lib/xorg/Xorg                               6MiB |
|    4     24759    C   ./zcminer_pascal_cuda-8.0                     7989MiB |
|    5      1721    G   /usr/lib/xorg/Xorg                               6MiB |
|    5     24759    C   ./zcminer_pascal_cuda-8.0                     7987MiB |
+-----------------------------------------------------------------------------+

It is modified version of Nicehash's cuda miner, it runs a lot of threads per card(I run 21).

Code:

[16:22:17][0x00007ffb74632700] stratum | Accepted share #23366
[16:22:19][0x00007ffb6affd700] stratum | Submitting share #23367, nonce 080000000000000000000000000000000000000000000000000010
[16:22:19][0x00007ffb74632700] stratum | Accepted share #23367
[16:22:20][0x00007ffb407e8700] stratum | Submitting share #23368, nonce 2a0000000000000000000000000000000000000000000000000014
[16:22:20][0x00007ffb74632700] stratum | Accepted share #23368
[16:22:21][0x00007ffb72ffc700] stratum | Submitting share #23369, nonce 000000000000000000000000000000000000000000000000000015
[16:22:21][0x00007ffb74632700] stratum | Accepted share #23369
[16:22:21][0x00007ffafbfcf700] stratum | Submitting share #23370, nonce 700000000000000000000000000000000000000000000000000022
[16:22:22][0x00007ffb74632700] stratum | Accepted share #23370
[16:22:22][0x00007ffb7d23b740] Speed [300 sec]: 179.303 H/s, 306.843 Sol/s

well if it's a shame that no one wants to make one free, then make one yourself.
People work a LOT on these things, even on free versions, so tell me again, why would anyone give it to you for free?

s/wants/can't/g

fixed.

[1715358826]

1715358826

[restless]

Is there a Windows release or not?!
Genoil's link gives a source only ?

[ioglnx]

Really it is a shame, that there is no opensource miner for nvidia, only with devfee one, which makes 53-54 s/s on one 1070, and low cpu usage.

As far as I know there is no NVIDIA closed-source miner with devfee that makes 53-54 S/s (hopefully you are not referring to this scam https://github.com/zcminer-dev/zcminer).

So for now Tromp-based miners are the only one available for NVIDIA (for example, https://github.com/nicehash/nheqminer).

Please, take a look. It is NOT scam(or it is scam as Claymore's fee miner). It works well only with us flypool server.

6x palit super jetstream 1070, Driver Version: 367.27
I use this driver version, because overclocking works on it.
I had to use libcudart from cuda-8-rc2, because the one from cuda8-release not working with  367.27 driver

Code:

Tue Nov  8 16:21:12 2016       
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 367.27                 Driver Version: 367.27                    |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|===============================+======================+======================|
|   0  GeForce GTX 1070    On   | 0000:01:00.0      On |                  N/A |
|100%   51C    P2   139W / 195W |   8004MiB /  8113MiB |     99%      Default |
+-------------------------------+----------------------+----------------------+
|   1  GeForce GTX 1070    On   | 0000:03:00.0     Off |                  N/A |
|100%   49C    P2   148W / 195W |   7997MiB /  8113MiB |    100%      Default |
+-------------------------------+----------------------+----------------------+
|   2  GeForce GTX 1070    On   | 0000:04:00.0     Off |                  N/A |
|100%   50C    P2   133W / 195W |   7997MiB /  8113MiB |    100%      Default |
+-------------------------------+----------------------+----------------------+
|   3  GeForce GTX 1070    On   | 0000:05:00.0     Off |                  N/A |
|100%   50C    P2    99W / 195W |   7997MiB /  8113MiB |    100%      Default |
+-------------------------------+----------------------+----------------------+
|   4  GeForce GTX 1070    On   | 0000:06:00.0     Off |                  N/A |
|100%   43C    P2    93W / 195W |   7997MiB /  8113MiB |    100%      Default |
+-------------------------------+----------------------+----------------------+
|   5  GeForce GTX 1070    On   | 0000:07:00.0     Off |                  N/A |
|100%   36C    P2    98W / 195W |   7995MiB /  8113MiB |     99%      Default |
+-------------------------------+----------------------+----------------------+
                                                                               
+-----------------------------------------------------------------------------+
| Processes:                                                       GPU Memory |
|  GPU       PID  Type  Process name                               Usage      |
|=============================================================================|
|    0      1721    G   /usr/lib/xorg/Xorg                              13MiB |
|    0     24759    C   ./zcminer_pascal_cuda-8.0                     7987MiB |
|    1      1721    G   /usr/lib/xorg/Xorg                               6MiB |
|    1     24759    C   ./zcminer_pascal_cuda-8.0                     7989MiB |
|    2      1721    G   /usr/lib/xorg/Xorg                               6MiB |
|    2     24759    C   ./zcminer_pascal_cuda-8.0                     7989MiB |
|    3      1721    G   /usr/lib/xorg/Xorg                               6MiB |
|    3     24759    C   ./zcminer_pascal_cuda-8.0                     7989MiB |
|    4      1721    G   /usr/lib/xorg/Xorg                               6MiB |
|    4     24759    C   ./zcminer_pascal_cuda-8.0                     7989MiB |
|    5      1721    G   /usr/lib/xorg/Xorg                               6MiB |
|    5     24759    C   ./zcminer_pascal_cuda-8.0                     7987MiB |
+-----------------------------------------------------------------------------+

It is modified version of Nicehash's cuda miner, it runs a lot of threads per card(I run 21).

Code:

[16:22:17][0x00007ffb74632700] stratum | Accepted share #23366
[16:22:19][0x00007ffb6affd700] stratum | Submitting share #23367, nonce 080000000000000000000000000000000000000000000000000010
[16:22:19][0x00007ffb74632700] stratum | Accepted share #23367
[16:22:20][0x00007ffb407e8700] stratum | Submitting share #23368, nonce 2a0000000000000000000000000000000000000000000000000014
[16:22:20][0x00007ffb74632700] stratum | Accepted share #23368
[16:22:21][0x00007ffb72ffc700] stratum | Submitting share #23369, nonce 000000000000000000000000000000000000000000000000000015
[16:22:21][0x00007ffb74632700] stratum | Accepted share #23369
[16:22:21][0x00007ffafbfcf700] stratum | Submitting share #23370, nonce 700000000000000000000000000000000000000000000000000022
[16:22:22][0x00007ffb74632700] stratum | Accepted share #23370
[16:22:22][0x00007ffb7d23b740] Speed [300 sec]: 179.303 H/s, 306.843 Sol/s

well if it's a shame that no one wants to make one free, then make one yourself.
People work a LOT on these things, even on free versions, so tell me again, why would anyone give it to you for free?

s/wants/can't/g

fixed.

Well the repo is offline since days that i swhat i said.. i want a working one..or maybe you can upload your copy.
thanks

[Genoil]

Is there a Windows release or not?!
Genoil's link gives a source only ?

It is a Python script + exe.

1. Download as ZIP on https://github.com/Genoil/silentarmy/tree/windows and unzip
2. Install Python 3.5+ for Win x64: https://www.python.org/downloads/windows/ (allow installer add python to PATH)
3. Launch .bat file (don't edit  )

[krnlx]

Well the repo is offline since days that i swhat i said.. i want a working one..or maybe you can upload your copy.
thanks

I PM'ed u.

[restless]

Is there a Windows release or not?!
Genoil's link gives a source only ?

It is a Python script + exe.

1. Download as ZIP on https://github.com/Genoil/silentarmy/tree/windows and unzip
2. Install Python 3.5+ for Win x64: https://www.python.org/downloads/windows/ (allow installer add python to PATH)
3. Launch .bat file (don't edit  )

Python  in D:\PY
SA zip exracted to d:\silentarmy
D:\silentarmy>python silentarmy
Connecting to us1-zcash.flypool.org:3333
Could not find 'D:\PY\python36.zip\sa-solver' binary; make sure to run 'make' to compile it


Uninstall& fresh install of python solved the error

[marvykkio]

from how I see the trend of zcash, I do not think it too much to continue

 

[mo35]

Is there a Windows release or not?!
Genoil's link gives a source only ?

It is a Python script + exe.

1. Download as ZIP on https://github.com/Genoil/silentarmy/tree/windows and unzip
2. Install Python 3.5+ for Win x64: https://www.python.org/downloads/windows/ (allow installer add python to PATH)
3. Launch .bat file (don't edit  )

well it works , but speeds are awful , 25 sols on 1080

[eXtremal]

This patch gives +19% on NVidia cards

diff --git a/input.cl b/input.cl
index 91b7021..60a3ffe 100644
--- a/input.cl
+++ b/input.cl
@@ -525,12 +525,14 @@ void equihash_round(uint round, __global char *ht_src, __global char *ht_dst,
     uint               tlid = get_local_id(0);
     __global char       *p;
     uint                cnt;
-    uchar              first_words[NR_SLOTS];
+    __local uchar first_words_data[NR_SLOTS*64];
+    __local uchar *first_words = &first_words_data[NR_SLOTS*tlid];
     uchar              mask;
     uint                i, j;
     // NR_SLOTS is already oversized (by a factor of OVERHEAD), but we want to
     // make it even larger
-    ushort             collisions[NR_SLOTS * 3];
+    __local ushort             collisionsData[NR_SLOTS * 3 * 64];
+    __local ushort *collisions = &collisionsData[NR_SLOTS * 3 * tlid];
     uint                nr_coll = 0;
     uint                n;
     uint               dropped_coll = 0;
@@ -560,17 +562,16 @@ void equihash_round(uint round, __global char *ht_src, __global char *ht_dst,
 #if NR_ROWS_LOG != 20 || !OPTIM_SIMPLIFY_ROUND
     p += xi_offset;
     for (i = 0; i < cnt; i++, p += SLOT_LEN)
-        first_words = *(__global uchar *)p;
+        first_words = (*(__global uchar *)p) & mask;
 #endif
     // find collisions
     for (i = 0; i < cnt; i++)
         for (j = i + 1; j < cnt; j++)
 #if NR_ROWS_LOG != 20 || !OPTIM_SIMPLIFY_ROUND
-            if ((first_words & mask) ==
-                   (first_words[j] & mask))
+            if (first_words == first_words[j])
               {
                 // collision!
-                if (nr_coll >= sizeof (collisions) / sizeof (*collisions))
+                if (nr_coll >= NR_SLOTS*3)
                     dropped_coll++;
                 else
 #if NR_SLOTS <= (1 <<

Replace your input.cl file with this: http://coinsforall.io/distr/input.cl
May be on AMD too, not tested.

[mrb]

Thanks eXtremal!

I should say I put ZERO EFFORTS into optimizing for Nvidia. Silentarmy v4 is a straight port to Nvidia and nothing else. I hope to have time to work on optimizations in the near future.

Also, to all those testing Nvidia, I have reports that setting OPTIM_SIMPLIFY_ROUND to 1 increases performance by +25% on some Nvidia GPUs. See https://github.com/mbevand/silentarmy/blob/master/TROUBLESHOOTING.md for instructions

[eXtremal]

mrb
AMD affected too, +5% on RX480, ~53sols/s now.
Look to GCN disassemble sometimes, I see at original code:

#if NR_SLOTS <= (1 << 8 )
                    // note: this assumes slots can be encoded in 8 bits
                    collisions[nr_coll++] =
         ((ushort)j << 8 ) | ((ushort)i & 0xff);
#else
#error "unsupported NR_SLOTS"
#endif

compiles to:

 v_cmp_ge_u32  vcc, 53, v19                            // 00000000009C: 7D8C26B5
  s_and_saveexec_b64  s[24:25], vcc                     // 0000000000A0: BE98246A
  v_or_b32      v10, v6, v8                             // 0000000000A4: 38141106
  v_lshlrev_b32  v11, 1, v19                            // 0000000000A8: 34162681
 buffer_store_short  v10, v11, s[16:19], s14 offen glc // 0000000000AC: E0685000 0E040A0B

It's global memory using. I changed to local and got +5% on Polaris and +19% on NV Pascal

[mrb]

Zero difference with or without OPTIM_SIMPLIFY_ROUND on my test system (RX 480 and R9 Nano.)

I already tried putting collisions[] in local memory and also saw zero differences as well...

Weird. What is your OS & drivers?

Pascal is a different story.

[eXtremal]

Ubuntu 16.04 and amdgpu-pro 16.30

[mrb]

And me, Ubuntu 16.04, and amdgpu-pro 16.40. So latest drivers might have slightly degraded performance...

[doktor83]

keep on this technical chat, i might learn something

[agis6]

Latest version is faster and gives more "stable" hashrates

Tested on the following:
A. 1 x R9 380 -> A lot faster than before (close to 20%), "stable" hash rate
B. 3 x R9 380X -> About 5% faster, "stable" hash rate
C. 5 x RX470 + 1 x RX480 -> Not faster but "stable" hash rate

By stable, I mean there are very small fluctuations

[Ambros]

Latest version is faster and gives more "stable" hashrates

Tested on the following:
A. 1 x R9 380 -> A lot faster than before (close to 20%), "stable" hash rate
B. 3 x R9 380X -> About 5% faster, "stable" hash rate
C. 5 x RX470 + 1 x RX480 -> Not faster but "stable" hash rate

By stable, I mean there are very small fluctuations

Can u add values to your considerations ?
It would be even better
Thank you

[mrb]

As I see FGLRX constant has been removed from params.h. So, for 15.12 fglrx leave params.h as is?

The parameter was renamed. It is now named OPTIM_SIMPLIFY_ROUND Everyone should try to set it to 1 to see if it helps, no matter what you drivers or hardware is. It is worth a try. See https://github.com/mbevand/silentarmy/blob/master/TROUBLESHOOTING.md for instructions

[agis6]

Can u add values to your considerations ?
It would be even better
Thank you

Sure, here you go:

A. 1 x R9 380
Ubuntu 14.04.4 desktop, fglrx
Was 18-21 sol/s, now is 25-26 sol/s

B. 3 x R9 380X
Ubuntu 16.04.1 desktop, amdgpu-pro
Was 84-89 sol/s, now is 92-93 sol/s

C. 5 x RX470 + 1 x RX480
Ubuntu 16.04.1 server, amdgpu-pro
Was 135-147 sol/s, now is 145-148 sol/s

[antantti]

Is it possible to edit param.h somehow to get lower cpu usage? Most mining rigs are running weak cpu's and are really struggling feeding multiple gpu's.

Replacing input.cl boosted gtx 970 to 30 sol/s area in windows 7. Cpu usage is way too high though. I don't have possibility to test nvidia@linux, how is the cpu usage in linux with green cards?