Topic: [ANN] sgminer v5 - optimized X11/X13/NeoScrypt/Lyra2RE/etc. kernel-switch miner - page 105. (Read 877846 times)

whats wrong with this, please help

setx GPU_MAX_ALLOC_PERCENT 100
setx GPU_USE_SYNC_OBJECTS 1

sgminer.exe --kernel bitblock -o stratum+tcp://stratum.westhash.com:3336 -u 12tyqFRW384n27ytyM77edUignkpGDDbZ9 -p d=0.001 -I 18 --worksize 64 -g 2 --gpu-powertune 20 --gpu-engine 1130 --gpu-memclock 1500 --lookup-gap 2 --auto-fan --gpu-fan 40-70 --temp-cutoff 85 --temp-overheat 80

Try memclock 1500, or if that is not stable, try 1250.

AMD gpus  have memory  timings set at 125MHz intervals, always ending at even 125MHz. Like this:

875-1000
1001-1125
1126-1250
1251-1375
1376-1500
1501-1625

 mining is mostly about random access latency, not about sequential reads. You get best random access speed at the high end of each range, like 1250 or 1500.

It's very likely that 1250MHz is faster than 1400MHz, because 1400MHz is so close to the start of the range and 1250 is exactly at the end..

"290 non x Elipa 1.488 Mh/s

          "gpu-engine" : "980",
          "gpu-memclock" : "1500",
          "xintensity" : "64",
          "nfactor" : "10",
          "worksize" : "64",
          "algorithm" : "Lyra2RE",

290x hynix 1.450 Mh/s

          "gpu-engine" : "1070",
          "gpu-memclock" : "1400",
          "xintensity" : "64",
          "algorithm" : "Lyra2RE",
          "worksize" : "64",


I've had the clock speeds all over the place it just seems like the 290x's don't want to leave 1.450 Mh/s

and yet the 290's are happy to run faster easyer

Very interesting.   I get about 2% gain on 7950 and need to use (mod % 2) with the case statements adjusted accordingly.
My 280X gains almost 6% as is, but the gain difference between (mod % 2) and (mod % 4) is pretty small, like 1-2 KHs

My SMix call is a bit different, I simply put the sub-calls inline so it doesn't bother with ScratchpadStore and ScratchpadMix.
Perhaps this fits nicer into the core and needs less swapping.  

I have tried, unsuccessfully, to further streamline the SMix, but any other way I do it, its either all HW errors or vastly slower. Any guidance here would be appreciated.

 

.....This is worth 20KH/s on my 280X......from 343KHs to 363KH/s at 1020MHz clock
.....now somebody needs to find 20KH/s more for me.... 

change the XORBytesInPlace call from
to
and change the function itself to perform some byte alignment checking

.....This is worth 20KH/s on my 280X......from 343KHs to 363KH/s at 1020MHz clock
.....now somebody needs to find 20KH/s more for me.... 

change the XORBytesInPlace call from
to
and change the function itself to perform some byte alignment checking

.....This is worth 20KH/s on my 280X......from 343KHs to 363KH/s at 1020MHz clock
.....now somebody needs to find 20KH/s more for me.... 

change the XORBytesInPlace call from
to
and change the function itself to perform some byte alignment checking

Thank's to Wolf0 for your .bin files

increas from 4.3 to 6.6 MHs
x11 r9 280x



but profit same as it was get 4.3 MHs
maybe diff raise like crazy

hopefully get a special .bin that not make a crazy diff 

Here is a small neoscrypt kernel improvement for free, since I am mostly doing X11 anyway.
It gave me a 5.8% speedup on my reference R9 290 card (with Stilt bios),
from 290.2 to 307Kh/s  at 800/1500 core/mem freq on Ubuntu 12.04 with stock drivers.
I didnt try it on my R9 280x cards, so please post your results if you try this.

You will have to mod the kernel as per the code below.
The bottleneck in this kernel is the way it stores the 128 intermediate results of chacha and salsa in global memory.
By doing the change below you are reducing stalls/latency by not making read/writes to same/adjacent memory banks.

Change:
void ScratchpadStore(__global void *V, void *X, uchar idx)
{
   ((__global ulong16 *)V)[idx << 1] = ((ulong16 *)X)[0];
   ((__global ulong16 *)V)[(idx << 1) + 1] = ((ulong16 *)X)[1];
}

void ScratchpadMix(void *X, const __global void *V, uchar idx)
{
   ((ulong16 *)X)[0] ^= ((__global ulong16 *)V)[idx << 1];
   ((ulong16 *)X)[1] ^= ((__global ulong16 *)V)[(idx << 1) + 1];
}

To:
void ScratchpadStore(__global void *V, void *X, uchar idx)
{
   ((__global ulong16 *)V)[idx] = ((ulong16 *)X)[0];
   ((__global ulong16 *)V)[idx + 128] = ((ulong16 *)X)[1];
}
void ScratchpadMix(void *X, const __global void *V, uchar idx)
{
   ((ulong16 *)X)[0] ^= ((__global ulong16 *)V)[idx];
   ((ulong16 *)X)[1] ^= ((__global ulong16 *)V)[idx + 128];
}

Here is a small neoscrypt kernel improvement for free, since I am mostly doing X11 anyway.
It gave me a 5.8% speedup on my reference R9 290 card (with Stilt bios),
from 290.2 to 307Kh/s  at 800/1500 core/mem freq on Ubuntu 12.04 with stock drivers.
I didnt try it on my R9 280x cards, so please post your results if you try this.

You will have to mod the kernel as per the code below.
The bottleneck in this kernel is the way it stores the 128 intermediate results of chacha and salsa in global memory.
By doing the change below you are reducing stalls/latency by not making read/writes to same/adjacent memory banks.

Change:
void ScratchpadStore(__global void *V, void *X, uchar idx)
{
   ((__global ulong16 *)V)[idx << 1] = ((ulong16 *)X)[0];
   ((__global ulong16 *)V)[(idx << 1) + 1] = ((ulong16 *)X)[1];
}

void ScratchpadMix(void *X, const __global void *V, uchar idx)
{
   ((ulong16 *)X)[0] ^= ((__global ulong16 *)V)[idx << 1];
   ((ulong16 *)X)[1] ^= ((__global ulong16 *)V)[(idx << 1) + 1];
}

To:
void ScratchpadStore(__global void *V, void *X, uchar idx)
{
   ((__global ulong16 *)V)[idx] = ((ulong16 *)X)[0];
   ((__global ulong16 *)V)[idx + 128] = ((ulong16 *)X)[1];
}
void ScratchpadMix(void *X, const __global void *V, uchar idx)
{
   ((ulong16 *)X)[0] ^= ((__global ulong16 *)V)[idx];
   ((ulong16 *)X)[1] ^= ((__global ulong16 *)V)[idx + 128];
}

Here is a small neoscrypt kernel improvement for free, since I am mostly doing X11 anyway.
It gave me a 5.8% speedup on my reference R9 290 card (with Stilt bios),
from 290.2 to 307Kh/s  at 800/1500 core/mem freq on Ubuntu 12.04 with stock drivers.
I didnt try it on my R9 280x cards, so please post your results if you try this.

You will have to mod the kernel as per the code below.
The bottleneck in this kernel is the way it stores the 128 intermediate results of chacha and salsa in global memory.
By doing the change below you are reducing stalls/latency by not making read/writes to same/adjacent memory banks.

Change:
void ScratchpadStore(__global void *V, void *X, uchar idx)
{
   ((__global ulong16 *)V)[idx << 1] = ((ulong16 *)X)[0];
   ((__global ulong16 *)V)[(idx << 1) + 1] = ((ulong16 *)X)[1];
}

void ScratchpadMix(void *X, const __global void *V, uchar idx)
{
   ((ulong16 *)X)[0] ^= ((__global ulong16 *)V)[idx << 1];
   ((ulong16 *)X)[1] ^= ((__global ulong16 *)V)[(idx << 1) + 1];
}

To:
void ScratchpadStore(__global void *V, void *X, uchar idx)
{
   ((__global ulong16 *)V)[idx] = ((ulong16 *)X)[0];
   ((__global ulong16 *)V)[idx + 128] = ((ulong16 *)X)[1];
}
void ScratchpadMix(void *X, const __global void *V, uchar idx)
{
   ((ulong16 *)X)[0] ^= ((__global ulong16 *)V)[idx];
   ((ulong16 *)X)[1] ^= ((__global ulong16 *)V)[idx + 128];
}

+1
on my 280x i get 308 with "improved" vs 317 before