On Windows, 8Gb of RAM is true minimum of RAM needed for hugepages to work.

AT this physical RAM size, possibility to allocate 1Gb of properly aligned RAM depends on how many and which applications start at boot and login. Even starting and exiting program at boot may cause mem fragmentatoin due to allocations made by other programs while exited one was running.

I recommend trying the consig on machine with 16Gb or more RAM, and once you know the right way to enable it, try on smaller machine.

I can get you GPU servers for much cheaper

SuperMicro 1U Server
Single Intel Xeon Hex Core SandyBridge E5-2620 (6/12HT Cores @ 2.00GHz)
Intel Xeon Phi 7120P Coprocessor /w 16GB GDDR5 Memory & 61 Cores
64GB DDR3 Quad Channel Registered Memory for Dual CPU (Max 384GB)
2x 240GB Intel Series 520 Gen2 SSD
Maximum of 6 HDD's
Hardware RAID 0, 1, 5, 10, 50 /w 1GB Cache & BBU
1Gbps Public Network Port (Dedicated)
Optional 1Gbps Private Network Port (Unmetered)
20,000GB/mo Public Bandwidth
5 Public IP Address via Secure VLAN (/29)
Remote Access IPMI Card (KVM, Reboot & Virtual Media)

SuperMicro 1U Server
Single Intel Xeon Hex Core SandyBridge E5-2620 (6/12HT Cores @ 2.00GHz)
Nvidia Tesla K20 GPU /w 5GB GDDR5 Memory & 2496 Cores
64GB DDR3 Quad Channel Registered Memory for Dual CPU (Max 384GB)
2x 240GB Intel Series 520 Gen2 SSD
Maximum of 6 HDD's
Hardware RAID 0, 1, 5, 10, 50 /w 1GB Cache & BBU
1Gbps Public Network Port (Dedicated)
Optional 1Gbps Private Network Port (Unmetered)
20,000GB/mo Public Bandwidth
5 Public IP Address via Secure VLAN (/29)
Remote Access IPMI Card (KVM, Reboot & Virtual Media)

there is lots of room to customize (ie making it cheaper by not requiring a dedicated 1gbps port)

Are you renting them out? If so, PM me your pricing.

Yes it supports it, its a K10 architecture. I am trying to use generic.

Earlier you mentioned you are testing on some AMD Phenom. Is this yours CPU?

http://www.cpu-world.com/cgi-bin/CPUID.pl?CPUID=6498

Can you post here output of HwInfo64 or other tool showing AES-NI is supported by your CPU?

Are you 100% sure you have yam-mmc.cfg in the same folder as yam.exe ?

yvg1900

Thank for the great miner
Does this result sound ok ?

MMC Agg. SPM: 24.648, HPM: 24.159; Rnds C/I: 152/16, Don. C/I: 79/0; Cfg/Wkr SPM: 24.648/1.6044, Cfg/Wkr HPM: 24.105/1.5066 231 rnds AV=2, ART=39825

Its a dual socket E5-2650 2Ghz 8 core 20M cache with 128GB Ram Baremetal running 64Precise

using numactl as you suggested have about the same total performance.
using yam-sandy-bridge
Here is my cfg file:

Code: [Select]

threads = 0
mining-params = mmc:av=2&aesni=on&m=16384
mine = getwork://walletaddr@work.mmcpool.com:80:8880:8881:8882:8883/mmc
mine = getwork://walletaddr@mmcpool.1gh.com:8080:8081:8082:8083/mmc
compact-stats = 1
print-timestamps = 0
change to av=1
got
MMC Agg. SPM: 26.581, HPM: 26.553; Rnds C/I: 2048/160, Don. C/I: 79/3; Cfg/Wkr SPM: 26.581/1.6854, Cfg/Wkr HPM: 26.554/1.6597 2127 rnds AV=1, ART=36152

Reasonable. I would also try with less memory and finetuning enabled (av=0) to see which AV will be selected.

Updated M7m version of yam miner available. It includes major performance improvement for PTS CPU mining - up to 2x speed in some configs.

yam M7m has improved performance on all CPU architectures. Most performance improvement made for Haswell (AVX2) CPUs.
New version has new algo variants for 512 and 1024 Mb mem setups. Make sure that you performed finetuning after upgrade to get most of your mach.
yam M7m does not support 2048M per thread for PTS mining. Maximum is 1024M per thread on all OSes. Please update your configs.

Downloads are at usual locations, as well as MD5 checsums and diginal signature files.

yvg1900

Updated M7m version of yam miner available. It includes major performance improvement for PTS CPU mining - up to 2x speed in some configs.

yam M7m has improved performance on all CPU architectures. Most performance improvement made for Haswell (AVX2) CPUs.
New version has new algo variants for 512 and 1024 Mb mem setups. Make sure that you performed finetuning after upgrade to get most of your mach.
yam M7m does not support 2048M per thread for PTS mining. Maximum is 1024M per thread on all OSes. Please update your configs.

Downloads are at usual locations, as well as MD5 checsums and diginal signature files.

yvg1900

Quote from: ManeBjorn on February 08, 2014, 06:26:30 pm

I used the generic YAM version as none of the others seemed to be the correct one for the Richland AMD CPU.
I do have the cfg in the same folder. 
I can't even get it to run at all it just opens and closes.  I have no idea what is messed up here.  I don't have any issues running any other command line mining software.
This one has me stumped.
I have been perusing the other thread that was mentioned to try and see if I can figure out what I am doing wrong as well.
Frustrating as I know this would boost my performance over mc2.
I cannot even get it to run in finetune mode either.

I'll leave this one to yvg1900 as I don't really know which miner should be downloaded for AMD (generic does sound like the logical choice though).
Also, are you running it from the command line or a batch file?
You could try to see what the error is either by adding "pause" to the batch file or simply running it from the command line and checking the printouts.

Again, Richland is NOT a microarchitecture name, it is core (platform) name. Your choice shall be bdver2 or btver2 (those for Bulldozer and Bobcat) versions. Richland cores seem to be all based on Piledriver microarchitecture, which supports AVX instructions and shall let you some more squeezing of CPU perf.

Also it supports AES-NI, so aesni=on shall be working.

What is important for these CPUs is to start with minimum memory setup (m=1024) and use finetuning to detect proper AV. Besides of commens made by me in AV description shall be valid for most cases, I strongly recommend relying on finetuning in determining proper AV for your system. After you are done with m=1024, it makes sense that you try more RAM, but after some point it may cause some perf degradation as mem bandwidth will become a bottleneck, so there is a balance.

Important note is setting up affinity even for single-CPU machines if you are not running as many threads as you have logical CPUs. Here is an excertp from readme.txt of upcoming version of yam that will be released in few days:

* Thread Affinity and Running less-then-cpu-count threads
If you are planing to run yam with number of threads less than your CPU count (for example, running 4 threads on 4-core CPU with HT enabled),
in some cases it is beneficial to manually specify thread affinity to prevent occasional core-sharing and cross-core thread migration. You may
want to ensure that number of CPUs enabled in your affinity mask matches number of threads you are using for mining, and that you have minimized core
sharing. To complete example above, logical CPUs (threads) 0 and 1 can be assigned to physical core 0, threads 2 and 3 to physical core 1, etc.
In this case when running 4 threads it makes perfect sense to set affinity mask such a way that only logical CPUs 0, 2, 4 and 6 left enabled for mining,
and guarantee than core sharing may not happen.

yvg1900