Quick Tips for Mining Scrypt Coins

This is a collection of tips from the scrypt mining community for configuring CGMiner.

Read the readme.

Do it. Now. This will give you a better understanding of what is going on and how to get a higher hashrate.

Shaders

This does absolutely nothing for your configuration. Setting a thread concurrency overrides anything you enter as a shader value. You could put in –shaders 9000 and it won’t matter. To quote the readme: “Setting thread concurrency overrides anything you put into –shaders and is ultimately a BETTER way to tune performance.” The reason for the –shaders Xxxx is find a working thread concurrency value on a test run of CGMiner. If you don’t enter a TC value, put in your correct shaders, run CGMiner, CGMiner creates a file in that folder showing you what it thinks your TC should be set at. The file is named something goofy, but part of it will say something like tc8192.

Commas

If you are running the same or similar cards, you don’t need commas. If you want a TC of 8192 for 4 cards, you can just set the TC to 8192 ONCE, and it will apply it all 2,3, or 4 cards. Now, if you’re running different clocks or trying to adjust temps on different types of cards, then yes, use commas.

Check GPUz

Sometimes, CGMiner has trouble overclocking, underclocking, and controlling voltage. I don’t know why, but it just does. For this, I really stress the us of MSI Afterburner. The problem is, CGMiner will still run, and not tell you that it isn’t actually undervolting your card. Check GPUz.

Conf vs Bat

Your .conf file is NOT the same as your .bat file.

-l 13

I for the life of me, I cannot understand how people are supposedly mining at -I 13. I’ve run rigs of 4 – 6870s, 4 – 6950s, 4 – 7950s, and now 3 – 280xs. None of my rigs ran at -I 13, they all ran at either 18 or 20. I suspect this has something to do with the difference between the .conf and the .bat files. Maybe someone knows the answer to this and can help me out.

-I 20 for the non-active (not plugged into monitor) cards, -I 13 for the one that hosts the desktop, but this is also depending on thread concurrency. at -g 2, much higher than that and it faces certain doom (frozen or blank screen, artifact hell, BSOD, etc..).

It comes down to “how fast can your GPU cores chew on numbers, they have to finish chewing on their set and free gpu memory before the next one starts..” So in the case of lots of GPU Ram compared to the # of stream processors, you’ll see there is a lot of “overturn” and therefore many gpu threads is optimal.

Vectors

Direct quote from the readme: “Vectors are NOT used by the scrypt mining kernel. SUMMARY: Does nothing.” A lot of people know this, but don’t always want to speak up! Faith in the mining community restored.

Overclocking and Tuning

As a general rule, memory is far more sensitive to tuning than a CPU/GPU core, as it runs much faster.

Computing components, when manufactured, don’t come out of the shop with a particular product name. Generally, they are stress-tested at the factory and given a “this will work reliably as product X”. Some components work extra well and are sold as “Product X PLUS EXTREME”.

Power draw and heat affect the performance of a component. If you push it past it’s capacity, it will underperform. There’s a fair amount of science to this, but basically you want to find the sweet spot for your component and just accept that’s how it ended up.

This wikipedia page may explain it a bit better:https://en.wikipedia.org/wiki/Overclocking#Factors_allowing_overclocking

Thread Concurrencies

Mess around with thread concurrencies more. You can gain 5% hash rate by fiddling with your thread concurrency for a few minutes. Over a month of mining, that would mean a whole extra LTC per 7970 or 280x. The same goes with core clock. Even 1-2 MHz up or down can change your hashrate (and almost always does).

Example: my 7870 runs with TC at 12021 and my core clock is 894 and memory at 1350. Literally raising or lowering either of those numbers by 1-2 lowers my hashrate by 5-15 KH/s. And 7950s run at completely different values than I’ve seen on the internet as well. So, please, tune your cards yourselves and you’ll find that extra performance.

As for memory, I think we all know that faster is always better (until you hit instability.)

TL;DR: Spend 30 minutes playing with thread concurrency and core clock to dial in your max hashrate. profit.

7870 examples: 371 KH/s with I=18. It will run at I=20 (383 KH/s), but randomly crashes sometimes. Same thing with higher memory speeds (1400 MHz gets me 400 KH/s).

Example clocks for a 680:

“intensity” : “13,13”, “vectors” : “1,1”, “worksize” : “256,256”, “kernel” : “scrypt,scrypt”, “lookup-gap” : “2,2”, “thread-concurrency” : “10240,10240”, “shaders” : “2048,2048”, “gpu-engine” : “1000,1000”, “gpu-fan” : “65-100,65-100”, “gpu-memclock” : “1500,1500”, “gpu-memdiff” : “0,0”, “gpu-powertune” : “-20,-20”, “gpu-vddc” : “1.100,1.100”, “temp-cutoff” : “100,100”, “temp-overheat” : “95,95”, “temp-target” : “85,85”, “auto-fan” : true, “api-port” : “4028”, “expiry” : “120”, “gpu-dyninterval” : “7”, “gpu-platform” : “0”, “gpu-threads” : “2”, “hotplug” : “5”, “log” : “5”, “no-pool-disable” : true, “queue” : “1”, “scan-time” : “30”, “scrypt” : true, “temp-hysteresis” : “3”, “shares” : “0”, “kernel-path” : “/usr/local/bin”