hashrate pool >> cgminer

[PeterL]

I just got my first mining rig - yeay! It's a very simple PC with 2x R9 290X inside. I spent today assembling the hardware and messing around with the settings. Thanks to the great community, it was fairly easy. Here's a few things I noticed:

• The ATI drivers work fine, as long as I stay in cgminer. If I cange things with the --gpu* switches or any of those settings in the cgminer UI, then I can't use cgminer again. I have to reboot. The SSH to my machine is also messed up, but that's quickly fixed by calling the command line reset.
• I'm mining tagcoins in the HashFaster pool. While my cgminer says I'm doing about 1200Khs/, HashFaster claims it's 2700! (Though it jumps up and down a bit, sometimes it's "only" 2000) I'd like the second to be true, of course. 1200 is way below what I expected (750 or more).

Settings: ./cgminer --scrypt -o stratum+tcp://tag.hashfaster.com:3335 -u bla -p bla -I 20 --shaders 2816 --thread-concurrency 28160 -g 1 --worksize 512

I hardly notice any difference in cgminer Kh/s when I play around with the thread concurrency or the gpu-threads, but it seems to make a difference for the pool. Actually, a thread concurrency of 1000 made the cgminer Kh/s go above 800 for the first time, but FastHash told me I was slow   A bit of explanation of the difference between the two and how to benchmark for the best results (coins mined!) would be helpful.

[iGotSpots]

Pool calculates hashrate based on your shares. If you got lucky right away and got a bunch it will sky rocket

[PeterL]

Thanks. The pool's so-called "Kh/s" has been fluctuating between 800 and 3500. Just a bit too crazy to be true.

Here's a quick copy&paste of my cgminer that has now been going for about 2 hours, stably, on both R9 290Xs at the same time:

 cgminer version 3.7.2 - Started: [2013-12-03 17:30:23]

 (5s):1.368M (avg):1.320Mh/s | A:155087  R:1280  HW:0  WU:1137.7/m
 ST: 2  SS: 0  NB: 28  LW: 1449  GF: 0  RF: 0
 Connected to tag.hashfaster.com diff 256 with stratum as user PeterJP.PeterJP2
 Block: 57872d9f...  Diff:37  Started: [19:42:13]  Best share: 219K

 [P]ool management [G]PU management [ S]ettings [D]isplay options [Q]uit
 GPU 0:  87.0C 3873RPM | 698.3K/697.5Kh/s | A:82974 R:768 HW:0 WU:607.5/m I:20
 GPU 1:  79.0C 3881RPM | 623.3K/623.0Kh/s | A:72112 R:512 HW:0 WU:530.2/m I:20

I set the GPU memory speeds manually to 1500MHz and the fan speed to 70%. GPU0 is throttling a bit (978MHz), while GPU1 is going at 1000MHz. I will try the powertune option, try the fan at 80% and then there's the optimized scrypt kernel files at https://litecointalk.org/index.php?topic=6058.0 that I can try. Anything else? I was hoping to get 1700Mh/s, maybe a bit optimistic, but 1500 should be realistic.

Another thing: right now, the PC is in my cellar, which is really cool (about 10C). But the case only came with one 8cm fan, so I think it's a bit short of fresh air. I'm planning to get at least a 12cm to mount at the from (push the airflow towards the graphical cards, exits through the 8cm, the PSU and all the covers of the expansion cards that I removed) and maybe another 8cm or 12cm to mount on the side of the case. That should further improve throttling. Maybe I can start pushing the GPU clock then.

A final weird thing: GPU0 is hotter and throttled, but it gets more than 10% more Kh/s than GPU1. How can that be? Both have the same settings.

Any idea or feedback is appreciated!

[PeterL]

Good news! I hit >1600 Kh/s, finally, by playing around with the concurrent threads. In the end, I set them to 22528, which is 8 times the number of shaders of an R9 290X. Here's the settings:

./cgminer --scrypt -o stratum+tcp://tag.hashfaster.com:3335 -u bla -p blo -I 20 --thread-concurrency 22528 --worksize 512 --auto-fan --auto-gpu --lookup-gap 2 --gpu-memclock 1500

I will still apply the kernel patch mentioned in my previous post. That should add a few extra Kh/s.

What I learned: decrease the thread concurrency until you start getting hardware errors. The lower, the better.

I will continue optimising, but not today.

[wirabhumi]

It's seem happen to me.

I got HIS R 290 (non X) last monday. It has 2560 shader, viewed by GPUz.
I spent last 3 days for tweaking and setting.
1st day i got 885 Khs, with --thread concurrency 32765 --gpu-engine 975 --gpu-memclock 1500 -I 20. It's rock and stable.
This card got 99,000 shared, WITHOUT Reject or stale. It has 70 C, and Fan only 65%.

You know in my country ( Indonesia ), It's pretty warm here. Ambient always @ 33-36 C.
My 7950 and R 280x always rise up to 78 C,  tough I 've undervolted them, 7950@0.965v, R280x @1.146v.

But the problem was occur. In my pool, hashrate goes up and down like a crazy. The lowest = 319 Kh, The Highest = 1.310 Mh.
Wide different, isn't it ?

Try to powertune +20, +30. there was no effect.
The effect only it will rise my electricity for 70w for powertune +30.
And temperature will rise 80 C <.

2nd day until now,
I try lower intensity, lowering Thread concurrency, trying downclock, update BIOS with the other brand. (ASUS, Sapphire, etc),
but it still no good side, rather than lowered hashrate @ 770 Khash.   

This R9 290 nice card, for performance and wattage.
But for pool hashrate instability, it's still became our research.

Thanks for this great forum.

[wirabhumi]

I am using PeterL calculating.
I change Thread Concurrency. Shader multiply by 8. So, 2560 x 8 = 20480,
Same Intensity, same gpu hardware config. and no gpu-powertune.

For 1 hours, It seems more stable,
The lowest hashrate goes low min= 700 KH, goes Up max 1.01 MH.
I' ll watch it, and report soon.

[PeterL]

About the pool hashrate: it's calculated based on what the pool receives. So if you are lucky and your cards find a lot of shares, then the pool hashrate goes up. With the same cgminer hashrate, I've had a poolrate of 50%-250% of what cgminer said.

I just ran the two 290Xs a night with the following settings:

./cgminer --scrypt -I 20 --thread-concurrency 22528 --worksize 1024 --lookup-gap 2 --auto-fan --gpu-memclock 1500 --auto-gpu

That worked perfectly stably at about 1630Kh/s.

I started playing with the engine clock:
975MHz: 1640Kh/s! CPUs started throttling.
950MHz: 1650Kh/s! Less throttling.
925MHz: back to 1640Kh/s again. 950 seems ideal.

Then I wanted to get rid of the throttling.  Setting the fan speed to 0-100 helped: the temperature decreased from 85 to 80C. I installed 3 additional 12cm fans and told them to blow as fast as they can in the BIOS, but their impact is limited (the GPU card fans blow way harder, so the 12cm fans can't get the air in quickly enough).

I played around a bit with thread concurrency and worksize, but this doesn't seem to have too much of an impact.

Next up was playing with the memory speed:

1600MHz: also works fine, but goes slower at just 1430Kh/s! Very strange.
1700MHz: card 2 acting strangely and a crash
1650MHz: works well, but again, not that fast.

So, if going up decreases speed, does a slower memory speed give better performance?

1200MHz: no, just about 600Kh/s per card. Maybe a bit too drastic.
1300MHz: idem!
1400MHz: OK, back up to about 650Kh/s
1450MHz: 750Kh/s...
1475MHz: 800Kh/s...
1525MHz: back down to 600Kh/s! Very strange.

I also tried 1510 and 1490 and concluded that 1500MHz is just perfect for my cards. So my final settings are:

-I 20 --thread-concurrency 22528 --worksize 1024 --lookup-gap 2 --auto-fan --gpu-fan 0-100 --gpu-memclock 1500 --gpu-engine 950

This gives me about 1650KH/s with two boards, so 825Kh/s each. And it's very stable just below 80C, here at room temperature. Time to move the PC back to the cellar where it's cooler

Thanks for the response and the tips!