Fury/Blizzard tuning and mods

[happydaze]

Hash rate is coming back up.  I'll let it run overnight at 342 clk.  340 clk might actually be the number for this unit until I get some thermal paste.  Your numbers look really good.

[KaTXi]

Do you know if the speed of the fan can be controlled by software (I guess not). Have anyone changed the fan for a good and silent one already? Does it make a diference in accoustics and temps?

[happydaze]

Mine is there with my Gridseeds so I hear them.  I think the Fury is more quiet.

[nst6563]

I've left the fan alone on my fury.  It's a lot quieter than the gridseeds fan and moves a good amount of air. 

[happydaze]

4.3   Readcount Timeout
After we send one command packet to the chained chips, that BIG chip will cover all the 32 bits nonce in a certain TIME. So we must send a new work to it before that, otherwise its hash power is wasted.

Cgminer’s readcount timeout is used for that. Every count means 0.1 second. We now set the readcount to be 3/8 of the covering time. It may be 3/4 later.

I'm not a coder at all but like to look through the driver file to get a vague idea of what is going on.  very vague idea  

Anyway, it looks to me they chose 3/4 in the current driver.  I'm still not sure of the affect.  Just guessing here but I think we'll see reduced errors as the dev's pick apart and improve the code.  There are a lot of variables to balance:  the number of chips, the difficulty of the work, the timing to send work to the pool so it is not stale and the timing of getting new work to the chips so they're not idle.

The current software is one size fits all, 6 chips or 96 chips, same software.  We might need a spinoff branch just for the 6 chip miner.  It would be great if we could get per chip tuning for the 6 chip miners.  

[nst6563]

I don't know C myself...I can read through it and get an idea of what's going on, but I don't know squat about coding with it.  I'm a vbscript/vb.net/powershell scripter by trade and that doesn't really help in the crypto world.

I could try changing that value in the driver and recompiling.  If I have time this evening I may do that just for kicks.  Like I said though....I'm not a C coder so for all I know we'll end up with cold fusion in a jar when I'm done with it.



:EDIT:
Actually...I couldn't wait so I opened the driver file and found this at the top....I'm wondering if changing this value would have any effect...

Code:

// The serial I/O speed - Linux uses a define 'B115200' in bits/termios.h
#define ICARUS_IO_SPEED 115200

[ZiG]

I don't know C myself...I can read through it and get an idea of what's going on, but I don't know squat about coding with it.  I'm a vbscript/vb.net/powershell scripter by trade and that doesn't really help in the crypto world.

I could try changing that value in the driver and recompiling.  If I have time this evening I may do that just for kicks.  Like I said though....I'm not a C coder so for all I know we'll end up with cold fusion in a jar when I'm done with it.



:EDIT:
Actually...I couldn't wait so I opened the driver file and found this at the top....I'm wondering if changing this value would have any effect...

Code:

// The serial I/O speed - Linux uses a define 'B115200' in bits/termios.h
#define ICARUS_IO_SPEED 115200

Hey nst6563,

This is the COM PORT SPEED...

Nothing to do with chip's hashing speed...don't touch it...

Cheers,
ZiG

[happydaze]

340 clk seems to be the max with my stock Fury.  HW errors almost double pushing it to even 342 clk.

I'll stop by a computer place today to see what thermal paste they have in stock.

[nst6563]

I don't know C myself...I can read through it and get an idea of what's going on, but I don't know squat about coding with it.  I'm a vbscript/vb.net/powershell scripter by trade and that doesn't really help in the crypto world.

I could try changing that value in the driver and recompiling.  If I have time this evening I may do that just for kicks.  Like I said though....I'm not a C coder so for all I know we'll end up with cold fusion in a jar when I'm done with it.



:EDIT:
Actually...I couldn't wait so I opened the driver file and found this at the top....I'm wondering if changing this value would have any effect...

Code:

// The serial I/O speed - Linux uses a define 'B115200' in bits/termios.h
#define ICARUS_IO_SPEED 115200

Hey nst6563,

This is the COM PORT SPEED...

Nothing to do with chip's hashing speed...don't touch it...

Cheers,
ZiG

I know it's the com port speed and nothing to do with hashing speed, my thought behind changing it was for the error rate.  Since each nonce returned has to pass through each chip in the chain before being passed to the controller perhaps raising the communication speed to a higher value could reduce any possible bottleneck in data movement.  Not likely, but even if it made a .5% difference it would be an improvement.
In the driver, it's hardcoded to 8 cores per chip, but the number if chips is variable.  So I'm not so sure a separate version for the 6-chip blizzard/Fury's and the higher chip count devices would be required.    Much like the gridseed 5-chip and 80-chip.  

[happydaze]

I was thinking sandors autotune ..............

You can specify multiple devices, but the length of the chips array must be equal to the number of chips on the GC3355 miners, Blades have 40 chips but you can only address chip0-7 (clusters of 5 chips), so the max is 8.

and

You cannot set the frequency of an individual chip on your G-Blade.

Autotune and a Fury / Blizzard would probably give a nice hash rate gain.

[happydaze]

Looks like two BFGMiner builds for Windows available for Zeus chip devices.

I've been using this one since it was released: https://litecointalk.org/index.php?topic=16301.msg180380#msg180380  It works well for me.  Ignore the HW number, instead look at HW %.

This one was released today and may still be being worked on: http://sgminer.builders/  I haven't tried it.  Discussion topic: https://hashtrader.com/t/bfgminer-fork-supports-for-zeus-miners/354/4

Edit:

3rd version now out.  BFGMiner 4.1 https://litecointalk.org/index.php?topic=16301.msg181331#msg181331

[schnauzr]

Has anyone gotten any mining software to work with IFMI pool manager and gotten the individual Fury statistics to actually show up on the status overview webpage?  I ask because I'm currently using starminer and while the overall hashrate does show with the oldass cgminer version that's been released by Zeus for these ASICs, it doesn't work 100% with IFMI.  This is what it looks like:

[nst6563]

I used starminer for a while and switched to Hashra's Lunar Lander image.  I only have one Fury right now, but the reasons I had for switching away from starminer were:
Slow interface.  It seemed to take forever to save a change or switch to different pages within it.
Not very intuitive in my opinion.  At first changing settings seemed to be obscure at best.
  It seemed that my miner would just stop mining for no apparent reason.  It would show the pool as dead, but my gridseeds running on a pi with Minera were all connected to it and running fine.

I've been using Hashra for most of a week now and compared to starminer it's:
Faster - overall.  no delays in changing settings or switching pages or anything else.  
Stable - I haven't seen it yet where it just says the pool is dead while everything else is still connected and mining to the same pool from the same network.
Cleaner more intuitive user interface.


Just my opinion though.  My suggestion would be to try them all.

If anyone has a link to a new image or system I'll add it to the first post (I added your's already happydaze).

Edit:  it looks like starminer is being actively worked on and the interface latency issues should be resolved with an upcoming release.

[schnauzr]

Does the hashra software show statistics for the individual Fury's on the web UI?

[nst6563]

I would assume so yes.  There is a section for current hashrate and average hashrate, and then below there is a section for individual miners.  I should have another fury later in the week so I'll know for sure, but check out the screenshots I posted on page 1 of this thread. 

[suchmoon]

Does the hashra software show statistics for the individual Fury's on the web UI?

Yes, it does show the circular chart for each, hashrate, A/R. I ran it with 4 Blizzards for a while.

[schnauzr]

Does the hashra software show statistics for the individual Fury's on the web UI?

Yes, it does show the circular chart for each, hashrate, A/R. I ran it with 4 Blizzards for a while.

Cool thanks for the info

[happydaze]

1,376 kH/s 24hr avg is where I'm at with 340 clk
A:8003 R:21 HW:5.8% for the current run so far.

I did get some thermal paste but might wait for my second Fury.  I don't want to stop this one working!

[nst6563]

340 is better than 328

I think I found the sweet spot on my Fury.  I edited the index.php for the Hashra Lunar Landar image to allow speeds higher than 350. I didn't need to go far though. 
At 355 I get an error rate of around 6%.  However, at 354 my error rate so far is at 2.95%...it's only been running for about 2 hours now, but in that same time at 355 I had an error rate of over 6%.  I'm on manicminer pool where you can set the difficulty and I have it at 304.

I'm going to let it run until tomorrow at 354Mhz, 304 diff and see what the numbers look like.

What's your ambient temp?  I have all my mining stuff in the basement where it stays between 70f-73f.  Slightly cooler in the winter (around 68f).

I know what you mean on letting it run....I'm waiting for the other Fury to show up so I can take the voltage readings all over the board.  I'll have a huge update after it comes in.  I just wish I had a good camera that I could get a decent high res macro shot of it. 

[BRADLEYPLOOF]

So, is there a version of CGminer that runs these?  I see CKolivas is a moderator...