Pages:
Author

Topic: Hacking The KNC Firmware: Overclocking - page 39. (Read 144343 times)

newbie
Activity: 3
Merit: 0
January 24, 2014, 11:25:06 PM
Quote
"What temps do you get on the chip itself? From the web interface and from your readings?  And what hash rate do you get per board?
If there is a better way to cool the down side of the VRMs, we'd be able to achieve better performance, any ideas?"

@Redbluebg
Well OC 231 in the ~23C ambient on stock settings WI ASIC temps are below 48C ( w/ ~175W DC, ~165 GH/s per ASIC, 1.23W per GH/s  and ~2.5% Reject + HW ...)

At this time I think forced air is most practical method to cool the VRMs. And it would be good to keep them dust free with brushes and pressurized air. Also keep/aim away the PSU so it doesn't blow hot air near the miner. I am direct driving with the PSU (HX750) cables to save on power loss and heat generation on the miner cables.

Immersing VRMs in oil bath (in synthetic Mobil1  0-20W Grin) may be an alternative for some of us Roll Eyes but not sure if it will accomplish much...

BTW, I thank to all those who help opening the KNC OC doors to us Grin


hero member
Activity: 784
Merit: 504
Dream become broken often
January 24, 2014, 10:27:56 PM
well i did put heatsinks on my vrms...IDK how it lowered the temp but it did...I saw 5c or so lower temps w/o changing anything...i try the 45 degree fan method too..but i was doing that before n after heatsinks so that wasn't the change in temp...and one module didn't even have a vrm fan on it so just with the heatsinks it droped...

Think i might try removing the stock hsf as i saw you can get a better fan placement on there...i just have mine proped up on the black shroud



hero member
Activity: 560
Merit: 500
January 24, 2014, 07:38:01 PM
45 degrees fans = best performances and best hashrate. thanks ! how many watts ( psu )251 setting requires ?
newbie
Activity: 9
Merit: 0
January 24, 2014, 05:56:40 PM
Here is how I have moved my fans, their are strapped with zip lock to the PCIe cables, temperature dropped with almost 20 degrees, now running under 40 degrees overclocked to 211, getting 485 GH/s. 231 and 251 are not working very good for my miner

http://i.imgur.com/yXdBhzT.jpg
http://i.imgur.com/aXkBuLb.jpg


Below are the images from OCT 4 VRMs with fans blowing about 45 Degree to the boards like Redbluebg doing. Running 231 with stock settings...

http://i.imgur.com/fLwkotI.jpg
http://i.imgur.com/jFLMBXt.jpg

Based on these images I am not sure if adding heatsinks to VRMs be really beneficial. There is a danger of reducing the air flow in areas really needs cooling.

Does anyone know where is the temperature sensor for the ASICs?






1BFkmxwdjH4PM8X2YUK8HTuN1nnQn4bwct



What temps do you get on the chip itself? From the web interface and from your readings?  And what hash rate do you get per board?
If there is a better way to cool the down side of the VRMs, we'd be able to achieve better performance, any ideas?
newbie
Activity: 3
Merit: 0
January 24, 2014, 05:17:18 PM
Here is how I have moved my fans, their are strapped with zip lock to the PCIe cables, temperature dropped with almost 20 degrees, now running under 40 degrees overclocked to 211, getting 485 GH/s. 231 and 251 are not working very good for my miner

http://i.imgur.com/yXdBhzT.jpg
http://i.imgur.com/aXkBuLb.jpg


Below are the images from OCT 4 VRMs with fans blowing about 45 Degree to the boards like Redbluebg doing. Running 231 with stock settings...

http://i.imgur.com/fLwkotI.jpg
http://i.imgur.com/jFLMBXt.jpg

Based on these images I am not sure if adding heatsinks to VRMs be really beneficial. There is a danger of reducing the air flow in areas really needs cooling.

Does anyone know where is the temperature sensor for the ASICs?






1BFkmxwdjH4PM8X2YUK8HTuN1nnQn4bwct

sr. member
Activity: 462
Merit: 250
January 24, 2014, 04:50:21 PM


It is minus 20 degree from overclocked, before my temps were around 57-60 now there are 37 to 40, both times with the same config 211. now the whole box feels cooler, the vrms are cooler, and the whole radiator is cooler, I think the readings are real.

nice, when I did something similar it didn't seem to do well without the shrouds is why I put them on the crossbar
maybe you have it at a better angle than I did

mine run pretty cool as well but 20c is a big drop
newbie
Activity: 9
Merit: 0
January 24, 2014, 03:56:46 PM
Here is how I have moved my fans, their are strapped with zip lock to the PCIe cables, temperature dropped with almost 20 degrees, now running under 40 degrees overclocked to 211, getting 485 GH/s. 231 and 251 are not working very good for my miner



not sure on those temp readings.. I saw without the shroud you get less drop and tilting down kept more heat on the boards

are you 20 degrees less than stock config?


you might also be pointing directly onto the temp sensor.  I forget where that is but if it is working for you don't move it !

It is minus 20 degree from overclocked, before my temps were around 57-60 now there are 37 to 40, both times with the same config 211. now the whole box feels cooler, the vrms are cooler, and the whole radiator is cooler, I think the readings are real.
member
Activity: 119
Merit: 10
January 24, 2014, 03:12:09 PM
@pedrosoft: Closed case. I have repositioned those "case" fans to blow to asics boards (leaning in front of asic fan), heatsinks etc. on vrmr's. Read my previous mails=) Now I'm having 471 GH/s with 2 modules. I have a feeling that the asic doesnt even get so hot, voltage regulator are what heats the whole thing.

Sometimes vrm:s start to dropt their amps to 24 amp, once the whole asic 0 was down to 24 amps, and hashing 415ghs. Stopped the miner, waited for vrms to cool and started again.

If you want to overclock, you cant start the miner on hot VRMs. You'll just get fault 96/97. VRMs needs to be cool when starting the miner when overclocking! If they are not your miner will settle for lower hashrate for rest of the session or those VRM-faults.



sr. member
Activity: 462
Merit: 250
January 24, 2014, 01:34:02 PM
Here is how I have moved my fans, their are strapped with zip lock to the PCIe cables, temperature dropped with almost 20 degrees, now running under 40 degrees overclocked to 211, getting 485 GH/s. 231 and 251 are not working very good for my miner



not sure on those temp readings.. I saw without the shroud you get less drop and tilting down kept more heat on the boards

are you 20 degrees less than stock config?


you might also be pointing directly onto the temp sensor.  I forget where that is but if it is working for you don't move it !
legendary
Activity: 1260
Merit: 1008
January 24, 2014, 01:33:07 PM
Ideas on cooling

I put air obstructions in my case so all air must pass the ASIC and VRM's.

For Sat find/make a cardboard box that fits between front (towards BBB) ASIC modules.
For Jup also find/make 2 smaller boxes to fill up outside edges.

Also the air that passes over the aluminum plate is not doing much cooling I direct it towards the VRM's.

Painters masking tape or similar wrapped around the heat pipes below the aluminum radiator will direct more air to the VRM's.

You could shove a wadded paper towel in there for testing.
Beware cardboard/paper burns, might not be ideal for long term usage in a box with a history of sparks.

Folks who drop fans to bar might consider blocking top(no fan area) of aluminum radiator also.

Aluminum tape/foil might be a better choice but then beware risk of shorts, keep it neat!

Make sure the cables do not block airflow also.

Oct. GE VRM's are designed for airflow from ASIC side to edge of board side.
KnCMiner is not ideal for OC airflow but it is what it is.
I'm glad they chose shortest lowV/highA copper runs making OC possible!

YMMV
Smiley

Care to share same pics?
newbie
Activity: 9
Merit: 0
January 24, 2014, 01:26:50 PM
Here is how I have moved my fans, their are strapped with zip lock to the PCIe cables, temperature dropped with almost 20 degrees, now running under 40 degrees overclocked to 211, getting 485 GH/s. 231 and 251 are not working very good for my miner

http://i.imgur.com/yXdBhzT.jpg
http://i.imgur.com/aXkBuLb.jpg
sr. member
Activity: 386
Merit: 250
January 24, 2014, 01:15:45 PM
Ideas on cooling

I put air obstructions in my case so all air must pass the ASIC and VRM's.

For Sat find/make a cardboard box that fits between front (towards BBB) ASIC modules.
For Jup also find/make 2 smaller boxes to fill up outside edges.

Also the air that passes over the aluminum plate is not doing much cooling I direct it towards the VRM's.

Painters masking tape or similar wrapped around the heat pipes below the aluminum radiator will direct more air to the VRM's.

You could shove a wadded paper towel in there for testing.
Beware cardboard/paper burns, might not be ideal for long term usage in a box with a history of sparks.

Folks who drop fans to bar might consider blocking top(no fan area) of aluminum radiator also.

Aluminum tape/foil might be a better choice but then beware risk of shorts, keep it neat!

Make sure the cables do not block airflow also.

Oct. GE VRM's are designed for airflow from ASIC side to edge of board side.
KnCMiner is not ideal for OC airflow but it is what it is.
I'm glad they chose shortest lowV/highA copper runs making OC possible!

YMMV
Smiley
newbie
Activity: 31
Merit: 0
January 24, 2014, 12:38:12 PM


Quote

Accepted:159374  Rejects:2242  HWHardWare Errors:4403
2242/159374*100=1.4% rejects
4403/159374*100=2.76% errors

1.4+2.76=4.16% total

304*4.16/100=12.65GH/s

304-12.65=291.35GH/s

Ah, we were doing the same thing at the same time. The second part of the figuring is correct (but I didn't include it), but remember that this is for my current settings at 211, to which I reverted this AM after the big discrepancy I was finding at 251. I'll try again and test. Thanks!
legendary
Activity: 3220
Merit: 1220
January 24, 2014, 12:34:42 PM
Quote
In the same way as rejections.
Out of the Accepted, x% were rejected and x% were hardware errors.

Ah, thanks.
So of my miner's total work:

1.406% are rejections
2.76% are HW errors
=4.166% of empty work.

The above is based off my current 211 configuration, but this difference would not cover the much wider gap of machine reporting versus poolside of my previous 251. So, I think the test should be extended to 231 and then 251. Perhaps I'll try temus's latest configuration, which is different altogether.

What is WU, by the way?
Thank you.



From https://github.com/ckolivas/cgminer/blob/master/README

Code:
The output line shows the following:
(5s):1713.6 (avg):1707.8 Mh/s | A:729 R:8 HW:0 WU:22.53/m

Each column is as follows:
5s: A 5 second exponentially decaying average hash rate
avg: An all time average hash rate
A: The total difficulty of Accepted shares
R: The total difficulty of Rejected shares
HW: The number of HardWare errors
WU: The Work Utility defined as the number of diff1 shares work / minute
     (accepted or rejected).

 BAS 1: max 67C 3.27V | 62.29G/62.19Gh/s | A:140813 R:256 HW:2860 WU: 852.0/m

Each column is as follows:
Temperature (if supported)
Fanspeed (if supported)
A 5 second exponentially decaying average hash rate
An all time average hash rate
The total difficulty of accepted shares
The total difficulty of rejected shares
The number of hardware erorrs
The work utility defined as the number of diff1 shares work / minute

The cgminer status line shows:
 ST: 1 SS: 0 NB: 1 LW: 8 GF: 1 RF: 1

ST is STaged work items (ready to use).
SS is Stale Shares discarded (detected and not submitted so don't count as rejects)
NB is New Blocks detected on the network
LW is Locally generated Work items
GF is Getwork Fail Occasions (server slow to provide work)
RF is Remote Fail occasions (server slow to accept work)

The block display shows:
Block: 0074c5e482e34a506d2a051a... Started: [17:17:22] Best share: 2.71K

This shows a short stretch of the current block, when the new block started,
and the all time best difficulty share you've found since starting cgminer
this time.
newbie
Activity: 31
Merit: 0
January 24, 2014, 12:31:05 PM
Quote
In the same way as rejections.
Out of the Accepted, x% were rejected and x% were hardware errors.

Ah, thanks.
So of my miner's total work:

1.406% are rejections
2.76% are HW errors
=4.166% of empty work.

The above is based off my current 211 configuration, but this difference would not cover the much wider gap of machine reporting versus poolside of my previous 251. So, I think the test should be extended to 231 and then 251. Perhaps I'll try temus's latest configuration, which is different altogether.

What is WU, by the way?
Thank you.

hero member
Activity: 574
Merit: 500
1.21 GIGA WATTS
January 24, 2014, 12:30:02 PM
cgminer version 3.9.0 - Started: [2014-01-24 15:25:51]
--------------------------------------------------------------------------------
 (5s):304.4G (avg):311.1Gh/s | A:159374  R:2242  HW:4403  WU:4346.0/m
 ST: 2  SS: 0  NB: 8  LW: 176865  GF: 0  RF: 1
 Connected to stratum-lb-usa48.btcguild.com diff 128 with stratum as user qweqweqweqweqwe_qweqweqweqweqwe34
 Block: 8c5d58f4...  Diff:2.19G  Started: [16:02:36]  Best share: 231K
--------------------------------------------------------------------------------

 [P]ool management [ S ]ettings [ D ]isplay options [Q]uit
 KnC 0:                | 305.7G/311.4Gh/s | A:159502 R:2242 HW:4407 WU:4350.0/m

--------------------------------------------------------------------------------


Accepted:159374  Rejects:2242  HWHardWare Errors:4403
2242/159374*100=1.4% rejects
4403/159374*100=2.76% errors

1.4+2.76=4.16% total

304*4.16/100=12.65GH/s

304-12.65=291.35GH/s
legendary
Activity: 3220
Merit: 1220
January 24, 2014, 12:23:17 PM
R: is rejected, and HW: is hardware errors. A: is accepted


A:159374
R:2242
= 1.406% rejections

How are HW errors quantified? How do I make sense of the 4403?

Thank you.

In the same way as rejections.

Out of the Accepted, x% were rejected and x% were hardware errors.
newbie
Activity: 31
Merit: 0
January 24, 2014, 12:21:10 PM
R: is rejected, and HW: is hardware errors. A: is accepted


A:159374
R:2242
= 1.406% rejections

How are HW errors quantified? How do I make sense of the 4403?

Thank you.
legendary
Activity: 3220
Merit: 1220
January 24, 2014, 12:17:14 PM
R: is rejected, and HW: is hardware errors. A: is accepted
newbie
Activity: 31
Merit: 0
January 24, 2014, 12:04:42 PM
Quote
You would have to get the information from cgminer on the device.

So either SSH onto the box or use the API and a third part App, or use Bertmod and it will show the Hardware errors (HW) on the web page.

Thank you. I'm able to SSH in, but I'm not clear on how to view errors once there. The "screen -r" view is this (keep in mind I just went down to 231, so the hashrate is reporting this stepped down as opposed to my original comments):

 cgminer version 3.9.0 - Started: [2014-01-24 15:25:51]
--------------------------------------------------------------------------------
 (5s):304.4G (avg):311.1Gh/s | A:159374  R:2242  HW:4403  WU:4346.0/m
 ST: 2  SS: 0  NB: 8  LW: 176865  GF: 0  RF: 1
 Connected to stratum-lb-usa48.btcguild.com diff 128 with stratum as user qweqweqweqweqwe_qweqweqweqweqwe34
 Block: 8c5d58f4...  Diff:2.19G  Started: [16:02:36]  Best share: 231K
--------------------------------------------------------------------------------

 [P]ool management [ S ]ettings [ D ]isplay options [Q]uit
 KnC 0:                | 305.7G/311.4Gh/s | A:159502 R:2242 HW:4407 WU:4350.0/m

--------------------------------------------------------------------------------
Pages:
Jump to: