BFL single - temp vs throttle

[vapourminer]

Ive noticed a lot of singles running in the mid to high 50's C. apparently without throttling. but mine throttles around 45 C.

so here are my questions. anyone know where the temp is read on a single? an average of the 2 chips? one chip? somewhere else on the board? what triggers the throttle? hardware errors? an internal temp thats not reported?

background: ambient 70 F (my basement).  originally I ran the 832 bitstream with no problems @ 44 C, but the 864 would throttle withing minutes, even with the temp never getting above 46. so.. I redid the paste with AC MX4 (the stock paste job was horrible, almost lumpy on one chip, and not much coverage on the other) and got a 2-3 degree C drop. loaded the 864 bitstream and it would run about 15 minutes and throttle, with the temp never getting about 43-44 C.

so

upshot is I now run it with a 140 mm fan against it and no sides and it runs at 40-42 with the 864  880 bitstream and no issues.

just curious, mainly.

[1714079364]

1714079364

[1714079364]

1714079364

[1714079364]

1714079364

[fortyniner]

I'm interested in these answers too.  My single is Rev E (shipped Aug 22) and came with 872 firmware.  Which was odd because it would throttle almost immediately.  I took the case off and got it stable with 832 firmware.  Watching the temperature in CGminer (often cycles up & down), I've not seen it go above 48.2C so I figure that must be about the throttle-temp-trigger.  However, if I load the 864 firmware and watch the temps, it appears that anything above 45C will trigger the throttle.

I should take the heatsink off and apply that Artic Cooling paste to see if that'll lower it any.

[freshzive]

I think it varies dramatically from unit to unit. Some seem to run fine at much higher temps than others. I sat all 4 of mine in the same exact spot and ran BFL's auto-tune package (which you should try) and got different max firmwares from all of them. In the end, I settled on 832 for three of them and 864 for one. Guessing this is due to slight physical differences in the hardware itself as well as the thermal paste/fans.

Here are mine:



You can see BFL 2 throttles occassionally (guessing mid-day when room temp goes up), but still faster than 816 firmware. These are in a closet, probably around 75F with a small fan blowing towards them.

Guessing I could get better speeds going the extra mile of removing the cases, adding better fans (and thermal paste), but overall I'd prefer not to mod them too much & I'm happy with the speeds. YMMV of course

[freshzive]

I'm interested in these answers too.  My single is Rev E (shipped Aug 22) and came with 872 firmware.  Which was odd because it would throttle almost immediately.  I took the case off and got it stable with 832 firmware.  Watching the temperature in CGminer (often cycles up & down), I've not seen it go above 48.2C so I figure that must be about the throttle-temp-trigger.  However, if I load the 864 firmware and watch the temps, it appears that anything above 45C will trigger the throttle.

I should take the heatsink off and apply that Artic Cooling paste to see if that'll lower it any.

What is your ambient room temp? This seems to have a dramatic effect on speeds. Also, having an external fan pointing at the units helps a lot.

Several of mine shipped with firmwares that weren't remotely stable in my dwelling, not sure why BFL doesn't use a standard (lower) firmware.

[n4l3hp]

Mine lowers the throttle threshold 2C for every higher speed firmware. (e.g. 800 = 58, 808 = 56, 816 = 54, etc, etc)

[fortyniner]

Autotune gave me 816 while it was still in the case. I agree, the temps at which these throttle vary widely from unit to unit.  Without a thermometer I'm unsure of my ambient temp.  With the windows open this morning, I noticed my single was running cooler so I tried the 864 firmware and learned that about 45C (or 45.5) seems to be the trigger on that one.  Here's a shot taken just before (or while) it throttled:

http://imageshack.us/a/img705/2486/cgminer.jpg

It appears n4l3hp is about right; tests on my single show throttle-trigger-temps of:
832 firmware - about 48C
864 firmware - about 45.5C
872 firmware - about 43C

[vapourminer]

I did try autotune. it seems too conservative for me, always went one speed step down from what I could actually run it at without throttles.

Mine lowers the throttle threshold 2C for every higher speed firmware. (e.g. 800 = 58, 808 = 56, 816 = 54, etc, etc)

that tracks with what Im seeing too. 880 bitstream runs 40-42 C with no throttles according to anubis and miner.php. tried the 896 bitstream and within one minute it throttled @ 40 C.

and threads I saw that mentioned running in the 50s C were usually running lower than stock speeds.

[dave3]

I'm running 2 BFL singles with the 808 firmware currently.  Ambient temperature is around 85°-90°F during the daytime, and cgminer reports the temperatures at 59°-62°C and they're not throttling.

Code:

 cgminer version 2.7.5 - Started: [2012-09-14 13:53:54]
--------------------------------------------------------------------------------
 (5s):1651.7 (avg):1585.7 Mh/s | Q:10607  A:47946  R:62  HW:0  E:452%  U:8.0/m
 TQ: 0  ST: 3  SS: 0  DW: 2453  NB: 658  LW: 233594  GF: 1  RF: 1  WU: 22.1
 Connected to http://us3.eclipsemc.com:8437 with LP as user ---
 Block: 0000028b8e6a63f106fc0868b5c62020...  Started: [17:29:57]
--------------------------------------------------------------------------------
 [P]ool management [S]ettings [D]isplay options [Q]uit
 BFL 0:  60.3C         | 803.7/792.8Mh/s | A:24170 R:29 HW:0 U: 4.04/m
 BFL 1:  62.0C         | 803.7/792.9Mh/s | A:23776 R:33 HW:0 U: 3.97/m
--------------------------------------------------------------------------------

I believe the throttles would show up under "HW" errors.

[vapourminer]

I believe the throttles would show up under "HW" errors.

they seem to, but you can also query cgminer via its API and get an official "overheat" count to be sure you dont miss any.

[PulsedMedia]

mine are running at 54-56c oO;
the default 832 firmware.

[n4l3hp]

Mine throttles every few minutes or so with 832 firmware at 50C. 

[vapourminer]

Mine throttles every few minutes or so with 832 firmware at 50C. 

ouch. is that what youre running now?

what is your ambient temp? have you cleaned the single of any dust build up? maybe redo the thermal paste? run it without the side cover?

all else fails, run a lower speed bitstream, it will give a higher long term hash rate.

[n4l3hp]

Ambient is about 34-39C at daytime, 25-29C at night.

It usually runs between 49-50C during the day but spikes up to 51C every few minutes between 11am to 4PM.

At night, runs at 46.6-47.8C but according to my digital thermometer, the metal plate on top of the chips are only at 36C.

I replaced the stock HS with a CPU tower HS w/ 2 copper heatpipes. Can run now at 816 during the hottest part of the day without problems, even 832 if its cloudy (can even run 864 when its raining   ). With the stock HS, can only run 800 max during the middle of the day (but throttles inside its case).

My x6500s are running at 37C, speeds ranging from 200-211 near the single using Thermalright chipset coolers.
http://www.thermalright.com/products/index.php?act=data&cat_id=8&id=107

[Morblias]

It's 53 degrees F out right now and I have 2 at 800 firmware. If one hits 43C it throttles. Kinda sucks, but I got 2 really nice ones.

53F outside with fan blowing in window onto singles:
896 firmware: 44C
880 firmware: 38C
800 firmware: 40C
800 firmware: 44C

The 800 firmwares cant be upgraded or they throttle, while the other 2 can hit 55C without throttling.

[davecoin]

Mine is humming along at 54 C on 832 firmware without any throttling.  Ambient temp is 60 F (15.5 C) right now .

[Askit2]

Since the trigger temp for throttling seems to happen towards the end of the work unit. I would assume and the last temp you get is at the end of the previous unit there is some variation. On my units this can be 1.2C higher at the end then the previous temp shown. Then back down the same amount. Maybe others temps are more stable.

The heatsinks seem less then optimal. The unit runs 58C on mine and it can handle most days 73F (22C plus a little). Exiting air temp with 864 bitstream is 90F (around 32C) over the heat pipes. Only 82F (28C) over the aluminum heatsink side. To me this says that less heat comes through the odd shaped heatsink then through the radiator like heatpipe section.

I got a semi Lucky unit. After some work I can run with only occasional throttles at 72F with the 864 bitstream. I have no idea why the >10C discrepency between units running the same firmware.

To be honest I had hoped autotune would adjust clockrates automatically based on temp. The time to reflash is longer then a handful of throttles add in testing and all and you can be waiting a long time to get back to making money minus throttles. As an example I can run 864 but during the summer the room this unit is in can exceed 73F. I can limit it by not playing xbox etc. I could downclock to 848 (given to me by BFL because I can not use 832(Power is wrong on one FPGA.)). But by night time the room can be in the mid 60s again. Back in the sweet spot for 864. Sure I would have loved to miss the 10-30 throttles a day but my average rate was at 848. Adjusting down manually would cost me speed and time. Even if everything went smoothly it took me usually 1 minute to flash the unit. That is 4 throttles. Now anything above 8 throttles would give me more speed but from one day to the next I couldn't be sure what one would be needed or for how long or when precisely. Had my unit at getting to say 57.8C clocked to 848 The temp would immediatly start down plus my tolerance would be some higher. Then at say 56C it could cock back to 864. To be very honest the unit likely could handle higher then 864 at night for several hours. Using a stair step logic the unit could every unit try to get the best performance possible.

I am hoping they have concidered this for their next stage of hashing units. If the asics could soft throttle and try to average to some speed it would likely be better for the end user. I do see one sad point their autotun usually ran a slower bitstream then the unit could run maximally. To me a small drop in speed because of higher heat (ambient) makes more sense then a flat out stop for 10 seconds to cool it down.

Sorry some is OT. I really think the variation is a tad excessive personally.