Topic: Butterfly Labs - Bitforce Single and Mini Rig Box - page 65. (Read 186946 times)

I think it's a problem when this happens :
BFL (add VAT to the price) -> customs (add EU VAT to the price) -> customer (WTF!? 2 times VAT?!)
It is possible to get a refund done for some part of the VAT, but I have no idea how exactly it must be done. Some stupid paperwork.

Coordial greetings make me angry!

I COULD FLIP A DESK IN RAGE (ノಠ益ಠ)ノ彡┻━┻

Throttling is heat-related. Should the environment get hotter than the acceptable
measure for the chip, then throttling will occur. There can be a degree or two difference in
throttle-threshold among different units (For example one unit may throttle at 32 Degrees
while the other throttles at 33 Degrees (Celcius) ) .


EDIT: Good Luck

The butterfly singles are excellent. Quiet little hashing cubes making bitcoin. 

I know it's throttling because the front light flashes intermittently when the unit throttles. I observed this happening.
In spite of the throttling the numbers are good.
You feel better about the numbers, but I feel worse.  I guess what was bothering me was Sunday, the U was 5.5-6, but cgminer was showing the usual ~815 MH/s. People need to keep an eye on U

If you are looking at the detail screen, those numbers are pulled from average Mh using your api.  The summary screen is calculating MH/s using accepted shares for the full uptime period.

In future I'll keep in mind to give much more importance to the variance.

Thanks for the info fred.

Second box should be arriving in the next few days.   I still often think something's wrong when I walk in the mining room and it's so quiet

Yes, thanks for pointing that out.  I need to look into this further.  Yes, not mining at p2pool. Generally I don't place much importance on numbers less that 24 hours.

fred0 - those U figures vs the MH/s figure in the first screen are fine.
Nothing to think that there certainly is a problem somewhere.

If you had run it for a whole day I'd still say the U's look OK - the highest to the lowest is around ... 10%
On my one day figures (with lower hash rates - roughly half of that) I get up to about +/-10% (I can't remember ever seeing it more than +/-10%)

But the main point being - the U figure has variance exactly like finding a block does.
Since shares happen 1.5million times more often than blocks at the moment, the amount of time to get a reasonably close to expected U value is of course a lot less when mining 1-difficulty pool shares vs 1.5million difficulty blocks.

However, a few hours isn't very accurate even at 800Mh/s
On the other hand, if you have been watching the 5s MH/s figure, that would be all over the place
I'd actually consider the MH/s figures you've shown quite reliable.

Does cgminer show the hash rate correct for Bitforce? I'm not sure. The numbers do look OK though.
I still need to do some work on the Icarus code to make it more accurate ... and thus understand where problems with that number could be when I do that - then I guess I might look at Luke-jr's code also if I find issues with the Icarus code (and own a BFL to test it

Thanks for the additional info Fred.

Unfortunately, today was uncomfortably warm in the atlantic northeastern US. Even now, it's about 80 F where the singles are located.  The 5970s are about to get shut off, that should cool the room off. Even at these temps, well you can see that only one throttling. It's at 747 MH/s now (BFL1)

I actually have seen it a couple of times today, use cgminer by all means, but keep an eye on the U, that is what signaled the discrepancy to me.

I have run ufasoft to compare performance with cgminer, and really noticed no difference.  However, needless to say with all my testing, none has been the 24 hour performance testing, but 3 to 8 hours typically, which should be pretty good.

Let me reiterate, I made a sizable purchase of singles, I have seen nothing to dissuade me from purchasing more.

Defenders of the single, I'm not attacking performance, of cgminer or single, I will continue to use both.

However, mh/s in cgminer is not reliable. I have been calculating the average mh/s from the number of shares submitted, then comparing with the average mh/s in cgminer, the two numbers are always different.

I don't know enough to isolate where the problem lies.
Devices

Summary


Notice that the mh/s on the device screen are pretty consistent, and are in fact the numbers returned by api call from cgminer (average mh/s).

The summary calculates the average mh/s by from averaging all accepted shares over the full submission period.

I would think that these numbers should be very very close, but they are different, still close.  I don't know enough about the working of cgminer to argue the point.

The throttling is heat related, and it is minor.

+1, it's only been down twice, once when my USB hub died, and another time for a reboot. Aprox. 73 degrees Fahrenheit ambient. Ufasoft (for now)

I have seen no evidence of throttling with my single. Been pretty constant at 829Mh/s. I have 4 more coming tomorrow so I'll have my first 5Gh computer running in the next 24 hours. 

My BFL hashrate seem to be commensurate with my GPU hashrate in terms of U and A in cgminer, just for the record.

You should check the validity of what you say first

Yes the MH/s may be wrong in cgminer - go complain to Luke-jr for that if it is wrong - he wrote the function that returns the number each time

If you want to give me a BFL (I've entered the raffle) I'll fix it if it needs fixing

However, your U: is only representative of your hash rate.
Yes it is SIMPLE maths to convert that random U: number to a random hash rate: U * 2^32 / 60
However, U: is directly related to the pseudo-random function of generating a block (it is the same thing)

Firstly what should be my usual comment on this: ask Meni to do an analysis if you want to work out it's variance over time.
However, after a day of hashing I usually find it is within 10% but it often jumps around up to 5% even after a day of hashing.

However ... in general ... as I have said a few times ... if you are hashing on P2Pool with a BFL ... DONT.
Coz then your hash rate will be all over the place and you will be throwing away a lot of work.

Edit1: oh and the throttling is done by the unit itself, not cgminer (in case anyone wasn't sure)
(though cgminer will turn off the bitforce if the temperature gets too high - above 'cutofftemp' which is actually a GPU option that defaults to 95°C - assuming your version of cgminer can read the temperature)

Edit2: Yes TheOtherGuy it is exactly heat related

Edit3: the option is --temp-cutoff

Fred,

Can you give us more details on the throttling? Is it heat related?

Well, I like the units a lot, but to say there is nothing left to talk about but avoiding sales tax, I find a little extreme.  After all how many threads are there for optimizing gpus.  Why is the average mh/s always off in cgminer?

The other thing that I noticed was that shipments have accelerated, so I think that we will see a lot more in the wild.

Lastly, my remark did have a bit of sarcasm, but the thread title may as well be called All things BitFORCE, because really, when looking at the posts, that does seem accurate.

Wait, Fred...so these aren't manna sent from heaven?

Throttling.  Inaccuracies in MH/s reading in cgminer.  I'll start.

Boy that throttling is a PITA.  I have one unit that always wants to slow down.

Don't look at the MH/s reading in cgminer, they are wrong.  You can only calculate an accurate hash from the Util. number.  Ok all you single owners go do some math and see how fast you are really hashing.