896 mh/s firmware release - Butterfly Labs

[JWU42]

Jealous - 5/6 of mine won't even run with 832 

That is within their 832 +/- 10% though

[Fefox]

Jealous - 5/6 of mine won't even run with 832 

That is within their 832 +/- 10% though

My room temp is 63f im sure if it was 75+ this wouldn't be so nice...


so far with the 896 firmware im getting a solid 888 Mh using bitminter.

one miner is 54c and the other 51c same temps as the two running 880 firmware +-.5c

[JWU42]

Yep - 63F room temp is a winner 

That will be me come winter...

That said, my units throttle well before the temp in cgminer hits 53C.  I do have one that runs warmer - 57C now - but is stable with 816 FW.  Avg temps from all units around 51-53C

[Cablez]

Yeah I have a range with the units.  I have 2 that will accept the 896 and the others can only do 880.  

[BTC-engineer]

Thanks for the new firmwares.

I've only tested it on one of my units before I will test it on the others too.
I took my 'best' unit into the test and skipped the 880 firmware and directly run the test with the 896 firmware.

My room temperature is ~26°C (~79°F) - no A/C or room fan.

My unit (with improved airflow/cooling) is running @ 896 without any problems. The reported unit temperature is still below 40°C.
I'm sure the test will run through the night without any following problems.  Tomorrow I will test the other unit too.

From my extrapolation based on my measurements and empiricism I expect that a 960 firmware could also run on this unit. 
So please release additional firmwares with higher clockrates too. I think it's enough to count up in 16 or 24 steps.

[JWU42]

Below 40C and 79F room temp - something isn't correlating...

[BTC-engineer]

Below 40C and 79F room temp - something isn't correlating...

It's correct 26°C room temp and 39°C reported unit temp.
But it's my best unit, the others have a larger temp delta.
As I said before, airflow/cooling was improved by myself...

[SgtSpike]

Excellent, going to try it on all four of mine now.

[JWU42]

As I said before, airflow/cooling was improved by myself...

What exactly did you do? 

Even my units with 768 firmware are at 48.6C in a room around 78-80F

[SgtSpike]

Well, 3 of the four were throttling and/or had inconsistent hashrates, so I'm just going to put them all back to 872 for the time being.

[rjk]

Don't forget to restart the devices with a power cycle after the firmware flash. I keep forgetting to do so, and no wonder it's getting the same exact speed now as it was before the flash.... doh.

[kano]

Don't forget to restart the devices with a power cycle after the firmware flash. I keep forgetting to do so, and no wonder it's getting the same exact speed now as it was before the flash.... doh.

Pity it's not like the ztex ... you simply tell the firmware what speed to run at with a command (while it's running)
In that case you keep increasing it until you get HW: errors and then step it back down.

[rjk]

Well shoot, it seems that 880 is the max for mine. Odd thing is that it gives errors without throttling, but cgminer doesn't report them. So cgminer makes things look like all is well, but in fact the device is returning bad results, and HW: remains at zero but share submission rate tanks.

[yjacket]

I don't think proper cooling has everything to do with the firmware working. I have plenty of cooling (never goes over 46) and my singles don't like any firmware over 832. They lock or have a sub 800 rate with anything higher than 832.

Anyone else have a similar experience?

[rjk]

I don't think proper cooling has everything to do with the firmware working. I have plenty of cooling (never goes over 46) and my singles don't like any firmware over 832. They lock or have a sub 800 rate with anything higher than 832.

Anyone else have a similar experience?

I'd tend to agree. I've seen cgminer report over 70 degrees with no throttling, but it still won't run the fastest firmware. I wonder what could make the discrepancy so large.

[SgtSpike]

It's kind of like overclocking a CPU.  You can have a CPU running at 25c, but won't overclock any further than 4 GHz regardless.  It doesn't have to do with temps at that point, just with what the CPU is capable of.

[BFL]

Yes, you're exactly correct.  These are full wafer yield, so there's a range of variance from chip to chip.

We sell the end product at their lowest common denominator performance which is 832mh/s +/- 10% @ 72f / 22c.  This is why you can go much higher in some cases (up to 1.05 gh/s which might explain our initial speed estimates), but can't be expected on all the units shipped.

[kano]

Well shoot, it seems that 880 is the max for mine. Odd thing is that it gives errors without throttling, but cgminer doesn't report them. So cgminer makes things look like all is well, but in fact the device is returning bad results, and HW: remains at zero but share submission rate tanks.

The HW: number is simple - any share returned by any device is checked (it is sha256'd)
If it is invalid, it is a HW: error, if it is valid, it continues down the chain of processing.

Thus a HW: error means the device's sha256 isn't functioning correctly somehow
In GPU terms, it usually means it's OC too hard.
As I mentioned above, with a ztex it means the MHz clock is up too high and needs to be stepped down a bit (that's what the cgminer code does)
But in both cases, the meaning is the same.

On the other hand, if USB device I/O returns an error, then of course that isn't considered a HW: error, since that is a USB issue.
Yes I've seen someone with 6 BFL's have them return USB errors regularly due to some problem with the USB configuration and the OS
But that is exactly that, a USB problem, not a HW: error (since HW: has that specific meaning and in this case the fix is not related to the device failing to sha256 correctly)

[rjk]

Is there some other metric that can be shown instead then, so that we can see when a share is invalid or not submitted due to throttle? HW stands for Hardware, as far as I can guess, so maybe you can have TH for throttle, and/or ER for error. I don't know whether they need different commands to detect or what, but EasyMiner shows separate counters for error rates and throttle instances.

[kano]

Is there some other metric that can be shown instead then, so that we can see when a share is invalid or not submitted due to throttle? HW stands for Hardware, as far as I can guess, so maybe you can have TH for throttle, and/or ER for error. I don't know whether they need different commands to detect or what, but EasyMiner shows separate counters for error rates and throttle instances.

I've asked BFL for details about throttling and what other commands there are that we don't know about.
Their reply was 'no'
They don't want better cgminer support for the BFL.
Vote with your wallet.