Topic: [ANN] Spondoolies-Tech - carrier grade, data center ready mining rigs - page 465. (Read 1260350 times)

Mine was getting 4.5 TH/s while it was being tested in Israel. It's now on two separate 120V 20A circuits. Shortly after I received it, I had it running at 4153 GH/s using 2672 W long enough to grab a stats dump, and I think I saw it go up to about 4.2 or 4.3 TH/s before I ran into a bug. This bug limited me to about 3.4 TH/s. Zvi provided a new firmware with a workaround in firmware 2.23 within about 12 hours of me reporting it to him. This firmware has hard-coded limits of 1080W DC and getting 4.0 TH/s (3967 GH/s precisely). Once Zvi releases the actual fixed firmware, I expect to be able to improve my hashrate even farther. I think 4153 GH/s should be pretty straightforward for mine. Mostly, I'm just doing this experimentation on 122V out of curiosity, since I'll have 500 kW of ~250V up within a week or two.

As for the 220/230/240V discussion, all three of those voltages are found in both North America and Europe. My brother's house in Seattle has 115/230V, for example, whereas our datacenter warehouse came with 480V and two indoor dry transformers which provide nominal voltages of 120/240V (which I'm using) and 110/220V (which the other suite uses).

Converting from two 120V outlets on different circuits in a 120/240V split phase system (typical for North America) to one 240V outlet as Adam suggested would be physically possible. However, I have never seen such a device before, and it would be tricky to use properly. Anyway, _{it's probably easier and cheaper to just host your miners with me.}

new firmware update for SP10 !

Zvisha you are the Jedi of the bitcoin's manufacturer !

edit : +10 Gh/s with the last update and it seems that the consumption is lower ?

By well over you mean what price? $10k?

$3895 plus shipping will be well over $4000 for 4.5TH/s in September, mid/late September since it is batch #2 is a pretty high price no?

Spondoolie-Tech didn't "have" to make the SP30 a pre-order. They could have made the product, like what other companies do, get an inventory, have solid specs and a working product, then sell it normally. It is the manufactures decision to use your pre-order money for R&D and making the product, but you'll never really know what you will end up with since it hasn't been made yet. ALL bitcoin manufacturers have the miners, "by the balls" in this industry now. Buying an expensive bitcoin mining rig(s) are worse then buying a car. At least you can sell the car at a later time for something.

Yeah, 65db is a bit loud for home use, but I don't care for data center use.

If the SP10 and SP30 have about the same noise levels, I'd imagine the SP30 would be a bit less of a whine. SP10 has 6x 40mm fans, plus one PSU fan. SP30 has 4x 80mm fans, plus two PSU fans. I would expect the case fans will be a little less whiney, but the PSU will obv be more noise.

Come winter, it gets very cold here in the north east, and it's very easy to keep ambient temps of our little (read: 150sqft) farming room down under 10C, even with 20kW of Antminers. Can we expect more than 4.5TH/s if we're on 220V, and with extremely low ambient temps?

4.1 th, according to feedback from this thread

Let me ask one: if a July/August SP30 is expected to sustain 4.5TH/s on 240V, what sustained hashrate can I expect from the same device when powered from dual 120V/20A circuits instead?

This graph clearly shows severall downtime periods in both sp10 and sp30. I am not sure it is miners fault, espessialy since we see that pool has effect and i dont see such downtimes in my test units. I would like some more client information on this issue. Collider, Please send me data from you minepeon pages (asics, settings and main) durring a long run.

Strangly the unit seems to have been more stable on p2pool than on a regular pool.

The 24h average for the first pic was right around 4450 to 4500

The second one was around 4200 to 4450.

I am not complaining about the recieved hashrate, I was just trying to give Zvi some idea of the unit performance.
What the data seems to suggest is that the unit became less stable after the first 24 hour mining period. (which seems counterintuitive)

Is the number in blue (4.5) the actual numerical value?
it is difficult to ascertain anything from the charts apart that they are probably somewhat in the vicinity of what they are supposed to be, looking closer to 4 than 4.5.

The quoted stats are from the second day of mining.

The unit was in fact online more than 24hours before the "yesterday" stats were recorded.

Here are the detailed stats:



Mining on p2pool. (above)



Mining on normal pool (the one the sp10 from previously was connected to aswell).

Yep, thanks for notification, I was also writing an email  .

Was just about to email, which would really be too many emails from me in any one week

We have unscheduled maintenance in our DC. It will take few hours at least until the machines will come online again.
We'll compensate for the down time.

Regards,
Guy

About $0.022/kWh here, once we get above 200 kW of usage. About $0.04/kWh before that. The SP30's efficiency doesn't enter into my configuration decision calculus yet. Risk of broken equipment does, but the chance of improving the performance of all 68 of our machines with some simple mods largely offsets that.


Thanks! I'll take a look.

Zvi, I'm expecting significant diurnal variations in ambient temperatures in our datacenter. Does minergate take ambient/intake temperature into account at all for determining optimal settings?

Perhaps it would be good to add an adjustable ambient temperature coefficient for some of the major variables, like the PSU power limit. If the thermal shutdown temperature for the PSU is 150°C (what is it actually? I have no idea), and ambient is 20°C, that would be a temperature gradient of 130°C for the PSU's waste heat to flow over. For the sake of easy calculations, let's say the PSU goes into thermal shutdown if you use settings that result in more than 130 W of waste heat being generated in the PSU. That would indicate that the PSU has a thermal resistance to ambient air of 1 W/°C. If you then dropped the ambient air down to 5°C, you would expect to hit thermal shutdown at 145 W instead; if ambient rose to 35°C, then you'd hit shutdown at 115 W. As I understand the limit finding algorithm, you'd end up running your PSU at settings which work in the worst-case scenario (115 W) even during the best-case scenario (145 W) if you saw high-amplitude temperature fluctuations.

Another option might be to first learn what settings work at one ambient temperature, then after that process has completed, create a few separate learning context variable sets (structs or whatever) for different temperature bins, using the original ambient temp settings as a template, and refine the new bins independently. This would probably be significantly harder than simple temperature coefficients.

....and if you/we/i/they ARE customers? 


only when i err or go too far do i feel the need to delete /edit my posts but i was unaware this was a tit-for-tat event...

again, nobody has 'complained of delays' nor asked for a 'queue' or 'exact dates for exact orders', merely how many units have been shipped.

so far it seems 4.

also nobody other than you has mentioned 'tons' in the wild.. in fact no-one has said much of anything other than you and sadly its mostly been based in egotistical aggressive marketing fantasy.

why do we have to go round and round??

I didn't explain myself correctly in mails:
In order to cause recalibration, go to "settings" tab and set high PSU limit (1370W is over the top for 95% of PSUs).
It will take it down again to it's possible maximum over the next day.


Note that the hash rate is more expensive the higher you raise the watts.
You can get ~4TH for 2300W, but the next 600W will only provide 0.5Th - much more 'expensive' hash-rate in terms of power.
So I guess for 2500 the system will give about 4.2TH. If you pay 0.15$ per kWT, you still better running 4.5TH (100$ more). If you pay over 0.35c per kWT, you better run 4.2TH.



Regarding the source code of miner-gate - everything we do is open-source, but playing with code voids the warranty:

git clone https://github.com/Spondoolies-Tech/spond1
cd spond1
make get # downoads dependencies
make init # applying patches, build minimalistic kernel and buildroot
make build # builds everything
make deploy # assembles the rootfs and builds the full kernel

You don't need to build the full image, you can only build packages/spilib/miner_gate and then scp binary to /etc/bin/miner_gate_arm.

lucky you =)
share with us your stats when it's running !