[ANN] Spondoolies-Tech - carrier grade, data center ready mining rigs

[bobsag3]

Sorry for the repeated question SP TECH, but I need to make a decision.   
I have 208V power (three phase 460 transformed to 208V).   Your post about SP30 having a programmed cutoff at 210, is that a hard number?   Will 208V see lower results?   If so, how much lower?   I can re-tap each of the transformer windings to get it up to 220V, but there is a cost to that and I need to figure out if it is worth it.

Zvisha will reply on Sunday.

208 is a bit low, and might have some small performance impact (~2%) - it really depends on exact voltage.
We will investigate next week to see if we can improve that.

My Units are not showing any noticeable impact on 208, Im happy to let you remote in and play with the settings if that will help

[jtoomim]

208 is a bit low, and might have some small performance impact (~2%) - it really depends on exact voltage.
We will investigate next week to see if we can improve that.

My Units are not showing any noticeable impact on 208, Im happy to let you remote in and play with the settings if that will help

2% is a little difficult to notice.

I think my SP30 might be running a little bit faster now on 260V than when it was doing burn-in in Israel (probably at 230V). The difference looks like about 1-2%, if it's not just my imagination. That might just be due to tweaking or a firmware update, though. I doubt it's due to cooler temperatures, because 93% of my datacenter's cooling equipment is still offline and we're running at 27°C intakes.

One of these days, I'll have to figure out the format for /mnt/mmc-config/rrd/hashrate.rrd so I can actually compute the numbers instead of guesstimating from the graphs...

In other news, the performance of my SP30 seems to vary by roughly 8 GH/s per 1°C difference in intake temperature. Last evening, when intake temps were around 22°C, I was getting around 4530 GH/s out of my SP30, but now, with intake around 28°C, I'm getting around 4485 GH/s. I'm looking forward to summer ending and improving our cooling.

[hashfastisawesome]

I've successfully migrated our SP30s over to our 258V supply. Performance is up.

Device ID	Hashrate (cgminer)	AC power usage	Top PSU lim (DC)	Bot PSU lim	J/GH
Mine	4527 GH/s	3024 W	1362	1367	0.668
Wanna1	4476 GH/s	2928 W	1345	1351	0.654
Wanna2	4459 GH/s	2960 W	1344	1347	0.663
Wanna3	4470 GH/s	2992 W	1349	1349	0.669

Thank you for the review. It seems that the competition is spreading lies about SP30 power consumption.

If you are talking about the knc newsletter, it is badly researched.

0,57 J/GH - KNC Neptune - 1950 watts at 3400 GH/s (reported average)
0,62 J/GH - Bitfury BF3500 - 2200 watts, 3500 GH/s (stated, unconfirmed)
0,76 J/GH - Spondoolies SP30 - 3200 watts at 4200 GH/s (reported average)

It also claims bitfury be 3500 does 0.62J/GH, which not even bitfury themsleves claim (it is ~0.775)

Edit:
Most of the reports I've seen of the 20nm competition are over 0.6 W/GHs
We're overclocking the extreme slow corners 28nm ASICs we got. The power ratio for October units will be 0.55 W/GHs

For those of you who don't understand semiconductors: You do not get multi lots and multiple wafers of an "EXTREMELY SLOW CORNER." Stop letting Guy take advantage of your lack of knowledge. Given the orders which they have in the pipe and what they have communicated, I would guess there are AT LEAST 12 lots (25 wafers per lot) in the pipe. You do NOT get 12 lots processed systematically to a slow corner unless you deliberately asked for it.

You're welcome

[jtoomim]

For those of you who don't understand semiconductors: You do not get multi lots and multiple wafers of an "EXTREMELY SLOW CORNER." Stop letting Guy take advantage of your lack of knowledge. Given the orders which they have in the pipe and what they have communicated, I would guess there are AT LEAST 12 lots (25 wafers per lot) in the pipe. You do NOT get 12 lots processed systematically to a slow corner unless you deliberately asked for it.

You're welcome

Your argument is essentially that TSMC might have poor precision, but they wouldn't have poor accuracy. Your argument is that TSMC would not have made a systematic error in their production process. I think it is quite plausible that TSMC might have had one of their machines poorly calibrated for the low voltage threshold transistor type that is so important for bitcoin mining ASICs, but less important for most other applications.

By the way, is that username chosen to deliberately troll people?

[Guy Corem]

My vague statement was correct. I don't reveal the full details on purpose.
In about 45 days (October batch) the SP30 will be 5.5 THs
(The limiting factor will remain the PSUs. The RockerBox has much more potential for over clocking)
This will be done without mask fix, just production tuning.

jtoomim more or less got it right.

Guy

[Guy Corem]

Can someone from spondoolies tell me whats going on with my hosted units? Order #1125

http://eligius.st/~wizkid057/newstats/userstats.php/1P19DeFn9dSkvv3d23y3BRJmmabTQiYHym

https://upload.robert-hager.at/uploads/static/eligius.PNG

 I am fine with this spikes as long as they level out, but it would be interesting what causes it? Especially would be interesting how three SP30 do 17 TH's ?  

We're compensating you for a unit down time.
We'll replace your hosted unit with another units early next week, after production. One of your unit has issues.

With all due respect, I do have some doubts about that program which pushes PSU to the actual limit through repeated shutdowns.
Could it be the culprit? Perhaps, it is better to limit power to 1300 or even 1250W?
It is not worth a few extra GH if it significantly increases the PSU failure rate (for the purposes of argument- I don't know if this is true or not).

Not related at all to the PSUs

[hashfastisawesome]

My vague statement was correct. I don't reveal the full details on purpose.
In about 45 days (October batch) the SP30 will be 5.5 THs
(The limiting factor will remain the PSUs. The RockerBox has much more potential for over clocking)
This will be done without mask fix, just production tuning.

jtoomim more or less got it right.

Guy

My point is that TSMC will not have 'accidentally' botched 3 separate batches of lots (August, September, October). Remember, these are the same 28nm lines that Qualcomm and Broadcomm use. The notion that it is TSMC's 'fault' is simply a convenient excuse to a design mistake. I have no doubt that TSMC is capable of a retarget of your LVT devices to meet your original performance target. However, that 10% increase you were hoping to achieve with a post engineering run retarget has disappeared. Guy, I applaud your ability to perform damage control.

jtoomim, yes my name is intentionally chosen to troll. That doesn't delegitimize my arguments. Source: I run fab sync for a major US semiconductor company. You don't have to believe me, but you can do your own research on the topic.

[Bicknellski]

My vague statement was correct. I don't reveal the full details on purpose.
In about 45 days (October batch) the SP30 will be 5.5 THs
(The limiting factor will remain the PSUs. The RockerBox has much more potential for over clocking)
This will be done without mask fix, just production tuning.

jtoomim more or less got it right.

Guy

My point is that TSMC will not have 'accidentally' botched 3 separate batches of lots (August, September, October). Remember, these are the same 28nm lines that Qualcomm and Broadcomm use. The notion that it is TSMC's 'fault' is simply a convenient excuse to a design mistake. I have no doubt that TSMC is capable of a retarget of your LVT devices to meet your original performance target. However, that 10% increase you were hoping to achieve with a post engineering run retarget has disappeared. Guy, I applaud your ability to perform damage control.

jtoomim, yes my name is intentionally chosen to troll. That doesn't delegitimize my arguments. Source: I run fab sync for a major US semiconductor company. You don't have to believe me, but you can do your own research on the topic.

Your expert opinion?

What specific degrees or experience do you have?

Why anonymously?

Not that I discount anything you are posting just trying to get a measure of your interest or bias potentially. There are people who have come on these forums to purposefully devalue brands so it be nice to know if this is intended as consumer protection versus narrative management for the purpose of degrading a company's image. The "truth" can be put in MANY different ways.

[Guy Corem]

My vague statement was correct. I don't reveal the full details on purpose.
In about 45 days (October batch) the SP30 will be 5.5 THs
(The limiting factor will remain the PSUs. The RockerBox has much more potential for over clocking)
This will be done without mask fix, just production tuning.

jtoomim more or less got it right.

Guy

My point is that TSMC will not have 'accidentally' botched 3 separate batches of lots (August, September, October). Remember, these are the same 28nm lines that Qualcomm and Broadcomm use. The notion that it is TSMC's 'fault' is simply a convenient excuse to a design mistake. I have no doubt that TSMC is capable of a retarget of your LVT devices to meet your original performance target. However, that 10% increase you were hoping to achieve with a post engineering run retarget has disappeared. Guy, I applaud your ability to perform damage control.

jtoomim, yes my name is intentionally chosen to troll. That doesn't delegitimize my arguments. Source: I run fab sync for a major US semiconductor company. You don't have to believe me, but you can do your own research on the topic.

I don't blame TSMC (or GUC). The ASICs we got and getting are within TSMC acceptence criteria.
My hands are tied here. With one WAT graph I can prove my claims, but I won't.
You're welcome to email me privately. If you're not working for a competition, I'll disclose the info after signing mutual NDA.
It's much more than 10% increase and it's not post engineering.
I'm as transparent as I can be here. No damage control, but simple truth.
Even with the underperforming slow ASICs we have the best miner in the market.

Cheers,
Guy

[Easy2Mine]

Even if you had faster running ASIC, he will still makes it look dark.
I am not a customers of ST yet. I have admire their enginering skills from the beginning.
If you took the time to look closer how well designed the case is and easy to mass deploy a big
farm with SP 30 is, compared with all the other options out there.
Saying that their chip is bad with no proof except telling that you have some skills is hard to believe.

[Collider]

My sp30 travelled from Israel to jtoomim´s WA hosting facility in only 2 days. (thank you Spond∞lies and FedEx)

A short review of my hosting experience so far can be found in the hardware hosting comparison thread:

https://bitcointalk.org/index.php?topic=622998.msg8361310#msg8361310


Also: I was faster than you, RoadStress

[murdof]

My sp30 travelled from Israel to jtoomim´s WA hosting facility in only 2 days. (thank you Spond∞lies and FedEx)

A short review of my hosting experience so far can be found in the hardware hosting comparison thread:

https://bitcointalk.org/index.php?topic=622998.msg8361310#msg8361310


Also: I was faster than you, RoadStress

So you guys suggest http://toom.im/ for best solution for SP30?
What happened to the Iceland data center? Aren't you using this for SP10?

[Collider]

So you guys suggest http://toom.im/ for best solution for SP30?
What happened to the Iceland data center? Aren't you using this for SP10?

I am still running one sp10 at the Iceland facility, and I will not change that one (also, I have a minimum contract length there).

Toom.im offers 76$/kW/m at the same 6 month contract lenght Thor DC (Advania, Iceland) offers. (Thor is 120$/kW/m though, and does not accept bitcoin or paypal payments).
Contracts with Toom.im are 90$/kW/m on a one-month term, with prices as low as 70$/kW/m.

I can say however, that I am moving an sp10 from spondoolies to Toom.im aswell.
Ambient temperature might be slightly lower in Iceland during the summer, but certainly not in the winter. While the sp10 is fairly temperature sensitive, sp30 isn´t.

[RoadStress]

My sp30 travelled from Israel to jtoomim´s WA hosting facility in only 2 days. (thank you Spond∞lies and FedEx)

A short review of my hosting experience so far can be found in the hardware hosting comparison thread:

https://bitcointalk.org/index.php?topic=622998.msg8361310#msg8361310
Also: I was faster than you, RoadStress

I will compensate Collider 

So you guys suggest http://toom.im/ for best solution for SP30?
What happened to the Iceland data center? Aren't you using this for SP10?

We are decentralizing!

[Guy Corem]

My sp30 travelled from Israel to jtoomim´s WA hosting facility in only 2 days. (thank you Spond∞lies and FedEx)

A short review of my hosting experience so far can be found in the hardware hosting comparison thread:

https://bitcointalk.org/index.php?topic=622998.msg8361310#msg8361310
Also: I was faster than you, RoadStress

I will compensate Collider 

So you guys suggest http://toom.im/ for best solution for SP30?
What happened to the Iceland data center? Aren't you using this for SP10?

We are decentralizing!

[Biffa]

My vague statement was correct. I don't reveal the full details on purpose.
In about 45 days (October batch) the SP30 will be 5.5 THs
(The limiting factor will remain the PSUs. The RockerBox has much more potential for over clocking)
This will be done without mask fix, just production tuning.

jtoomim more or less got it right.

Guy

My point is that TSMC will not have 'accidentally' botched 3 separate batches of lots (August, September, October). Remember, these are the same 28nm lines that Qualcomm and Broadcomm use. The notion that it is TSMC's 'fault' is simply a convenient excuse to a design mistake. I have no doubt that TSMC is capable of a retarget of your LVT devices to meet your original performance target. However, that 10% increase you were hoping to achieve with a post engineering run retarget has disappeared. Guy, I applaud your ability to perform damage control.

jtoomim, yes my name is intentionally chosen to troll. That doesn't delegitimize my arguments. Source: I run fab sync for a major US semiconductor company. You don't have to believe me, but you can do your own research on the topic.

How do you know that they didnt get all the SP30 chips in one batch? And it just takes this long to build and ship them. Plus I think its just August and September, the October batch is purported to be back up to 5.5TH so is probably separate. Thats the way I read it anyway.

[mrpark]

My vague statement was correct. I don't reveal the full details on purpose.
In about 45 days (October batch) the SP30 will be 5.5 THs
(The limiting factor will remain the PSUs. The RockerBox has much more potential for over clocking)
This will be done without mask fix, just production tuning.

jtoomim more or less got it right.

Guy

My point is that TSMC will not have 'accidentally' botched 3 separate batches of lots (August, September, October). Remember, these are the same 28nm lines that Qualcomm and Broadcomm use. The notion that it is TSMC's 'fault' is simply a convenient excuse to a design mistake. I have no doubt that TSMC is capable of a retarget of your LVT devices to meet your original performance target. However, that 10% increase you were hoping to achieve with a post engineering run retarget has disappeared. Guy, I applaud your ability to perform damage control.

jtoomim, yes my name is intentionally chosen to troll. That doesn't delegitimize my arguments. Source: I run fab sync for a major US semiconductor company. You don't have to believe me, but you can do your own research on the topic.

How do you know that they didnt get all the SP30 chips in one batch? And it just takes this long to build and ship them. Plus I think its just August and September, the October batch is purported to be back up to 5.5TH so is probably separate. Thats the way I read it anyway.

I don't think TSMC makes too many mistakes, they are not worth 106 Billion dollars for nothing.

[Biffa]

My vague statement was correct. I don't reveal the full details on purpose.
In about 45 days (October batch) the SP30 will be 5.5 THs
(The limiting factor will remain the PSUs. The RockerBox has much more potential for over clocking)
This will be done without mask fix, just production tuning.

jtoomim more or less got it right.

Guy

My point is that TSMC will not have 'accidentally' botched 3 separate batches of lots (August, September, October). Remember, these are the same 28nm lines that Qualcomm and Broadcomm use. The notion that it is TSMC's 'fault' is simply a convenient excuse to a design mistake. I have no doubt that TSMC is capable of a retarget of your LVT devices to meet your original performance target. However, that 10% increase you were hoping to achieve with a post engineering run retarget has disappeared. Guy, I applaud your ability to perform damage control.

jtoomim, yes my name is intentionally chosen to troll. That doesn't delegitimize my arguments. Source: I run fab sync for a major US semiconductor company. You don't have to believe me, but you can do your own research on the topic.

How do you know that they didnt get all the SP30 chips in one batch? And it just takes this long to build and ship them. Plus I think its just August and September, the October batch is purported to be back up to 5.5TH so is probably separate. Thats the way I read it anyway.

I don't think TSMC makes too many mistakes, they are not worth 106 Billion dollars for nothing.

I don't know whether the mistake was in manufacturing or in design, all I meant was that it appeared to me that the concept of them making 3 batches with the same error would, as you say, not make sense. Either from SPtech's POV or TSMC. So I was suggesting that the erroneous batch was a single batch for Aug/Sep production, hence why it is being corrected in the October/Nov batch of chips from TSMC.

So say it was a production error, its only one mistake, not 3, not "too many" just one. Of course it might not have been TSMC's fault at all, but again thats not what I was talking about.

[Searing]

My vague statement was correct. I don't reveal the full details on purpose.
In about 45 days (October batch) the SP30 will be 5.5 THs
(The limiting factor will remain the PSUs. The RockerBox has much more potential for over clocking)
This will be done without mask fix, just production tuning.

jtoomim more or less got it right.

Guy

My point is that TSMC will not have 'accidentally' botched 3 separate batches of lots (August, September, October). Remember, these are the same 28nm lines that Qualcomm and Broadcomm use. The notion that it is TSMC's 'fault' is simply a convenient excuse to a design mistake. I have no doubt that TSMC is capable of a retarget of your LVT devices to meet your original performance target. However, that 10% increase you were hoping to achieve with a post engineering run retarget has disappeared. Guy, I applaud your ability to perform damage control.

jtoomim, yes my name is intentionally chosen to troll. That doesn't delegitimize my arguments. Source: I run fab sync for a major US semiconductor company. You don't have to believe me, but you can do your own research on the topic.

How do you know that they didnt get all the SP30 chips in one batch? And it just takes this long to build and ship them. Plus I think its just August and September, the October batch is purported to be back up to 5.5TH so is probably separate. Thats the way I read it anyway.

I don't think TSMC makes too many mistakes, they are not worth 106 Billion dollars for nothing.

thats the catch...it is NOT a mistake by TSMC if they are as spondoolies said 'within spec' just at the low end of that ...I"m sure if they were truely out of spec
spondoolies would have other options....such stuff is likely written by TSMC to benifit them in slippage one direction or the other on performance anyway imho.

Searing

[Guy Corem]

We're not blaming anyone or any company. The nature of the business doesn't allow shuttles and first productions runs.
As explained, all the wafers lots until and including September suffers from the same, easily fixable, production issue.

Guy