This either means the chip is designed with a slight fault, or certain production methods at TSMC caused the chip to underperform. (or some mixture of the two)

The comment really suggests the latter.

Guy also seems to be fairly confident that it is not a packaging issue, and that it can be fixed for future chip batches (which would be great news.)

If TSMC is at fault, spondoolies might get the new and improved mask at a discount.
If it is more a question of implementing certain requirements posed by TSMC, it should really have been corrected during (initial) tapeout.

Again, this is pure speculation (which I actually resent) so please wait for spondoolies to comment and don´t take any of my suggestions too seriously.

Obviously overdelivering is always better, and yes, their previous estimations of "6TH per sp30" are now going to be very costly for them.

On the other hand, spondoolies raked the cash in for 6TH miners, and are now going to find a way to make every customer whole.

While overpromising could be considered a rookie mistake, keep in mind that even bitmain overpromised on their S3 by about 10%, offering refunds in that amount.

They didn´t have a worse efficiency though and atleast my S3 (which I bought just for fun) actually does 500GH after some tweaking.


The tests from spondoolies, including stress tests under high temperatures, should give everyone a solid number as to what they can expect from their sp30s.

I am still a little unsure as to how they can be certain that the chip isn´t as good as it should be (and not some other component is at fault) without specific chip testing, but they probably found a way to eliminate certain possibilities.

However, I doubt the foundry is at fault, but spondoolies would then be able to hopefully get some compensation.
I also don´t know how they have ruled out bad /non-optimal chip packaging as a source of lower efficiency.
(I know that certain materials used to connect the inner connections with the pins at the outside of the package can be problematic and create a  slightly lower efficiency).

This is all pure speculation and once spondoolies has solid numbers and a compensation plan for customers, I am sure we are going to learn more about the specific nature of the problem(s).
That is if this doesn´t give away any of spondoolies IP.

If there is nothing to tell, why would they tell you "no update yet"?

There is obviously some major testing and firmware building needed to address the issues, and spondoolies wants to be certain that provided specs are accurate and reproducable.

July batch will be shipped in July and final specs will be released in July.

Exact valuation aside, I am certain nobody will contest that Bitfury is probably the biggest player in the asic mining industry.

They have easily produced the most hashpower and have enormous profits.

They also seem to be holding the highest percentage of the network.

They have been the nearly uncontested leader in W/GH right till the Neptune, even though they were on 55nm all the time.

They wont. UPS will collect payment for VAT when delivery takes place.

The S3 manual doesn´t mention any reset switch...

This I didn´t do. Is there anything specific like switching it for 5 seconds?

S3 controller boards do not utilize sd cards.

@ bitmain

Please tell me how to get rid of the following issue:

The web interface only shows this on startup:

the website isn´t working.

yes, I see no real danger of a unit "lasting shorter", as it might influence the asic life cycle by around 20% adversely, but normal asics last 5 years easily.

Seeing as they shouldn´t be able to generate more than electricity cost in 1-2 years (conservative estimate), this really is nothing to worry about.

As guy already wrote, things should be clearer within the next 4 days.

He might not be able to give reliable power consumption and speed estimates as the firmware should still be tweaked by Zvisha for optimum performance under the changed circumstances.

Anyway, I am sure we will see some news once the first units ship, and I for one will be reporting on the hashrate and performance (that I can observe from the pool page).

Looking forward to my (hosted) sp30.

Compensation for less efficiency might be best given (as an option) in the form of cheaper longterm hosting, but let´s wait and see how the final specs will turn out.
(Therefore only the lower total hashrate would still need compensation)

In the meantime I wish spondoolies the best of luck in trying to configure the miner and firmware as best as possible.

To be fair, I didn´t do any exact thermal measurments (or exact GH measurements) on my DCDCs, so it could possibly have worked at that GH without the additional heatsinks.

I just had them lying around and thought additional heatsinks couldn´t possibly lower the hashrates, so I attached them.

I was also thinking about maybe reapplying thermal paste, but without access to my webinterface I couldn´t even say whether that lowered temps.

When I open the unit the next time I can take some pics.

I hardly think they intended to give too high specs, but promising a very high hashrate and efficiency that then cannot be met obviously isn´t beneficial for them either.

I think once the product specs are confirmed (and in a competitive range) they will still be able to sell their coming batches, let alone chips.

Obviously, most of the profit on the sp30s should be gone (for sp) with compensation for lower hashrate and efficiency in order.


If they have managed to produce a good chip they will still be able to do additional bulk sales of those, and if the sp30 doesn´t turn out too horrible, those will surely sell aswell.

yesss!

Well, the only thing broken is that the webinterface isn´t available anymore.

It runs at around 505GH average after I put small heatsinks on the DCDCs (I had them lying around) and some overclocking, obviously.

Erm how do I reflash the firmware from SSH....

Has anyone got the download address and instructions for a firmware reflash?

Mine shows this error when I want to access the webinterface:


It runs stable over 500GH, but I would like to access the webinterface again

This is one of the problems you run into when not testing chips and / or having test boards with one chip.

It is harder to make out the reason why the chip is performing not as simulated.

Then again, they seem confident that they will be able to get the unit working on the ordered pcbs, so the problem can´t be too big a deal.

Let´s see what the results of the more detailed analysing is, and we will all be wiser.

Actually, it would call for a higher refund as the power efficiency would also be impacted by either 17% or 33% (in your example).

It would call for an additional 500$ refund if the power efficiency would be off by 17%. (calculated on a basis of 100$/kW/m hosting cost and maximum power draw of 2.7kW at the wall, over one year only)

This is a conservative estimate, but let´s wait until final specs are released.