[Work in progess] Burnins Avalon Chip to mining board service

[guitar82]

You can also use the 12v supply going to the motherboard, I think there are breakout boards for that as well.

[tigerbit]

You can also use the 12v supply going to the motherboard, I think there are breakout boards for that as well.

Duh!  Good point, I have a homemade board I designed for my FPGAs, although I only use it for switching & providing min loads to 3.3v & 5v.  Lots of 12v there too!  Time for v2.0!

[AMD FTW]

Didn't Burnin say you could power 3 boards with one 6 pin adapter? With overclocking I think it would only be two now, but that is what I thought. I am not sure how they wire though.

I think it was two boards, but that was before overclocking fever took over

Officially one 6-pin PCI-E connector has 75W power output, which burnin already exceeded with higher clock rates. I guess he wasn't using single PCI-E connector though.

That's the rated spec of a 6 pin pci-e power connector, it can and will output more wattage. An 8 pin pci-e connector is rated for 150 watts and the only difference between a 6 pin and 8 pin is the addition of 2 more grounds to the connector.

Yes but even after making a splitter for x1 8pin to x2 6 pin that's still going to present a power distribution problem. 12 boards 1000W modular psu, I'd need six PCIe8 leads onto the PSU to stay legit.  Would it be worth looking at x2 smaller PSUs? Kinda redundancy - degraded service.

It's no problem at all as I've used 1 sata to 6 pin pci-e power adapters for years on my gpu bitcoin rigs setup as well as the molex to 6 pin pci-e adapters as well

[cimpex]

power distribution boards will be very usefull (for 24pin ATX and 4/8pin mobo) including on/off switch.
another question: is it problematic when several rails are chained together? maybe chain them all together like:
input: 24p ATX + 4/8p mobo + all available pcie --> output: 10-15x 6p pcie

[jlsminingcorp]

@burnin

When you launch your shop, I think it would be wise to disable the order button in the beginning for a certain time (1 or 2 days), so that all the guys can have a look at the offering, the different items and how to combine them. That would prevent possible wrong orders done in a hectic rush and also the trouble you would have to correct the orders. When questions arise you can answer them in advance.

Just my thoughts.

Thats a great idea since burnin already stated that he will give out a warning about the opening of the shop 24 hours before. When he opens the shop with buying disabled everyone has 24h to calculate and maybe even prepare the basket.

These seem like really good ideas. "Timed-release" shopping things are always a bit of a nightmare though (very stressful for all). I remember being one of the idiots who got out of bed at a f**king stupid time in the morning to try to order a RasPi on the day the stores opened, only to spend a couple of hours unintentionally contributing to the massive DDoS on RS and Farnell's sites that brought them crashing to their knees (RasPi fever, I apologise ). It would be a real shame to see something similar happening to burnin's site when the payment section opens. I'm sure burnin will do a far better job than RS or Farnell, who were ridiculously unprepared, but I think we all know that his site is going to be busy!

Is there any way to make things work more smoothly than everybody scrabbling around to be the first to submit an order on a hugely overloaded website and getting frustrated when it doesn't go through properly?

I have a slightly more radical suggestion (it may not be sensible, so please be gentle with your replies ). What about taking orders for a couple of hours after the payment option opens (or longer if things are really busy) and then randomly assigning orders placed within this time period to slots in the order queue (like a lottery, names out of a hat for 1st place, 2nd place and so on). You could then repeat this for the next time period and so on until things quieten down and it's safe to switch to first-come first-served. This way there would be an incentive for ordering early (being within an earlier time period will mean a better position in the order queue), but there wouldn't need to be a mad rush to order within the first 30 seconds of opening. Your position in the order queue is pretty much random when things are hugely busy anyway, so why not formally randomise things? This could reduce the stress levels a lot and reduce the pressure on burnin's servers. Any thoughts?

[HeRetiK]

Didn't burnin post somewhere that the orders will be processed in the order in which the group-buy chips arrive? As in - it doesn't matter when you hit the buy button, but rather when your group-order batch arrives and when you placed your group-buy order?

[jlsminingcorp]

Didn't burnin post somewhere that the orders will be processed in the order in which the group-buy chips arrive? As in - it doesn't matter when you hit the buy button, but rather when your group-order batch arrives and when you placed your group-buy order?

Yes, I'm pretty sure you're right about the group buy batch being one of the limiting factors (if your chips aren't there he can't assemble), but if I understood correctly burnin's plan was to assemble each board within a particular group buy batch based on when orders are placed on his site. This would mean that there is an incentive to be "quick" in placing your order.

Edit: I think I found the relevant message (https://bitcointalk.org/index.php?topic=179769.msg2402705#msg2402705).

[Exoskeleton]

Didn't burnin post somewhere that the orders will be processed in the order in which the group-buy chips arrive? As in - it doesn't matter when you hit the buy button, but rather when your group-order batch arrives and when you placed your group-buy order?

Exactaly. Unless all the chips arrive on the same day there is little point to being first in line. If zefirs batch 1 chips arrive first, and your not in that group, you will be skipped over. First in line, first to be skipped. This will keep happening untill your batch of chips comes in to burnin.

Its likely that being first in line won't get you ahead by more than a few days in your individual group buy. I took a look at some of the assembly gear being utilized by burnin and I think 1,000 units per week (20,000 chips) is underestimating what can be done.

I know every day counts, but I don't think the distinction is as big as most think. The real hold up here will be the delivery of chips from Avalon. I fear that the gap in shipping between batches may be bigger than expected.

[SebastianJu]

@jlsminingcorp... the lottery would only lead to even more connections because everyone would hit F5 or so to get one of the tickets. If an order went through it would be easier for the webserver.
But in fact i doubt that the webserver will have to do much. I mean how many users will order? I guess over 200 from my groupbuy, maybe 1000 more form others? And not all will buy in the first minutes. So i think the attack on webserver wont be that hard.

@Heretik... the orders will be worked up after time of order. With a timeframe to being able to pay through wire too, which can take some days, and to not having disadvantages against payments with bitcoins) and the second order priority is if your chips already arrived at burnin's place. If you are in line but your chips arent there others are done first.
It would be a nightmare to check first if every chip that arrives was bought before or after other chips. Especially with the involvement of groupbuys.
By the way... burnin stated that he will do it this way.
There wont be many many buyers anyway. Many boards ok, but not so many buyers.

[CumpsD]

This is the actually final message of burnin on ordering:

https://bitcointalk.org/index.php?board=76.0

"Order date/time will (within reason) now be the determining factor (within a batch)."

[RoadStress]

"Order date/time will (within reason) now be the determining factor (within a batch)."

I have a feeling that "withing reason" means that people who order in the same day (00-24 hour period) will be shipped according to chips batch number and people who ordered different days will be shipped according to order day. So if 2 people from chips batch 1 and chips batch 4 order the same day will be shipped the same day, but someone from batch 1 ordering the fifth day will be shipped the fifth day. (as long as there are enough chips for everyone, if not then chips batch number is the first)

[SebastianJu]

"Order date/time will (within reason) now be the determining factor (within a batch)."

I have a feeling that "withing reason" means that people who order in the same day (00-24 hour period) will be shipped according to chips batch number and people who ordered different days will be shipped according to order day. So if 2 people from chips batch 1 and chips batch 4 order the same day will be shipped the same day, but someone from batch 1 ordering the fifth day will be shipped the fifth day. (as long as there are enough chips for everyone, if not then chips batch number is the first)

Why do you think so? Its very easy for burnin to see the ordernumbers in his shop, but it would be much work to find out in which batch a customer bought, when this batch was ordered and so on. I, in burnin's shoes wouldnt do the work, and i dont see why that should be done. Burnin doesnt have to favor early chip buyers. The only advantage he has is when he got money for his business. That would be an order.
In my understanding burnin wont care about batches, batch order dates and so on. I think he wrote this already in the thread. Its simply too much work without any advantage.
The only advantage an early buyer might have is, that avalon is sending the batches in order of ordering. So the chips might appear at burnin earlier.

[RoadStress]

Why do you think so? Its very easy for burnin to see the ordernumbers in his shop, but it would be much work to find out in which batch a customer bought, when this batch was ordered and so on.

No particular reason. Your version sounds reasonable too.

[eros]

Burning boards have ~320 parts and pnp machine handles ~22000cph. So ~70 boards per hour. He can handle over 10000pcs avalons in normal 8hrs workday. But broblem arise from postage. It takes lot of time to pack boards and post to many customers. If average order is ~200 avalons ~10boards it means 50 pack/post per day. And he is alone and have other work too. But if things gone well he handle all orders in week.

Burning test pcb bottom are full painted. Orginal avalons have solder square in every chip locations. I think solder is better heat conductor than paint. Maybe he corrected it production boards..?

[burnin]

Burning boards have ~320 parts and pnp machine handles ~22000cph. So ~70 boards per hour. He can handle over 10000pcs avalons in normal 8hrs workday. But broblem arise from postage. It takes lot of time to pack boards and post to many customers. If average order is ~200 avalons ~10boards it means 50 pack/post per day. And he is alone and have other work too. But if things gone well he handle all orders in week.

Burning test pcb bottom are full painted. Orginal avalons have solder square in every chip locations. I think solder is better heat conductor than paint. Maybe he corrected it production boards..?

You are correct about the pnp machine (i am up to 405 parts now btw).
But i am a professional, and as such I do what I can do best and outsource the rest!
Therefore i'll use a fulfillment service to do the packing and shipping.
I know that any delay will cost YOU, my customers real money and therefore getting your stuff to you ASAP is a priority for me.
Exposing the copper on the bottom will increase the thermal coupling but i want to keep the heat sinks isolated.
(Important for water cooling, electrical corrosion)



Shop Timing:
I agree that this is not the optimal strategy for things like this.
But with all in all 500 People ordering stuff things should not break instantly.
(did choose a beefy server)

And i have some numbers to go with those from yesterday:
Slightly different air cooling setup therefore different temperatures with air cooling. (fan placement)
TL;DL : 450Mhz [9Ghash/s] - STABLE
But at the cost of 94Watts of power.

Air:
431 - 54, 48, 1.30V, 87W, stable
450 - 56, 48, 1.30V, 90W, HW Errors
450 - 57, 52, 1.34V, 94W, slightly increased error rate compared to what i normally call "stable" but close enough

Water:
450 - 54, 32, 1.34V, 94W, slightly less hw errors then with air

[RoadStress]

TL;DL : 450Mhz [9Ghash/s] - STABLE
But at the cost of 94Watts of power.

Air:
431 - 54, 48, 1.30V, 87W, stable
450 - 56, 48, 1.30V, 90W, HW Errors
450 - 57, 52, 1.34V, 94W, slightly increased error rate compared to what i normally call "stable" but close enough

Water:
450 - 54, 32, 1.34V, 94W, slightly less hw errors then with air

Grats for being the first one to hit that speed stable! Great job!

Edit: burnin please plan the PSU so that it accommodates with the overclocking figures and with stacking. thank you.

[tigerbit]

And i have some numbers to go with those from yesterday:
Slightly different air cooling setup therefore different temperatures with air cooling. (fan placement)
TL;DL : 450Mhz [9Ghash/s] - STABLE
But at the cost of 94Watts of power.

Air:
450 - 57, 52, 1.34V, 94W, slightly increased error rate compared to what i normally call "stable" but close enough

Water:
450 - 54, 32, 1.34V, 94W, slightly less hw errors then with air

That's amazing.  Well done!  450MHz on air cooling - Bitsyncom seemed to suggest this would not be possible on air.  Congratulations - now hoping for a stable and high quality production run to repeat your brilliant achievements.  @94W I'm not sure I'll actually want to go that high due to PSU sizing, but it just shows what a quality product you've achieved here.  Thank you.

[ecliptic]

May want to provide 12V on both the PCI pin and with some crazy barrel plug -> 12V SATA power adapter or something - looks like those 12V inputs are wired together

[isimme]

Edit: burnin please plan the PSU so that it accommodates with the overclocking figures and with stacking. thank you.

+1
Yes Please!
-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

Burnin,

You said in regards to air cooled over-clocking:
"Slightly different air cooling setup therefore different temperatures with air cooling. (fan placement)"
-Would this work with stacked boards or is the fan placement maybe unconventional just to get that umph?

You said:
"slightly increased error rate compared to what i normally call "stable" but close enough."
- What does that mean, maybe the numbers or the log/report from cgminer would clear this up?

-What is your personal opinion or gut feeling on longterm air cooled over-clocking at
 450 - 57, 52, 1.34V, 94W?

*I want to thank you for all your efforts, professionalism, and forthcomingness with info, updates, and questions!*

[f.baasch75]

And i have some numbers to go with those from yesterday:
Slightly different air cooling setup therefore different temperatures with air cooling. (fan placement)
TL;DL : 450Mhz [9Ghash/s] - STABLE
But at the cost of 94Watts of power.

Air:
431 - 54, 48, 1.30V, 87W, stable
450 - 56, 48, 1.30V, 90W, HW Errors
450 - 57, 52, 1.34V, 94W, slightly increased error rate compared to what i normally call "stable" but close enough

Water:
450 - 54, 32, 1.34V, 94W, slightly less hw errors then with air

WOW! Excellent, excellent! Burnin, that is a really impressive result, congratulations!