Pages:
Author

Topic: [Open Source Hardware Project] Hive & Wasp Prototype Development - page 5. (Read 16215 times)

hero member
Activity: 924
Merit: 1000
Update:

From the last meetings we had on Saturday and Sunday our project EE has clarified some the issues that could have potentially caused prolonged delays in redesign of our own boards or would have caused reliability issue with our units. This is a direct result of heat and power  issues related with the chip packages of all newer 28nm chips that are soon to hit the market next month and early in 2014. We are now more confident in our prototype potential release dates.

BitFury Wasp Prototype

+  16 to 24 chips  for 300 gh/s to 440 gh/s depending on the configuration.
+  4U Server configuration
+  Design for the power on the Wasp currently being down so 3d renders for the prototype are a week away.
+  By December the prototype testing video should be out.
+  By December the prototypes shipped to a datacenter for longer term testing and troubleshooting.
+  The DIY & Licenced Production could start as early as January for these units should there be a market for them.

Avalon Gen I Wasp Prototype

+  16 to 24 chips for 50 gh/s to 75 gh/s depending on the configuration. (Gen II chips would be higher)
+  4U Server configuration.
+  Design for the power on the Wasp currently being down so 3d renders for the prototype are a week away.
+  By December the prototype testing video should be out.
+  By December the prototypes shipped to a datacenter for longer term testing and troubleshooting.
+  The DIY & Licenced Production could start as early as January for these units should there be a market for them.

A1 Wasp Prototype

+  A1 designs will be applied to the BitFury Wasp design and should take less than a week.
+  4U Server configuration
+  Design for the power on the Wasp currently being down so 3d renders for the prototype would occur in late December.
+  By Late December the prototype testing video should be out.
+  By Late December or Early January the prototypes shipped to a datacenter for longer term testing and troubleshooting.
+  The DIY & Licenced Production could start as early as January for these units.

Minion Wasp Prototype

+  Minion designs will be applied to the lessons learned from the A1 Wasp design and should take less than a week.
+  4U Server configuration
+  Design for the power on the Wasp currently being down so 3d renders for the prototype would occur in January or earlier.
+  By Late February the prototype testing video should be out.
+  By Late February or Early March the prototypes shipped to a datacenter for longer term testing and troubleshooting.
+  The DIY & Licenced Production could start as early as Late February or Early March for these units.

We are keen on working with chip manufacturers to get their chips into our modular design as well as working with group buys, diy'ers or EE's who have previously handled production of Avalon, BFL, BitFury or others boards that have made it to the marketplace. Our intention is to put out the wasp and hive as open source software and hardware and provide support for those who wish license our units for production. Given that our modular system does have a number of other uses depending on the design of the Wasp mining bitcoin will not be the only thing in the designs repertoire. There are also future plans in the near future to adapt the basic design with add ons to allow for scrypt mining but there is functionality well beyond cryptocurrencies such as a robotic controller.

Our modular miner design strengths include:

+  No requirement to change firmware software for each new Wasp.
+  Firmware will be loaded during fabrication for each new Wasp.
+  Multiple varieties of Wasp chip set mining together in the same modular unit.
+  Short lead time to production of new chips with the modular design.
+  Scales easily and a rapid deployment of hash-power.
+  Protection against obsolescence.
+  Resale value of the components for other purposes beyond mining bitcoins.

If you are keen on getting a sneak peak at our designs that are up on the Zoho page before we close the pages to paid members... then drop me an email and I will add you to the Zoho Project page. Membership to the The Wasp Project Collective is always open and investment in the Wasp and Hive is ongoing and we welcome paid members to join either the design or project management team on this project. We welcome questions and we hope to provide evidence of our prototype miners hashing soon. Come join us.
sr. member
Activity: 434
Merit: 265
The Wasp Project Next EE Meeting - Wasp & Hive.
------------------------------------
The Wasp Project Next Organizational Meeting.

Let's challenge the market. Happy to be a part of "The Wasp" project.
hero member
Activity: 924
Merit: 1000
Update:

Our BitFury prototype should be testing by the end of next week. Next meeting 6pm Pacific Time Seattle USA on the teamspeak server. Minute notes from today's meeting will be posted later in the Zoho Project page. Members welcome anytime on the teamspeak channel to talk with the EE's. The current designs are being done in PCB123® - Free PCB Design Software from Sunstone found here: http://www.sunstone.com/pcb123.aspx

Notes on Chips Order of Prototype Development for the Wasp:

1. BitFury chips in hand will be arriving this week and will be put into prototype boards by end of the week and live development update will be available via VPN for members interested in the design process.

2. Our EE in Utah will work concurrently on the Avalon chips he has in hand already the development of this Wasp will be over the next few weeks updates will follow at the next meeting.

3. A1 chips still from what we know will ship in December but we are still waiting on chip specifications. Designs can be modified with in a few hours of specs being dropped and we have contingencies already in place for a variety of different specs of the A1 chips.

4. BA chips are still on schedule to ship in February and chip specs should drop soon.

As always we are still open to anyone willing and interested to join our group we have a $100 USD membership fee and if you want to create a new project or support the current Wasp / Hive development a 2 BTC opt in will cover the costs of the prototype development as well as chip purchases. Drop me your email and I can add you to the Zoho Page.
hero member
Activity: 924
Merit: 1000
Meeting slightly delayed... will start on Teamspeak in about 1 hour and 10 minutes at 8:30 pm Utah Time. One of the EE's still on the road heading home.
hero member
Activity: 924
Merit: 1000
Given the rapid increase in USD/BTC price we are setting our membership fees at $100 USD lifetime membership in the group and lowering the chip buy in for the A1 chips to 2 BTC. We are basing our prices on Bitstamp. That should be easier for members to get involved in the project.

We have purchased the domain www.thewaspproject.com so if you are keen on helping build the site ground up join The Wasp Project today and we hope to see everyone at the EE briefing tomorrow. Lots of great news will be on offer I am sure from both Utah and Washington. Exciting times for us.

That sounds better than the 1 BTC price for membership you had before.


1 BTC was set when the BTC / USD price was reasonably low not at nearly $400. We all have to adjust. Tomorrow when it is 50$/BTC at least we have the base at $100 USD.
legendary
Activity: 3416
Merit: 1912
The Concierge of Crypto
Given the rapid increase in USD/BTC price we are setting our membership fees at $100 USD lifetime membership in the group and lowering the chip buy in for the A1 chips to 2 BTC. We are basing our prices on Bitstamp. That should be easier for members to get involved in the project.

We have purchased the domain www.thewaspproject.com so if you are keen on helping build the site ground up join The Wasp Project today and we hope to see everyone at the EE briefing tomorrow. Lots of great news will be on offer I am sure from both Utah and Washington. Exciting times for us.

That sounds better than the 1 BTC price for membership you had before.
hero member
Activity: 924
Merit: 1000
Given the rapid increase in USD/BTC price we are setting our membership fees at $100 USD lifetime membership in the group and lowering the chip buy in for the A1 chips to 2 BTC. We are basing our prices on Bitstamp. That should be easier for members to get involved in the project.

We have purchased the domain www.thewaspproject.com so if you are keen on helping build the site ground up join The Wasp Project today and we hope to see everyone at the EE briefing tomorrow. Lots of great news will be on offer I am sure from both Utah and Washington. Exciting times for us.
hero member
Activity: 924
Merit: 1000
Update from the EE in Seattle

+ Full schematic/BOM ready for review for BitFury, A1 chips this weekend.

+ Firmware is progressing - the 8-bit AVRs are being used to emulate the various target hashers, by acting as slave SPIs and passing the work info to a PC. Right now, we're just reading and writing, but shortly we should be able to pretend to be a hashing chip by doing the hashing on a GPU and sending the results back... will be very slow, but will verify the firmware even before we get boards back.

+ 32-bit AVRs should be in today, and the test setup running shortly thereafter.

+ Please keep pressing Zefir / Bitmine for A1 data - there are a lot of questions critical to the design that Bitmine hasn't answered yet!

------------

Meeting on Teamspeak server will be at 7pm on Saturday Utah time with both EE's and support team. All members on the Zoho Project page are welcome to attend.
hero member
Activity: 924
Merit: 1000
November Prototype Wasps.


Avalon Wasp (Being Designed in Utah, USA)

Avalon Gen I Chips In Hand
6.4 Gh/s (16 chips)

BitFury Wasp (Being Designed in Washington, USA)

BitFuryChips In Hand
40 Gh/s (16 chips)


The Wasp could hold more than 16 chips but for the prototypes 16 will be enough to demonstrate the Hive and Wasp modular design and help us work out design issues before we get to the A1 and Minion chips in December and February respectively. We will have to Engineers working on the Wasps.
hero member
Activity: 924
Merit: 1000
Probably through Group Buys or DIY resellers for Avalon / BitFury Wasps & Hive (Late November December / January)
Probably through Group Buys or DIY resellers or potentially through chip fabricators for A1 / Minion Wasps & Hive. (A1 in December / January | Minion February)

Also this will be open source so you could always DIY as well once we are sure everything is functional. We will not be releasing anything until we are sure it is functional in terms of the Open Source Hardware of this design.

We could sell units through our website controlled by local members. Meaning local members could fabricate and sell units in your country and manage the logistics to cut down delivery time and customs issues. Although at this point though we are concerned first with having working prototypes for the Hive and Wasp.

Following up on the OP I just had a short conversation with one of the EE's and the design for the BitFury / Avalon Wasp could hold 16 or more chips. The Quad A1 is now more likely to be a Hex A1 or 6 chips per blade. With a 6 slot Hive so a fully populated Hive with 6 x Hex A1 Wasps could run 1.5 Th/s.

Note it is already "OVERPRICED" as one just has to check out the costs per chip in Zefir's A1 chip buy. Given the difficulty, cost of production, chip costs etc these units will still have one very significant advantage over any other mining systems in that any new chips that come out can be designed into Wasp's which can then be plugged into the same Hives already running say Avalon Wasps or BitFury Wasps. Turn around in weeks, after a new chip is shipped, to mining with that new variant of the Wasp. Modular is the way to go so you are not trapped into one type of chip from one manufacturer.
legendary
Activity: 924
Merit: 1000
Think. Positive. Thoughts.
When can I buy one? Smiley
hero member
Activity: 630
Merit: 501
Miner Setup And Reviews. WASP Rep.
Looks very promising.
hero member
Activity: 546
Merit: 500
Owner, Minersource.net
Happy to be a part of this project Smiley
hero member
Activity: 924
Merit: 1000
As the Wasp Project team moves closer to the working prototype we thought we would post up some details here of the 4 chip A1 Wasp currently in the works. Our modular design will be able to use any ASIC chip that is sold to the community past, present, and future. We will be using Avalon Generation 1 and BitFury chips (even a KnC chip) to test the Hive and Wasp capabilities and as soon as the A1 Bitmine Chips ship December or when the Minion Black Arrows ship in February we will have the ability to fabricate Wasps based on those chips in a very short turnaround time.

The Wasp Project is an open membership collective and we are committed to Open Source Hardware and Software and we are very interested in working with chip fabricators, diy'ers / hobbyists as well as miners looking to upgrade their "ageing" and "inefficient" systems. If you want to learn more or even become a member of the team please check out our Project thread. We should be testing the Avalon and BitFury Wasp (and potentially a KnC Wasp) prototype with the Hive in November and we will post more details in our Project thread here: https://bitcointalksearch.org/topic/the-wasp-28nm-asic-miner-open-hardware-development-project-299255

This is exciting since there is the potential, for the community especially those with BitFury and Avalon Gen I chips, to buy or build these units in a few weeks time given the prototypes work out well. We are hoping that our modular system will allow our community to develop a variety of Wasps as new chips become available and up-cycle your mining rigs using the Hive modular design.


Quad or Hex A1 Wasp System Architecture


Introduction

Our Quad A1 Wasp is comprised by five main sub-systems: the A1 hashing chips and their SPI interfaces, the USB interface, the firmware-controllable power supplies, the power distribution circuitry, and the debug, monitoring, and display subsystem.

The Bitminer A1 hashing ASIC is the workhorse of the blade, with 32 fast hashing engines and an SPI based interface for controlling them. Each chip has its own chip select on the SPI bus mastered by the xmega microcontroller.

Control of the entire blade is performed over the USB bus, which is managed by an Atmel ATXmega256A3U microcontroller. The USB connection is a client of the system's main computer/controller, where the mining software runs and connects to the Internet for pools.

Firmware on the microcontroller configures the buck power supplies to provide individually tailored voltages for each A1. The buck controllers also provide many safety features to protect the A1s during hot-plug or power supply failure.

Backplane voltages required by the Wasp's circuits other than the hasher internal power are all provided either directly from the backplane or through the use of LDO linear regulators. This Wasp must provide the VIO for the hashers and regulators, voltage references for the the A/D circuits monitoring on-board voltages, power for the LED indicators, and power to the connectors for diagnostics and monitoring (the optional local console & display). Almost all the ancillary voltage sources are switched and controlled by the xmega, to provide protection for hot-plug circumstances.

Finally, the monitoring and display subsystem is tasked with monitoring temperatures on the board and the power circuits, monitoring voltages, and displaying their various status levels through single- or multi-color LEDs, which can be pulse-width modulated by the xmega.

Voltages and temperatures on this Wasp are monitored through the xmega's A/D subsystem, and through TWI bussed thermometer circuits, as well as through individual sensors in the ASICs and the buck controllers. Provision has been made to attach additional daisy-chained TWI thermometers for the cooling system to on-blade headers. There are several LEDs on the Wasp's edge furthest from the edge connector, used to provide visual identification, board health, operational modes (hashing, debugging, programming, unplugging), and two push-buttons associated with unplugging and manual mode switching. Finally, as mentioned previously, there is a header for the attachment of an external diagnostic and maintenance console, and a firmware debugging and programming header can also be populated.

A1 Hashing ASICs

The bitminer A1 hashing ASICs have 32 on-chip hashing engines for mining SHA256 based eCoins. These engines can run at somewhere between 0.5 and 0.95V, and the higher the voltage, the higher the internal clock can be configured. The chips are nominally rated at 25GH/s each, at 0.65V, and up to 40GH/s each at 0.75V, if all 32 engines are functional. The seller guarantees only that at least 30 engines are functional, but makes no guarantees about any clock speeds or voltage survival beyond the baseline. Our implementation of the voltage controls and the clock controls performed by the microcontroller allow us to automatically characterize and then optimize the running environment for each chip individually, under firmware control. Since the controls can be changed using the communications protocols, a user can also re-configure his blade while it is hashing, in order to lower the power consumed while increasing the ratio of hashes to Watts. The firmware will also, if configured, manage the power and clock to protect the A1 from overheats caused by cooling failures, for instance.

All communication between the microcontroller and the hashing chips is conveyed over the SPI bus. While the A1 has pads to allow the SPI loop to contain more than a single ASIC, there is no benefit to that feature for us, and it involves significant costs in PCB routing - potentially even forcing a change to a four-layer board. Instead, we buffer the SPI bus to each A1, and simply assign a separate SPI select line to each. This allows us the firmware simplicity of talking to a single hasher at a time, and works better than slaving all the hashers to the slowest one's SPI clock or polling interval. It also prevents long shift chains that can tie up the polling system and lower the utilization of the hashers, because one that finishes while others are being re-initialized, must wait up to several hundred microseconds for its new job. It's also cleaner - we don't know what errata are involved in these new chips, and so treating each one independently gives us the best chance at having a fully functional blade, without very costly hot-air rework to replace a poorly performing chip.

The Atmel xmega series contains a USART which can be used as an SPI bus master with DMA, relieving the microcontroller of the duties of servicing the SPI inputs and outputs. This - along with a similar DMA capability on the USB port - makes the microcontroller very responsive to new commands from the mining software and enabling separate work queues for each A1, which are all moving at different speeds, potentially. We don't have to run at the lowest common denominator, and hence can get more hashing out of any given blade than a design that is limited in that fashion.
 
USB Interface

The particular xmega microcontroller selected for this design has an on-board native USB interface whose outputs require only external anti-static protection and noise filters to implement a USB 2.0 Full Speed (12Mb/s) interface. This interface is configured with two device descriptors, making the blade a compound USB device. One device is basically only used for normal mining specification of hashing jobs, and for returning any golden nonces that have been discovered. This makes the miner software simple, and allows it to conform to a single protocol without regard to the type of hashing chip that the Wasp uses. The descriptors also allow for self-identification, as required by the USB specifications, and can show the mining software what opaque configuration data block it might load, to initialize normal operation.

The second of the compound devices is a more complex management interface. It functions to enable FLASH programming, as well as overlay loading and invocation, and can duplicate the monitoring and maintenance reporting for voltages, temperatures, and power supply availability, etc. from the first device. Firmware licensing verification operations are performed using this interface, and it also provides an interface for running diagnostics that can override the first device's commands - or disable them completely.

Power Generation and Control

This model of Wasp, like every model, uses PMBUS-configurable buck power supplies to provide the variations of main internal operational power to the hashing chips. The IMVP-7 specification has resulted in many viable alternative buck controllers, since they are used primarily in the production of CPU point-of-load regulators, both for laptop PCs and desktop PCs where the graphics processor is implemented on the CPU chip. Because most graphics engines take the same low voltages as the processors, at nearly the same high current levels, there exist dual buck controllers such as the TPS59650 which can create two independent power supplies of the same voltage range, individually programmable throughout the range (0.1-2.5V, in 50mV steps!), at 40A current levels with very low ripple, and monitoring/recovery circuits for over-current, over-voltage, under-current, under-voltage, and synchronized soft startup - all for about $3/per chip, in medium quantities!

Quad A1 Wasp derives all its power needs from the edge connector(s) on its PCB. All backplane configurations provide the following voltages:



This provides drive power for the buck controllers, which generate programmable voltages in the 0.1V-2.5V range at currents around 30A for each section. It, too, is switchable by the Xmega, to support hot-plug. Higher amperage (60A) supply is possible with a second, optional power-only edge connector.

+5Vsb is used to supply all the power to the microcontroller, through a 3.3V low-dropout linear regulator. It is always on, unless the backplane power is completely unplugged, and provides constant availability for the USB connection. +3.3V is used to power the buck controllers, any level translators, and most LED drivers. +5V is the standard switched power that is used only for needs greater than the standby power can supply. It is currently unused on the Quad A1 Wasp. +12V is the main power source for the hashing chips, once it has been conditioned by the buck controllers. With four A1 chips running at turbo speeds, this rail will need to supply the blade with about 160-175W, or somewhere around 18-20A. Other proposed Wasps may use more power, and require the addition of a second power stage connector on the motherboard, bringing the total available power for +12V to 60A.

Power needs for the Quad A1 Wasp involve a few additional voltages derived from the backplane supplies. Low current needs are met with low-dropout linear regulators, while the high current, variable voltage needs of the hashers are met with switching buck regulators, one per A1 chip.

This type of buck controller is perfect for driving a pair of A1's, and the PMBUS is a well known specification for controlling them, which even has pre-built firmware libraries available from Atmel to serve as the basis of firmware development on the Wasp. The Quad A1 Wasp would have two copies of the circuit in Figure-1, below, which would share the xmega between them:
A1 Wasp Power Template

Figure 1: Template for Internal power generation and control for 2 A1's



The dual buck power supply circuit would be a slight simplification of the Texas Instruments reference design available from the manufacturer at TPS59650 Reference Design.
Pages:
Jump to: