Timelines, delays and why shipping is only now happening in June!
by, 06-07-2013 at 12:08 PM (29115 Views)
Ok, so after seeing the almost incessant carrying on about Jalapenos being shipped before Little Singles/Singles/Minirigs/the kitchen sink as well as people complaining that it has taken so long for BFL to deliver anything, I've decided to write this blog entry in the hope it might get a few more people to understand why we are at the point we are today.
First up, I work(ed) in IT, doing all sorts of things from helpdesk to business analysis and project management. I have worked as a coder, designer, script following helldesk monkey, 2nd/3rd level support and resolution, system architect and a whole pile of other things. I have managed projects that have gone well, and projects that have been complete disasters. I have applied what I have learned to how I've seen things progress at BFL and below is what I think has caused the delays and issues we have seen. These are my own opinions, based largely on what I've seen on these forums, in the shoutbox, on grnbrg's twitter feed and on IRC.
First up, we have the product. Development of a dedicated device using non standard parts to mine bitcoins faster than anything else that currently exists while using less power than anything that has come before it to do the same task. BFL has had some experience doing similar, develping a dedicated device that mines bitcoins using an FPGA as the core processor. Faster and more efficient than a GPU, but really little more than a cut-down developer board for the same chip that many other devices use in the world. Requires development of a nifty little case, custom board and some firmware to run the thing. Total requirements: Engineer to make the box, electrical engineer and/or computer scientist to work on the FPGA, firmware and board design. Potential risks of supply are minimal, risks of the FPGA being unsuitable are minimal as it can be reprogrammed easily. SHA-256 engine design can be purchased from 3rd party, further minimising timeframe.
Lets now look at the ASIC. Also requires a nifty case (easy). Requires redesign of engines to take full advantage of ASIC design and minimise power consumption while maximising hashing output (difficult). Requires board to host ASICs, which is dependent on design of the ASIC itself and can not really be done beforehand. Requires firmware to run the whole thing (easy) however this requires at the very least an idea of the board design and ASIC specifications.
Critical path: develop ASIC, develop board, develop firmware.
This is what BFL has pretty much done. The original timeline for the project may have been about four or so months, assuming preorder was when the project kicked off. Where things became unstuck was when things went wrong.
ASIC and board design can to an extent be run in parallel, once you know the physical shape and number of pins on the ASIC, as well as what the pins will do (power, data, ground). This would have been going along just fine, until it turned out that QFN packaging was unsuitable for the ASIC that had been designed due to overheating. I'd guess a crisis meeting would have taken place which would have discussed options on how to proceed, such as:
cancel orders, refund everything and stop the project
proceed with QFN design of ASIC and lose the targets on power consumption and hash rate
replace QFN design with something else (BGA) that will overcome the physical constraints.
design a whole new ASIC that will work with QFN packaging.
All options would have had their drawbacks. Cancel orders and refund everything would probably have meant the end of BFL, if not at the time in the future when competitors developed working designs. Not to mention the loss of reputation. Proceeding with the QFN design would allow a product to market sooner, however it would be well below promised spec (cue class actions, refunds, whinging, mayhem, etc). Replacing the QFN design would require a rework of the current wafers (if even possible) and a complete rework of the board design. Any existing boards would be useless and money would be lost on any parts that could no longer be used, as well as requiring additional time to repeat the development processes. Designing a whole new ASIC to use the existing QFN packaging would take even longer, have no guarantee of success and might not have even been possible with the technology available.
BFL proceeds with the change in chip design. Existing board design is thrown out the window and started again, although many chunks could be reused. New boards need to be ordered from the supplier, and new chips need to be fabricated. Decision needs to be made on how to test the modified/new wafers, testing before cutting and mounting will cost an entire wafer, but will be faster than waiting the whole time for the foundry to do its thing. Foundry and other external suppliers already have existing commitments and will only go as fast as they choose to, no matter what BFL does. Time progresses and finally after almost having to restart the project from scratch (or at least half way through), a product is finally built. It gets tested, but doesn't perform to spec. Power requirements are a lot higher than predicted for the promised hashrate. This is bad news for the Jalapeno line which is suppose to be USB powered.
Further crisis meetings occur, except this time for the boards. Testing shows the USB powered jalapeno will not work, it just pulls too much power. Parts of the board are getting hotter than they should and nobody can figure out why without doing even more testing, all of which takes more time. The USB powered jalapeno is scrapped and the little single external powered design is used instead, just with fewer lower spec chips. This will cost more but at least it will work. Boards are reworked to suit
Little Single, Single and MiniRig designs are determined to require a redesign of board to support more chips run at lower clock rate. (Not having more information available, I'd guess this is to do with inefficiencies and heat generation when running the ASICs at the highest possible speed). This means yet another board redesign, not to mention further fabrication, sourcing and all that is involved with essentially making a whole new product.
Instead of the original four month timeline, the project has now blown out by a factor of two with the board and ASIC redesign. Add to that delays with the production of the chips (at least another month) and discovering the end product still did not meet original design specifications (a month or more on trying to get the power issue under control). Doubling the project timeframe brings us to February 2013, adding two and a half more months, to April/May 2013. Now add the discovery that a whole new board is needed (add another month, the longboard looks like two shortboards stuck together so redesign should be faster) and you get to June 2013. The Jalapenos, while no longer USB powered, fundamentally work at the hashrate promised. BFL is now way over their original timeframe.
What do they do?
Left, right and center you have people, customers and competition calling you a scam. Customers are asking for refunds or completed products. There is only one way out of a situation like this if you want to survive and that is to deliver what you have got, even if it isn't the whole end product. Therefore, BFL starts shipping Jalapeno units, even though the LS/S/MR aren't ready yet. The competition is already shipping product and new competitors are on the horizon, so it is the only way to claw back market share and restore some semblance of customer faith. By shipping the Jalapenos you silence the people screaming scam, and also put a product out to market that will attract more customers, as higher order numbers ensure you can order larger batches from suppliers and get them more quickly. It also ensures that customer complaining is tempered by those who now have devices to hash with. While proceeding with Jalapeno production and shipping, continue work on the MR as higher priority (as cancelled orders of these will hurt a lot more), followed by Single then Little Single.
Total Trackbacks 0