So.. we are starting to get some power numbers:
https://hashfast.com/second-tests/“We’ve got two dies up on this. Right now we are clocking it at 700mhz and .84v core voltage. It’s doing 248Gh/s on only 2 dies! Half a Golden Nonce.
Consumption is around 300 watts total and 78C die temp.
So... we are looking at ~600W total for around ~500GH? Obviously the 25% extra hashrate is nice, but I thought we were expecting the total power to be under 1J/GH.
I guess that isn't going to happen unless we underclock.
So essentially roughly the same power efficiency as KNC, shipped late and at higher cost.
So KNC wins, and will make much more money if there is ever a round #2 match...
Let's see if Cointerra gives HF an ass kicking at half the cost, they might even deliver by the end of the year
This is mostly about KnC and how they differed from HF and CT in the chip design and business models so it may be TL:DR for those who don't care but this is just my observations IMHO.
Hope to see better numbers but it's kinda different comparing KnC hardware with HF and CT as KnC's 28nm chip with 4 cores only runs at 110-125 Gh so 4 chips are necessary to make a 550Gh unit so there is no room for expansion in the case at all, especially with the huge heat sinks that they used as well as the circuit board for each chip. Even if they used water cooling it would be hard to imagine how they would cool just another two chips (having a total of 6 chips) and the boards attached to them and still fitting it in a 4u case.
I'm guessing that they got their machines to market faster because they most likely did their entire design on an FPGA then just ported it directly to an ASIC so it had some of the advantages of an ASIC in terms of power consumption and what not but they already knew that their chips would not run as fast as building an ASIC from scratch. So they ended up with a product that would fill a 4u case and max out at 550Gh (well 400Gh initially) with no possibility of expansion even though their controller card took 6 inputs and the firmware and gui on the beagleboard supported all 6 ports as well.
As some people who ordered multiple machines from KnC noticed that some of the controller pcbs had all 6 ports populated with headers and some not although they all worked so if they wanted to they could solder a couple of headers on and it would work to run another 2 chips. Who knows why they had some controller cards with only 4 headers and some with all 6 (cost saving on a 10c part?). I'm not an engineer or anything but have done a good bit of research just for the sake of due diligence, just ordering 1 of their batch 1 jupiters is like 3 months salary for many people so it's no joke to order multiple machines.
So, with that (feel free to correct me if I'm wrong), it looks like KnC put their chips on a circuit board with the VRM, and everything else to run the chip and most likely used a serial interface from each chip pcb to connect to their "controller" pcb that really was there to just basically act as serial a hub for the 4 chips (or up to 6) and used a custom gui built on a stripped bare beaglebone black (no hdmi, usb etc with only way to access the beaglebone is through the network connection) to manage the serial connections linked to their proprietary board. I don't think that the pcb which they used to connect the chips and BBB did much but flash some leds, act as a serial hub and allow voltage (and other) data and settings to be sent to the chips from the BBB. I'm sure someone with the right skills could connect those serial connections to a serial to usb adapter and just ensure that the right data is sent to the chips and it would work with a regular usb interface. The as far as I'm aware, "firmware updates" are on the BBB not to any eprom on the pcb or anything.
The bad part was that even using the highest recommended PS the Corsair 850w 80+ Platnum their controller pcb hub is very sensitive to any voltage spikes. As many people knew there were a good bit of machines that needed RMAs including mine. There was a power surge and even with surge protectors 2 of my PSUs got burnt out and also 3 of their cards. After almost a week of back and fourth "troubleshooting" it was determined that it was their custom PCB was the issue as when I plugged the BBB on to one of my other working miners that did survive, it worked without any issue.
Previous to that it was confusing as their board would still provide power to the BBB and buttons but whatever chip they used to communicate with the ASIC chips and BBB was blown (but BBB was actually not harmed) as re-flashing the BBB did nothing. I'm sure this happened to many others as well and when they finally issued an RMA it was almost a week later, then sending the parts back to them and back would mean almost 2 weeks or more of no mining. I was just lucky enough to realize that all 6 headers on their controller card could be used so I was able to make super long cables and connect 2 chips to one board and 2 to another but I did have to solder some more headers on a couple cards as those only had 4 headers from the factory. I was lucky enough to have purchased enough miners from them to get all the chips running, and still running so it shows I have 6 chips showing in the web gui and get about 850Mh which is just about right for 6 chips. So I only ended up with about 2-3 days of down time and not 2-3 weeks.
This worked great for them at the time because they were able to actually deliver a product very close to their actual marketed release dates, the bad thing is that they skimped on just about everything else and aside from tweaking the pcb for the batch 2 machines I think they knew their chips were going to be an EOL very quickly from just these couple of batches that were sold. For the price they could have included a PSU etc but there was no room in the box anyway so I think it was more about that and the way they had engineered the project from the first place.
Their 20nm chip may be different and perhaps they can get all 3000Gh on one chip or perhaps 1500Mh on 2 chips by building an ASIC directly. That is what I believe CT and HF did, which is the reason that they are getting the same speeds on just one chip allowing easy upgrades and lots more room and space to save on wattage and all that stuff.
It does look like they did have some talented engineers that built their chip but some shitty business models and sales execs who don't even know their own specs on their products or even researched their competitors' products. CT may have been more honest about the time frame but still they charged quite a bit.
On the other hand HF totally screwed all the B1 B2 and most likely all their buyers with just lies. You would think that they would have purchased miners from the other competitors just to see how they had built their machines so that they would be able to say exactly why theirs are better than the others.
When I asked HF about their chips and boards and also the issue about power spikes that may kill the entire board as it just has 2 pcie connectors directly plugged in and brought up KnC and how their controller was very sensitive to power spikes the answer I received was "well, I don't know anything about the KnC miners but I promise you [said in emphasis] WE OVERBUILT THIS PRODUCT [at the same time pointing and even enthusiastically tapping on the plexiglass as if it somehow it made the chip better...!?] THERE IS 15 POUNDS OF COPPER ON THIS BOARD and there is PLENTY OF ROOM TO OVERCLOCK AND IT WOULD NOT FAIL" because they used Seasonic power supplies, which is the OEM builder of the Corsair PSUs that I had connected to the KnC that died and it was the Platinum rated PSUs as well. When I said that he was saying that it was most likely that I may have used a PSU that was not superior to theirs but when I said that Seasonic BUILDS the Corsair PSUs he did not even know that was the case.... and the water leak was still not even wiped off, the worse part was that the case had thumb screws so they didn't even need a screwdriver to open the case, why didn't they just wipe the water off in the first place?
I just hope that their real engineers really did make a good product.
Wow! Five hundred gigahashes/sec per chip. With room to overclock. 
