Okay, after some thought I've decided to post part 2 of the interview with KnC in it's own thread, only because since posting it last night half-way through the other thread and with replies mounting on top and little views I'm assuming most are unaware it, and the YouTube links are there, there's a lot of people here with a vested interest, not to mention that after 10 pages of typing I think i've suffered enough internal bleeding of my two typing fingers...
Seriously though, mods, if you're not happy with this, or if this in anyway contravenes forum rules, or is not seen as relevant enough for it's own thread, please delete this part II, and not the original part I. I will literally cry if that goes missing, i've put so much effort in typing it.
I just figured it's easier for people to find, read, inform or discus this way...also Marcus fired me an email at 7am today correcting me on a few mistakes I made. Most of which I was aware there was likely to be an error, i.e. acronyms I couldn't fully hear. I kept replying those points, audio was sketchy - changes are in redEDIT: Videos can be shown in 1080p - though that's caused some problems for some members, so please check your video quality selection in YouTube first!First here is a video of Marcus discussing die and package size (dreaded portrait mode letterbox, sorry wasn't aware at the time):
http://www.youtube.com/watch?v=by-je8XRCdY&feature=youtu.be
So here is the next part of the Q&A's; Retro 72's questions:--
--
retro721) Did they design the Mars chips or buy them off the shelf? Sam: It's an FPGA
Me: Yeah...so you programmed it
Marcus: It's an FPGA, so you buy them off the shelf, you program them, we have proven the fastest design, possible, available is err...is running at 106gh, we get 130gh out of it. So obviously we redesign them to make them faster.
Me: Ok, that, that was something he hadn't asked about, but I was gonna, because we were looking at a 30% jump on the pushing the chip, so with respect to what you've done with the FPGA, how does that translate to the ASIC?
Sam: (Pointing at Mars) They're 30% higher.
Me: They are 30%?
Sam: 29% is about right.
Me: Ok
Sam: And we still haven't finished. We still haven't finsihed.
Marcus: No, no, no. We've...we have more to squeeze out of them, which we will do later on, just for the fun of it, eh...
Other forum member interupts: So 175, 350 is a minimum?
Marcus & Sam: Minimum.
Marcus: That's a minimum.
Other forum member: So it could go upto like 420 for the big one?...4...could...?
Marcus: Could.
Sam: Could.
Me: Let's not throw figures around. (General laughter) It's not worth the hassle...but err, so respect to...is that 6gh/s standard?
Sam & Marcus: No.
Sam: 48 Chips.
Marcus: No, no. 48 divided by 306. Calculator. That's 5, 5.1.
(EDIT:
48 multiplied with 106MHash (original design) is 5.1, we wanted to show that we can run faster).
Me: Ok that's cool, that's what I wanted to get at, only because you were saying 30% beforehand and said that 6gh/s.
Sam: 6.2.
Me: Ah yeah...6.2. No you said 6...
Sam interrupts: No we said 6 if we did use some of that gain within the 6.
Me: I didn't know whether the gain was included or not.
Sam: We said 6, knowing we could over achieve 6. So now we're running at..I think we had 6.8...
Daggeteo confims: Yup.
Sam:...earlier we were going to your wallet, right?
Daggeteo confirms: Yup.
Sam: And we haven't finished yet. We stopped at 6.8 to do these demonstrations (open-day, not first
Mars vid)...
I interupt: On a stable...
Sam: On a stable, 6.8, constantly. Now. When you guys have gone, we're gonna crank it up, scale the machine and squeeze it all in, and see what we can get out of it.
(NOTE: Sam spoke the latter part of that last sentence so fast, i'm not sure if he said “scale the machine and squeeze it all in”. It sounds like that. I've listened to it over and over and I cannot make out what is said)
Another member: You should do a burn out.
Me: You should film a burn out test on it.
Sam: If we have time.
Daggeteo: So how will we know?
Sam: Your wallet will let you know, put it that way. (General laughter)
Daggeteo: Hopefully...but I mean em
Sam: Yeah of course, but what the statement was six...
Me: Maybe when your wallet starts smoking...
Sam: The statement was six, that's all we wanted to achieve. We achieved the statement, anything
else is just a bit of fun.
Me: Yeah, no, the only reason I was labouring on that point was because in the literture you said 250, 350 for Jupiter, 17..err...
Sam interupts: You still have the gain on top.
Me: So...but you'd only mentioned 6 before for Mars, so I din't know what you meant wheather that was prior, or as you were in a rush to get Mars out, that you'd not, you know...achieve what you'd wanted to achieve, blah, blah, blah...ok...umm...I'm gonna just go through these...questions.
Marcus: Yeah, go ahead, you...
Me interrupting: Word for word, so, you know, I know you'd pretty much answered, basically
--
2) They seemed to imply they were Altera chips. Did they design them for Altera?Sam: You just buy them from Altera.
Me: Yeah, I know.
Sam: They're FPGAs
Marcus: They're FPGAs; Altera Stratix 4
(
http://www.altera.com/devices/fpga/stratix-fpgas/stratix/stratix/stx-index.jsp)
EDIT: (
They're FPGAs; Altera Cyclone 4
(http://www.altera.com/devices/fpga/cyclone-v-fpgas/cyv-index.jsp))
Me: I don't know where he was going with that...
Sam: No I...I mean sure, some people don't understand the FPGA chip. You buy the chip, you put the code inside.
Me: Yeah. I'm aware of that.
--
3) If they did not design the chips, how is the Mars prototype relevant to the asic designs? Marcus: Well...err...there was two big reasons, a) we wanted a proof of concept to show them that we can build this type of complex design extremely fast.
Me: Umhum
Marcus: Not six months, or four months.
Me: Good, what...
Marcus interrups: We did it in weeks.
Me: Ok good...could you...so have you got an actual day...
Marcus: We were talking about that, weren't we. When actually did we start? To me...
Sam: Actually start? April the...20th? I think it was six weeks man. From then you guys did your pre-study to when the product arrived with the chips actually mounted.
Marcus: So proof of concept to show the community this is what we can do, that this is our speciality. So that, we can design the FPGA faster than anyone else.
Sam: Six weeks to go from zero, to what you see in front of you.
Me: Yup, that's cool...
Marcus: We need to show proof...
Me: Yeah
Marcus: The second reason why it was that we needed something that can verify the ASIC code on, cos when...since we are going directly to ASIC and not the
NPW, no prototypes from now, nothing, we need to be 100% sure, not 99%, we need to be 100% sure that the RTL code and every single line of the RTL code that we have done is 100% fully tested. The only way to do that efficiently is to run the same code on an FPGA. We can run it slower, that doesn't matter, but we need to make sure that we run it on an FPGA live against the err..CGMiner because we get billions of test vectors running through that FPGA and that would take us decades to do the simulation, to have the same data. So we are running simulation in parallel of course, but we are able to use the Mars as an ASIC prototype, that's what every semi-condductor guy does today. You put many FPGAs in parrallel, yo try to verify as much of the design as possilble into the FPGA. Luckily our designs are possible so we can test 100% of the RTL in an FPGA...
EDIT: (
"NPW" should be "MPW")
Other forum member interrupts: In an FPGA?
Marcus: Yeah.
Sam: Which you've already done.
Marcus: Which we've already done...doing...yeah. I mean normal guys, dual ARM etc., etc. say you can never squeeze in exactly that ARM processor in the FPGA, so you will always have proportions of the design that you can't test in an FPGA, and that's when the risk goes up and you have to do a LBW...
EDIT: (
"LBW" should be "MPW" (Multi Project Wafer))
Me: So that there was going to be my next so where are the weak links in the chain basically?
Marcus None (laughs)
Me: So...your, your that confident?
Marcus: Yes
(General laughing)
Marcus: Yes, yes, absolutely...
Sam: We have tested absolutely everything. We are over-cooling it, we are over powering it; 1000w
Marcus interupting: Oh yeah, we have margins on each.
Sam: We have margins on margins.
Marcus: we have margins and margins on margins just to make sure that we are reducing the risk throughout the whole cell. I mean we have redesigned the RTL code for the ASIC in a way so that we make it easier for the backend guys to do the backend to reduce risk.
Andreas: The devices have to run 24/7 as well. So we have to be confident that when people put the device on the shelf, can run it not for only one day, but hopefully a year down the line.
Sam: In Dubai, in an apartment, without the air-con on!
Andreas: We give them something that is maximum of everything it might work for a while, but then it might break because it was too hot, so we've taken a lot of margins.
--
4) Did they design and fab the PCB?Marcus: Yes...well...it depends on what you mean by fab? We do the PCB schematic, we also do the PCB layout, but we do not manufacture the PCB board itself. That is done in ASIA so, but we have in-house PCB expertise doing the full schematic and layout.
--
5) Is the software they developed for the Mars, directly applicable to the asics?Sam: Which software is the question relating to? CGMiner is not our software. I..I..I mean I guess the FPGA code is our code inside. The software is CGMiner. We've written a driver for Mars to work with CGMiner.
Marcus agreeing: Yes
Other forum member: I think he means the FPGA code because if thats appliciable to directly anything
Sam and Marcus: Yes and no.
Marcus: It's a configuration, I mean in that FPGA we only have one core instantiated, of course in a very very large ASIC we have multiple cores.
Other forum member: Yes multiple cores
Marcus: Basically, like I said, we have designed the RTL code in a way to cut down lead time to our backend and fab time that is different from an FPGA code that we have been running, but we are veri, we are using the Mars product to verify the ASIC RTL code to it's full extent. But you can't really do the same, when you design something to be specifically targeting an FPGA, you need to rewrite it. It will not be as efficiently implemented in an ASIC and vice-versa. So we have two different sets. We have one that we are showing the performance enhancement of the Mars, that is FPGA based and we have designed that code to directly fit into the FPGA. You then have the ASIC code, that we are verifying on the Mars product, but that's a unique code, we are verifying it.
Another forum member: Ok, just verifying?
Marcus: Verifying.
Me: I'm really looking forward to typing this up...
(General laughter)
Me: But how...how um...are you working closely with CG Miner? I mean how...
Sam: Well, no, we didn't we didn't just want to say CGMiner, we are working with lot's of developers of mining software, we're not going to put one above...
Me: The..the only reason I ask...is...in terms of your verification issues, as it were...um....it's always great to hear, well for them to hear (forum members) to hear that, that these other respected known members will say,
Sam: acknowledges: Yes, I agree...
Me: “Yup, we've seen this...that they are working on this...”
Sam: Yup, we've contacted the major developers, there on board, they're gonna be given our boxes over the summer that will be functionally identical apart from performance. And then when the chips arrive in September then they will be given the boxes appropriate for testing on.
Me: Is there a reason...is that ah...ip reason that you don't want them to see such things as...
Sam: No, no, no, basically we want to give to them a box that looks identical, to Jupiter, behaves identically in every single way apart from hashing speed. So, they can get their interface, they can get their code inside, they can get their automatic firmware (unknown word? - poor audio) processes working with us, they can do all that kind of stuff, and then we'll be using that for product stuff, like airflow, cooling and other things. And then when the chips do arrive, we will put two chips in the right board, and send them out to them as well. But we don't want to...
Marcus: So we order... (something I can't make out as Sam speaks again – he's physically closer, so louder)
Sam: Exactly
Marcus: ...we've all ready got the surrounding components, power, dcdc modules, the on chip linux software, all that is already developed, tested, ready to go.
But obviously we need their code to work on the box as soon as it's delivered. I mean, it's unfair to ship out a hundred boxes and say, “Sorry guys, you need to wait for CGMiner to finish”, so they can do all their testing with the other boxes first.
–
6) What problems did they actually run into with the Mars? Sam: We just wanted to over achieve.
Andreas: It was more umm...getting the code in...and
Me: There were no problems...?
Marcus: Well it was more CGMiner...hiccup
EDIT: (
Well it was more our CGMiner driver ....hiccup)
Me: ...and spelling mistakes?
(General laughter)
Me: Have they been solved? Have you been able to fix your spelling mistakes?
Sam: Only, the board still works, right? I mean c'mon...
(Holds up vacant Mars PCB. General laughter)
Sam: ...plug in all it's little things, I mean it'll be fine, it'll still do it's job. I think I might use it as a mouse mat...it's quite an expensive mouse mat...hahaha!
Me: Is there going to be loads of dodgy green paint over the...
Marcus: We need tipex...
(General laughter)
--
8)Do they translate/ have any bearing on the asic design? Me: I don't know where this...
Sam: I guess you need a previous statement.
Me: Well yeah the, the last phrase...
Sam: Um the errors...well there were no errors so there was nothing to translate to the ASIC design.
--
9) Did they learn anything significant from these issues?Sam: All of the coding...
Marcus interrupts: Spellcheck!
Sam: Spell check, please, let someone else do it...umm...well ah basically it's just proof of concept of the modular design, wasn't it? The airflow going across the top...
Marcus interrupts: I mean, of course, going back, we're not just sitting here saying that; everything we tested worked the first time, like always in development you test something, and argh it doesn't work, why? You go in and fix it, you hook up the
logic (
nortic?) analyser, etc. You say, “Ohhh...we err misplaced the
pin p-mapping on the ah FPGA, etc., etc...normal development.
EDIT: (
"logic analyzer" not nortic, change "p-mapping" to "pin-mapping)
Other forum member: Since you guys did six weeks, no major problems?
Marcus: No major ahh...because otherwise we wouldn't have been able to deliver it in six weeks. You know, that's our statement. Try to beat six weeks...
--
10) They seem to be significantly cheaper than all the other options out there. Why is this?We need to compete against global hashrate. We don't really have a direct competitor on cost at the moment, and so; we know our devices will arrive in September, and unfortunately we need to take the money now, and give you a competitive product in September.
Me: So this is ...this is...your assumption that it, with all due respect to what's happening with...with hashrate, chance are you're still gonna be...
Sam: Making money for people, yeah. We need you to make money from our products so that you buy a second one...
Another forum member interrupts: So you can make money on...
Sam: So we can make money.
Me: I think...I think where they are getting with that, say with respect to ASICminer, right now, who are milking as much as they can at the moment...
Sam: Right now, they are the only product you can buy, really significant product you can buy, and sure, we would love to say; “here is a USB stick, with an ASIC on it, buy it from me, seventy-five coins”...and I could probably sell it, but I'd much rather we concentrate on this and give a longer term approach and a more reasonable price tag.
Me: Yeah, ok, umm....
Sam: Although, it's still quite a lot of money.
Marcus: Yeah...we...we want to be in this for a long run...
Me: Yeah.
Marcus: We...we want to be in this game for five, ten years.
Other forum member: Considering all that money could've been tied up with BFL pre-orders, you made a good move.
Sam: I think so. We came hard. Not with price, but perfrormance and dropping to 28nm straight away, not developing a 100nm product. I mean we're going to the very front..,
Me: I mean, you're not undercutting and squeezing your margins to the point of non existance, are you? I mean you're still going to...
Sam: We still have to make money.
--
11) Will their prices be fixed, or will they follow the trends in the BTC price/ROI?
12) Will the prices be raised after this initial pre-order period or are these the final RRP?We're not going to raise out prices...not, not...
Marcus: No. No.
Sam: We've no intention of doing that now.
Me: I mean I know from a couple sort of email conversations with you...
Sam: Yup.
Me: ...err...you've happy to expand the product entry
Sam: Higher, and possibly lower. Um...we do not want, or have any intention of competing with thee $50-100 market...
Me: Well how about $500, that seemes to be...
Sam: Ahh we maybe...maybe down to something like that.
Me: I mean that...
Andreas: It's about support level as well, because we want to be able to operate good support, and if it get's too many people...
Me in agreement: Yeah, yeah...
Andreas: ...we might not be able to handle the support load.
Me: I think a lot of people want to see something around the $500-$750 for instance because that's in the range of two graphics cards...which is what a lot of the common miners are.
Sam: Yeah, yeah, but...and, and we know, if we do, if we drop our...if we do create a product at that level, suddenly we have 30,000 customers, instead of, you know, 2,000 customers up here. We'd have to scale that massively, and it becomes a much larger, larger accommodation. What we might me willing to do is have a Jupiter, or a Saturn and actually split it so you can purchase one from us, say you guys, you wanna come up with a company, you buy 10 jupiters...
me: and distribute...
Sam: ...and chop them into pieces and you sell a piece of Jupiter, because we're gonna see that next year, you're not going to buy much more than a 175, or a 350, because...
Other forum member interupts: Wh...wh...wh...what would a $500, $700 machine make...in, in January?
Sam: Exactly. Not 175, but if you took a larger machine that was say $10,000 or $20,000 dollars then...
Me: You have the economies of scale and then you just split...?
Sam: ...and then you can split that up, and then you guys can buy one of those resell it around your neighbourhood, or to whoever you want and sell out hashing power. So that's the way we kinda see it going...
Me: Ok
Sam: So in theory we see our prices changing, not dollar per hash changing too much, we will always try and be competitive there, but we see our entry prices actually getting higher...
Me: So these little coffee cup warmer things we see people...
Sam: We're probably not gonna bother with things like that, no. No.
Me: I mean, one question that hasn't been asked at all, which I think is crucial to how we've seen with everything operate at the moment, is what are you doing with customer service? How are you gonna be handling...because I don't know how you are going to handle all the orders you've got right now...but we see the clusterf**k that BFL are dealing with at the moment and...um...and they can't, I can't see how...god knows what their email must look like, but yeah it must be going nuts!
Sam: Yes, yes and there is a, there is a good reason why we're not selling $200 products is basically because of the customer service and support that that will generate. We're going to try and stick to the high end of the market, so they are basically the business sellers. The people who are seriously wanting to do it. The customer service there is less, if you like, and we are going to scale to meet that. We've already got more customer services guys now today, that have started.
Me: Do you also assume that, or do you expect your customers to have a certain level of...
Another forum member: ...IT intellect?
Andreas: We expect a proportion of them will have some kind of what is involved about...
Me: Some things will need to be sent backwards, and forwards, which is the case with you know... (Avalon).
Sam: There certainly will be a case of...err we will do what ever we can do remotely. Absolutely.
Me: Yeah.
Sam: Um the device... Jupiter's effectively two devices, and if half...if one half of it breaks you just ship us that half back, and we'll ship it out...um the other idea we have to make our devices a little more friendly is they're gonna come ina, in a box, Marcus calls it; “a fence”, cos it really is to stop your fingers going in, and not much else. There is gonna be a case around it. It's gonna have some power connectors and a, an an ethernet connection, and the idea being is you will come back to our shop later in the year, put your wallet address next to your account, we will then preconfigure that box, with your wallet address, so when your going to a mining pool that you choose from a drop down list, so we will automatically code the little Linux machine, so when it is plugged in, turned on (Shows chip board), that little machine here, it will start hashing against that wallet, via that mining pool, via that worker that you've specified.
Me: So it's plug, and play basically...
Andreas: Yup.
Sam: We're doing the hard work, we're doing the software, we're doing the web interface, why would we not make it easy as possible?
Marcus: Plug in mining.
Andreas: It's supposed to be a self-independent box, there shouldn't be a need to do a lot of stuff, around it. Some people are concerned that they; “Oh so I have to be a geek to run that one” but, but that's not the idea. We wanted it to be something that you plug in, and then you have it run.
Me: No, I mean, in terms of how I see it, the way I have seen other companies behave Avalon have in no uncertain terms in the Bitcoin 2013 thing, said, you know, we love being the tech geeky hardware developers, we don't want to have to deal with customer service, so that's what you're getting. If you're happy with it, you know. BFL can't handle, their order volume must be...insane...and umm, yeah, no, it's gonna be a mixture of...of...of...ability levels.
Sam: Always is and language levels. Literally, an English manual would be useless to most of our customers, so they can't Google translate that...um...but online videos, and demonstrations are much easier, and the simpler we make the product, with the simplest web interface possible, for the, for the simple members, but also an SHA access for those that really want to play with the Cgminer, you know with some more slashes and stuff, of course you will as well as well.
Me: So presuming you two (Sam and Andreas) are front of house...?
Sam: Pretty much...
Me: So how hands-on with customer service will you be with that in that case...how will you be able to cope with that?
Sam: We will be overseeing the teams that deal with the customer service underneath us.
Andreas: We...we have previous experience, not for the Bitcoin area, when it comes to support, but we have previous experience from the finance industry, from the IT industry when it err...takes how to handle support it in an efficient manner, and how to set-up support, so we can focus on the right things and give out information when it's needed.
Me: So, on the record, you are comfortable with support issue and you're going to do whatever you can?
Sam: We're going to scale towards it, yeah.
Me: Cool.
Sam: But...we're not going to sell a product so cheap that it attracts too many customers that we can't support.
Me: Yeah.
Sam: That...that's the important distinction. That we will target our customer support.
Andreas: We've got to think of our phone as well, if it's constantly calling about things that …
Marcus: But what can go wrong? Plug the ethernet in....
Me: Noo, don't, don't go jinxing yourself...you've got that on record now!
(General laughter)
Sam: We do...we do get questions like; “what is an ATX power supply?” So we do get questions that we need to answer at a basic level.
–
Mars Video:
http://www.youtube.com/watch?v=BgNPIY9L8s4&feature=youtu.beEnjoy!
(Note: This
has been freakin' exhausting!!)
