Topic: Mini Rig card = 2 x Altera Arria II EP2AGX260 (Read 39231 times)

You're probably right.  I can't believe nobody ever said anything negative about BFL before I sent them $30k in mid-March.  I was excited and hopeful until I read your post just now, but I guess I'm screwed because even if I somehow luck out and manage get the two mini-rigs I ordered, BFL must STILL be a scam.  *SOB!*

My bet is they dont want to earn money now and use ASIC scheme to get more money later. It would possibly the best con job ever.

How many minirigs have they actually shipped? They already take preorder for ASIC and announce trade-in program.

Who told you this?

Maybe the low margins makes the mini rig ($15000 for 25 GB/s  -- $0.61 per M/h) good value for money product is to get people to pre pay.
So BFL balance sheet  very looks good for the vulture capitalist opps I mean venture capitalist. I would think the trade in is because there will be a high margin on asic rigs from day one and you would expect the new asic products to half in price after the trade in period expires .

I wonder what the quantity of Minirigs they have sold is? 30 would get them up to a quantity of 1000 FPGAs, which might be another price break for them.

Even then, they probably aren't making a ton of money on Minirigs if they're even paying $250 per FPGA. That would still be 8500 in FPGAs, and other components in the system would likely push it up closer to $10k. If they sell 30 FPGA MiniRigs, that would only leave 150k to pay for the space, keeping the lights on, paying employees, etc. There might be money in it, but not much.

200-500

Just for curiosity, what does 'moderate quantity' means (probably for such largish FPGAs)?

that means if you produce a stuff cost you 100$, and it's a low quantity product. then in a general way, sell it at 200-300$ will make ends meet.
it doesn't matter what BFL do. they could make the singles and minirigs at a unbelievable low cost, or just don't want to earn any money "now". who knows.

certainly not.
by experience, as a small amount product (bitcoin mining equipment will NEVER grow up to a traditionally BIG market you know), if the final price is twice as the cost, you will run at loss unless you careful calculation and strict budgeting, and it must be a easy to make stuff. sell the product at least triple of the cost will much better for a flat.

If we made some wild assumptions that these were it, and that they could get them for $300 each, that means:

$300 x 17 cards per rig x 2 chips per card == $10,200.00

Mini-rig price: $15,295.00 minus $10,200.00 == $5,095.00

So are the PCBs, supporting components, PSU, enclosure, and fans enough to eat that ~$5k? Seems like there could be a decent bit of profit here.

certainly not  , 600USD (this is only chip cost) for only 1.5GH/s is unacceptable.  plus 100$ additional charge, what is your expected final price for a 700USD cost product? 1000USD? 

and, A2 is a obsolescent product, A5(Arria V series, 28nm) is much cheaper (20%+)than A2 (at same capacity), but BFL didn't use the faster and cheaper one. maybe it's really hard for them to find enough 2nd hand A5 FPGAs.

for now though, let's  see what will happen on spartan6 this year.

From what I understand A2 GX260 are capable of about 750GH/S  ie 1.5 GH/s for 2 = bfl new single (use 17 of them in mini rig)

ie if they where sold as singles  at about $1000  for 1.5 GH/s ---

maybe you should consider using A2 GX260  in your next FPGA development

just contact with an altera FAE ( is on of my friends  ), he is unwilling to discuss about the price, but leak a message that the A2 GX260 is about 300 USD at a moderate quantity. 

And how much would a factory wafer respin cost from an old maskset, ngzhang? Do you know any numbers there?

OK, here are some grapevine news.
i asked a seller in the market today, who is one of my best friends.

the EP2AGX260 F780 is about 2000USD for a new one (supply is nervous), and 200~350USD for a used one (no problem for a 500pcs order).

 

So do I, after reading their FAQ about the Custom ASIC.

Would definitely bet on the latter.

Nice find. It'd be really interesting to to be a fly on the wall of BFL's offices to see their design flow. Do they design a product and source it, or do they wait and try to source an amazing price on a lot of powerful hardware, and then design a product around it?

Ah, now I see. Makes sense, then. (And gives me some ideas to non Bitcoin related projects!)

Yups, that's right. But someday, I'll be able to afford one (or two! ). Just thought I was missing something (and I was missing more than something!) on why they were so expensive when compared to graphics cards that someone once said to be almost a sort of FPGA. (Excluding mass production costs, of course)

No, not implying anything, just trying to understand why the actual design instead an 30 years old mainframe type of design where you plugged boards in a backpane.

Once again, this was just a thought on "Why not?" but, yeah, that would be the main idea.
Already answered also:


Tonight, I'll be able to sleep calm, now that the world makes sense!

Thank you all for your patience and I'm sorry if I bothered someone with my doubts, I prefer to ask instead of making assumptions, and as far as I've been able to search and read, I've never seen anybody asking this.
I also hope these answers to my questions become a reference for future FPGA designers.
It's great to be part of this community!

Actually, the 3.0 spec allows up to 150 watts, but there are few boards that support that.

Pre-3.0 specs are 10 watts all the time, up to 25 after boot, and up to 75 on request of the installed device, I believe, if what I read is accurate.

Yeah, but you're limited to 150w total across the entire motherboard. Same reason you're boned if you try more than two 5970/6990/7990 and aren't using powered risers.

Can't pcie slots provide up to 75W of power?

With the way the minirig is assembled, they couldn't make it that way with backplanes anyhow, it'd require actual cables.

A PCI-E-based board dedicated to mining would essentially just be an existing two or four Spartan 6 board with a serial to USB chip plugged into a USB to PCI-E host chip and all power supplied off a PCI-E 6 pin plug fabbed on a standard PCI-E board shape.

Backplanes are expensive. Reusing SATA is the cheapest industrial internal connection by far. Its cheaper than using DB9 serial connections as well.

As for you being unable to afford it... well, thats the way it is. These parts cost money. There is no cheap solution.

Maybe I misread you, but are you implying that a 15,000 dollar mini-rig could be made cheaper by not using SATA cables for their I/O (which doesn't actually use SATA, just the cables)? Or are you just wanting a PCIe FPGA that you throw inside your computer just like a video card?

SATA cables are far cheaper than a rigid backplane solution. They also allow flexibility in creating your own layout or putting the stuff in your own case, if that's what you want to do too.

As for SATA connector's that's quite interesting, but it leaves me another question: as far as I understand, there is no need of high speed communications to feed the FPGA/ASIC and it's always needed a PC to access the pools and feed the jobs, so, why don't they use the PCI bus inside the computer? Or even the PCIe?
Other designs I've seen in this forum, use Molex connectors, and I still don't get it, why not ISA/VESA/AGP/PCI/PCIe type connectors?
IMHO, it would simplify the creation of rigs with a new and easily expandable design.
You could have a backpane which would get the work with some very cheap microcontroler (even a PIC with Ethernet shield?) where you would plug in small boards containing only some kind of eprom to provide an unique serial number (so the microcontroler would forward the work to the proper chip) and the FPGA/ASIC itself.
Even the power bus would be standard and could be populated with fan connectors.

Sorry if the question seems too stupid, I'm just trying to understand why people chose the designs they have, which seem to increase greatly the costs, reducing the resale value.
All the designs I've seen are too expensive for me, as far as the cheapest FPGA I've seen, capable of mining, costs $300, way more the ~$150 I would be able to spend.

re SATA plugs: they're just reusing an industry standard connector. It has enough pins to run serial over, so its fine. External 4x SATA/SAS has often been ran over Infiniband connectors because its cheaper and higher quality than using eSATA, and I've seen other things reuse SATA plugs. Its not particularly a new idea.

You can't reuse ASICs because they're designed specifically to ONLY for this. Thats why its an ASIC.

BFL FPGAs COULD be repurposed, but BFL requires encrypted bitstreams.

Latelly, I've been seeing pictures of BFL's in this forum with an Sata connector.
I even saw one picture some one posted with lot's of connectors and, although I realised their purpose is to provide a communication link, the first thing I thought they were was to connect an hard drive and create hardware encryption like this one: http://www.cast-inc.com/ip-cores/encryption/sha-256/index.html
So, two thought rushed through my brain:
- Why can't we use one of this to mine?
- Why can't we re-purpose BFL's and *ASICS to do hardware encryption?
The latest thought would impact the resale value, right?

I'm wonering if it is considered by BFL to be a thought of miners to purchase the Singles and/or Mini-Rig to maintain some kind of resale value in the event that bitcoin has some hypothetical collapse (somehow). Specialized ASIC's would have no resale value whatsoever, whereas having high end Startic chips wouldn't be as bad.

Thats funny, I originally wrote 20, but thought that was too low for some reason.

eBay, the only place you can get $600 parts for $25 just because its an enterprise part and some IT department in some company somewhere is in a protracted war against their accounting department.

I'm not sure, but wasn't it 1.05GH @ 20W.
However, I understand what you want to say and I also have my doubts.
Let's see with what they really will come up. The technical data (e.g. MH/W) will show us what technology they are using....

LOL no not quite. I meant ringing up Altera and saying "Yo, we want 10K chips, start the foundry pls"

And my board was bought on eBay, that's how I got it cheap.

They also announced 1gh in 40 watts.

Yes, Altera has the Hardcopy program.
But BFL has stated to announce full custom ASIC's.
An Altera Hardcopy solution is not the thing which is named a full custom ASIC.
 

Doesnt Altera offer Hardcopy? BFL could very well be selling ASICs.... just SASICs instead of the kind we wanted.

Well if they have access to wafer chip manufacturing, then why would they manufacture FPGA when they can do ASIC ?
I think more likely they got a good deal on these chips somewhere.

I remember guy with Extraordinaire rig, had a part that was advertised for $2k officially, yet a phonecall got it for $600.

Yes, I'm sure they do.

Ah yes, and subscribing 

YABMTTF (yet another BFL message thread to follow)

I'm fairly sure based on compiling the comments made over a long span of time from BFL, that one or both of these chips are (probably unchanged stock versions) standard FPGAs that were spun up for another wafer run. I have no reason to believe that they scored an amazing deal from a distributor that had them in stock.

Where BFL stated? No.  If a direct admission of BFL is the minimum level of "proof" well your right.  All of them are custom manufactured chips.  Obviously.

You do? Can you show me where BFL stated that they didn't have the single chips manufactured?

Why would the timeline suggest the actual chips are being manufactured to order?  We know the Singles used off the shelf chips and had similar long timeline.

So if you look at BFL's timeline for delivering mini rigs, it would suggest that these chips are indeed being manufactured. And if this is the case, and they are not buying second hand chips, then the ASIC would not be a far stretch from their process for delivering products now.

Just a thought.

Well, $2k+ is maybe a little bit too high. On the Altera website you get the chips starting from $1.6k, without volume.
My experience is that you get a (much) better price for Altera FPGA's with a quote from the official distributes. However I doubt that BFL is directly ordering this chips over this distributors. Also this chips are or getting legacy with the Arria V series. So BFL is using for the second time special connections to get "old" generation FPGA's to a VERY good price to make mining products.

The announced BFL ASIC approach is fairly different to what we have seen from BFL up to know...

Interesting work.

Okay, maybe I see why, most places that sell these chips or the boards they come on have a price tag of $2k+ for these.
They must be getting a really nice deal of them. 

Then you probably have found out their source of chips.
I admit I don't have the same fascination of figuring these stuff out but good detective work.

Yup. The one I had linked to was a much smaller Arria II with only 45k LE.

I'd agree with you.
However I'm not totally convinced, as the chip size is a little bit different, the main processor is actually smaller on the altera FPGA board.
I'm not refuting that the techy data probably backs it up of course.

Edit: you changed the picture from your original post. It does look more similar now...

Compare this Altera Arria II EP2AGX260 FPGA against the 2 chips on BFL's Mini Rig card shown in https://bitcointalksearch.org/topic/m.906496 :




The marking near the bottom edge of the package (blurred in the reference pic) is obviously Altera's Lot Number and Trace Code. The chip's package is identical. Same 4 asymmetric golden dots in the corners. Same PCB traces under the green soldermask on the 4 borders. Same arrow in top left corner. This is an Arria II for sure.

Which one though? Well we know the Single is based on the Stratix III EP3SL150 (142k LEs at 65nm) and mines at 832 Mh/s. BFL claims a Mini Rig card mines at 1.5 Gh/s, so its FPGAs must have 1.8x more LEs, or about 260k... And there is one model matching exactly this prediction: the Arria II EP2AGX260 (257k LEs at 40nm), see http://www.altera.com/devices/fpga/arria-fpgas/arria-ii-gx/overview/aiigx-overview.html

Arria II being 40nm also makes it close to the 45nm Spartan 6 in terms of efficiency (about 20 Mh/Joule), which is exactly what BFL claims (25 Gh/s at 1250 W for the whole rig).