Topic: [DB.RCLMR] DeepBit "Reclaimer" ASICs - page 4. (Read 46423 times)

there ARE going to be big mining rack farms, so someone might as well cater to them. when you're looking at mining operations as a business i could see these being useful after a certain point.

I concur, I allready have a PC, so I actually don't wan't the miner to be able to mine standalone, what/who is the intended audience, multibillion rack farms!?

1U and 2U rackmount case hardware is *EXPENSIVE*, with 3U you can still use standard heatsinks etc. Maybe with some luck some models on 2U.
A proper low height heatsink is going to easily cost 60$ alone.
Then add a properly sized PSU there, and your 750W Corsair is starting to look cheap.
Then for lower height you need to use smaller diameter fans which are by ORDER OF MAGNITUDE less efficient, infact, in modern 1U server the biggest electrical expense can be at times the fans!
Because adding depth to fan is not efficient, but a must do in small diameter, and to drive sufficient airflow you need to spin it at really high rpm these fans can  use considerable amount of electricy to even reach half of what an 230mm fan @ 2.5W can do. Some say these might use even upto 10-15W *EACH* and you need like 4 of them (that's why they are also RPM controlled)

Since depth of a rackmount chassis is not set you can make half depth chassis, and you can load any rack in standard setup DC from both sides, mount them from both sides and drive the hot air to the center of rack for gravitational dispension of heat by rack top ventilation hole, creating a natural rack zone hot and cold aisles
How well it works depends on the DC, some DCs don't have hot and cold aisles, others does. I would use one without hot + cold aisles for this setup other than which sucks hot air directly from top of the rack and keeps everything else cool

Ofc, to get it to work efficiently you need fill up the whole rack, which means that even on 42U rack you need 27 (1 slot for the switch) of these, or atleast near so putting separator plates makes sense
That's 75600$ investment per rack for a total of 2.16TH, at a operating cost of roughly 500€ a month depending upon location, electrical rates etc. plus network capacity. many DCs include 10-100mbps with the rack, not really free but "cheap", and backup connection probably will cost about 30-40€/mo for crossconnect + 50-150€ a month depending on provider, how low they are willing to go.

If someone wants a low cost location for several Us we should have spare capacity in near months

Reclaimer RM will contain same boards as smaller devices, plugged into a backplane. Technically it's possible to shrink it down to 2U, but 3U would also allow usage of common ATX PSUs with top fans when more powerful devices will be available.

Note that this 3U device is expected to provide much more hashpower with additional modules installed - not all the space inside is occupied at 80 GH/s.
All Reclaimer devices are designed with expansion in mind so if someone releases faster products we can just switch to soldering more chips on same modules or plug more modules in RM.

RM enclosure design is not done yet, so changes are possible. I can agree that 2U may be more suitable for collocation or filling 19" racks entirely. We will check if mining module, connector and backplane fit safely in 2U height and post updates in case of any changes.

I saw this detail. But isn't it irrelevant? All you should need to drive 2 x Single SC is a wallet-sized ~3W Raspberry Pi.
My question is why would Tycho need a full-blown 3U enclosure for similar hashing capacity?

I am genuinely curious and trying to learn more about the product to evaluate my investment choices.

I think you're missing one small (but important factor):

Are you really going to need 3U of electronics/fans/etc for 80 Gh/s? BFL supposedly will be able to do 80 Gh/s with 2 tiny Single SC (which will be the same size as 2 current Singles), which suggest BFL will be massively more efficient in terms of Mh/Joule...

I'll give you #1, but 2 and 3 arn't really that true.

Compare to 2 SC Singles:
80GH/s for $2600 - requires a PC to connect to.
Or the Reclaimer   RM:
80GH/s for $2800 - does NOT require a PC to connect to.

Those 2 smaller ones tho? The 4 and 8 GH/s units? Those might be hard to sell at those prices.

$1299 / 40 GH/s = $32.475 per GH/s
$2800 / 80 GH/s = $35 per GH/s
I wouldn't say that this is "considerably more expensive" or "claim of performance lower".

But I still prefer giving minimal numbers yet to be on the safe side.

I think the issue the others are trying to make is that at present:

1.  Your device is estimated to deliver at least 3 months after BFL.

2.  Your device is considerably more expensive then BFL.

3.  Your claim of performance is lower than BFL.

4.  You wont provide any more details than BFL.

I accept that BFL has made claims which they currently wont prove, but at this stage, if you want to compete, you need to either claim a better/ cheaper product, or wait until BFL fail to deliver.  Given the two options I think most people would currently pre order from BFL.

Even with suboptimal designs only bruteforce tasks are possible.
You can't use it for general SHA calculations because that would require completely different IO for high bandwidth: you'll have to channel gigabytes per second in both directions.

Let's imagine that we want to mine bitcoins on a general purpose SHA256 device. Without using midstates we should load 80 bytes of block header first (two chunks of 64 bytes each), then get first hash result - 32 bytes. Then we load it again as a 64 byte chunk and get our result as 32 bytes. This equals to 256 bytes of total I/O per one dual hash, and just for 1 GH/s that would be... ~2.048 terabit per second.
May be there is a little mistake somewhere, but anyway making general purpose hasher and bruteforce hasher is not the same.

If You don't start with the 3rd round and have the option to send the output of the first complete sha-2 pipe you can use the chip for general sha-2 calculations. this is suboptimal but only a small overhead with the advantage of a new market. But You already have the design and probably MPW based results so it is too late to change anything.

Yes, I have simulation results, but I wouldn't like to publish them because this estimation may be incorrect.
AFAIK our competitors aren't disclosing this kind of information too.
Of course it's much more effective than FPGA products, at least.

As for more technical details: yes, more information will be added.

I can't imagine how this can be done with a custom ASIC product. Either it's suboptimal design consisting of generic packet hasher and separate nonce incrementation module (would only work on SHA256-hashes anyway) or it's not completely truth.
May be their representative was talking about something else.

I remember their announcement of BitForce product that mentions "medical image processing", "Monte-Carlo method" and so on, but that was FPGA and it's technically possible with appropriate bitstream provided. Just as with any other FPGA board, but without GPIO pins.

Presence of competitors is NOT a reason for me to not do anything.
If some of them will start shipping I'll consider changing specifications of our devices, but now it's too early for that.

Making different ASIC products is a good thing for protecting the network, not to mention the competition itself.

Not sure, just asking regarding other uses, BFL once stated that their ASICs can do certain type of password cracking which is similar to  SHA256 processing, but they purposely disabled it so they don't run into export limitations.

What kind of passwords ?
No, this is a specialized device and it can be only used for mining, proof-of-work generation and PoW signatures.

Will the ASIC support password cracking ? this would be a major benefit over other ASICs

Looks like the 80 gh unit is going to be bigger than BFL's two SC + a working laptop?

+1