I want to revisit the topic of Scrypt ASICs. The reason for this is two-fold: first, mining income from Scrypt mining with GPUs is basically at (or just above) the break even point right now for most of the major coins, and that's only if you happen to have someone inexpensive power. Second and perhaps more importantly, Alpha Technologies sent out an email stating they are now entering the final 60 days of development on their Scrypt Viper ASICs, and thus the call has come for the final 70% payment from those that pre-ordered. As I mentioned a while back, I went in 50-50 with a good friend on one of the then-25MH Viper ASICs, and while I was nervous I figured it was worth the chance.
The payment request is sort of a good news/bad news scenario, unfortunately. The good news is that the ASICs are nearly done, right? The bad news is that some of the stuff surrounding the final couple of months is sounding all too familiar to me. I remember the Butterfly Labs fiasco, and while they did eventually ship the products and they delivered the expected/promised performance, it took so much longer to ship that even the somewhat early orders basically got hardware that hardly mattered on arrival. Today, 5GH of SHA256 hashing power is practically worthless, and even with power requirements of "only" 40W or so (which is a far cry from their earlier target of <5W), at the current difficulty level you would net around $2 per month.
The reason I mention Butterfly Labs is that, despite better communication, Alpha Technologies is basically unproven as far as cryptocurrency mining hardware is concerned. And what really makes me nervous is that they've gone from 25MH to 90MH (5 to 18MH on the lesser Viper), and now they've pushed the limits of credibility by promising 250MH (50MH for the small Viper)! I don't want to look a gift horse in the mouth, but we're now talking about a 10X increase in performance over the initial specs, all "due to increased competition". The competition of course comes from various places: KnCMiner, a company that has at least proven capable of delivering SHA256 ASICs in quantity, is offering 300MH for $10000. That's good but other companies are already shipping, including Gridseed (from various sources), Innosilicon, and Zeusminer.
While it's possible that A-Tech has managed to improve performance dramatically, much of the increase comes from trimming "extra" features like an integrated LCD display and doubling the number of ASICs instead -- all with a higher power requirement and a larger chassis of course. Ultimately they are giving away a lot of "free" extra performance to batch one customers. If they deliver that will generate a lot of good will, but there's a second possibility. Let's assume you have a bunch of preorders, and you've been paid a lot of money but still have costs and you need to get more money to finish the project. What do you do? Well, you can't really tell your customers, "Oops...we underestimated our initial costs; please pony up so we can finish!" So instead you tell them that you're "nearly ready to ship -- please remit the final 70% of your payment." And to further entice everyone to pay as quickly as possible, you then nearly triple the promised performance of a product where you already increased performance by 3.6X!
I really, really hope that they aren't going to disappear into the night with all the money from their customers, but let's be real: it wouldn't be the first time something like that happened in the world of cryptocurrencies! The size and power requirements of course have jumped, so what was going to be a 1U 25MH system became a 4U 90MH system, and now it will apparently be something like an 8U 250MH system... and it will require nearly 2000W (via three 750W PSUs last I heard)! It's a gamble still, as are all of the other non-shipping Scrypt ASICs, and if I lose it will hurt -- but thankfully income from other cryptocurrencies has basically got me covered.
Now with that all out of the way, let's talk about the performance and cost of running these Scrypt ASICs compared to standard GPU mining. This is why Scrypt GPU mining is about to die (assuming the ASICs are actually delivered). We'll start with my standard mining rig (3xR9 280X), then look at the ASICs from Gridseed, Innosilicon and Zeusminer (these are shipping; Gridseed is readily available, and the other two are both shipping as well now), and wrapping up with Alpha-Technologies and KnCMiner (not shipping yet).
Graphics Mining (3x280X):
Hash Rate: ~2.1MH
Price: ~$1500 for a complete system
Efficiency: 2.8KH per Watt
Daily Income: $2.60 (LTC)
Power Cost: $1.80
Hash Rate: ~70KH per chip, 300KH per "mini USB miner"
Power: ~1.3W per chip, ~7W for mini USB (~60W if hashing 9GH SHA256 as well)
Price: $1079 for a 5.2MH 80-chip blade that draws 140W
Efficiency: 37KH per Watt
Daily Income: $6.47 (LTC)
Power Cost: $0.34
Hash Rate: 86MH
Power: ~550W (?)
Price: ~$12000 for a complete system
Efficiency: 156.4KH per Watt
Daily Income: $107 (LTC)
Power Cost: $1.32
Zeusminer Thunder X3:
Hash Rate: ~28MH
Price: ~$3100 (no PSU?)
Efficiency: 26.7KH per Watt
Daily Income: $35 (LTC)
Power Cost: $2.52
A-Tech Viper (250MH):
Hash Rate: 250MH
Power: ~1900W (?)
Price: ~$10000 for a complete system
Efficiency: 131.6KH per Watt
Daily Income: $312 (LTC)
Power Cost: $4.56
Hash Rate: 300MH
Power: Unknown? (1500W?)
Price: $10000 (no PSU)
Efficiency: 200KH per Watt
Daily Income: $373 (LTC)
Power Cost: $3.60
So first things first, in terms of efficiency even the Gridseed ASISs are about 13x what you'll get from a good GPU mining rig. The Zeusminer is actually the least efficient ASIC and it's still 10X what you'll get from GPUs. Move to the Innosilicon and A-Tech and we're looking at around 50X better efficiency, while if my power guess on the Titan is accurate we're talking over 70X more efficient mining compared to GPUs. When these ASICs arrive en force, GPU mining for Scrypt basically dies, and it will die fast. The only "good" news is that the initial ramp up in production means it will take a bit longer before things really escalate, and as I've mentioned in the past, GPUs can still be used for other algorithms -- Scrypt-N, Scrypt-Jane, X11, X13, etc.
There's another aspect we need to discuss, of course: how long to recover your investment with these ASICs? Well, assuming income remains constant (which is way better than any realistic best-case scenario), we're looking at nearly six months for the Gridseed, 113 days for the Innosilicon, 95 days for the Zeusminer... and if it really delivers, only 32.5 days for the Viper 250MH! KnC Titan would be the best result of course, given it's the highest hash rate, paying you back in around 27 days, but we're still likely three or more months away from the Titan shipping. Of course, by the time the Viper and Titan ship, I suspect profitability of Scrypt mining might be down another 50% or more, but even if it's 60 days to break even I'll be quite happy.
The only problem is whether or not the darn things actually ship in July (or 2014 if they pull a Butterfly Labs). Let's just hope that doesn't happen.... And if you didn't take a chance on a Scrypt ASIC six months back, now you can be sad... and glad at the same time that you're not out $10,000 and wondering whether you'll actually receive the promised hardware -- on time or perhaps ever.
If you like reading these blog posts but don't want to subscribe to my thrice-weekly newsletter, please consider making purchases through my Amazon Affiliate links!