Bitmain to release Antminer F3 Ethereum ASIC miner with 72GB DDR3

P

pclausen

Gawd
Joined
Jan 30, 2008
Messages
697
According to Technews, Bitmain is about to release the Ethereum ASIC miner F3 in Q2 or Q3 2018.

“Bitmain is about to release F3, the ethereum ASIC miner. It’s reported that every miner is mounted with 3 mainboards. On each mainboard there are 6 ASIC processors, each of which has 32 1GB DDR3 memory. Therefore, one unit of F3 miner contains 72 Gigabyte DRAM memory.”

Read more about it here:

https://bitcointalk.org/index.php?topic=2932282.0
 
ryan_975 said:
Moving to proof of stake requires a fork. If there are enough ASIC ethereum miners by then, the fork could be ineffective.
Click to expand...

There are two parts to the forks support miner support and invester support. The pos fork will have suport from investers and due to the nature of it it will also have miner support as a result. I don't see how the old fork will stay sustainable with just the miners and can't see too much money going into it. With that said etc MAY stay profitable for these things
 
Stoly said:
I thought ETH was ASIC proof
Click to expand...

Nothing minable is asic proof. The goal is to make the algorithm so memory intensive that it's not profitable to buy enough high bandwidth mem to make the asic
 
Orddie said:
but it's on the Internet...

perhaps they have some insider info into how far out POS is for ETH
Click to expand...

If that was the case we would be looking at late 2018 early 2019 fork. If that is the case then if anyone believes this is credible they should buy as many gpus as possible as they would hit roi before release or fork
 
Guess they are planning a fork regardless of when POS is fully deployed because there is also difficulty bomb so they have to fork to diffuse the difficulty bomb otherwise this antminer is pointless and will never ROI.

haven't looked at the thread but I would guess one can estimate hashrate based on memory configuration.

edit: looked up difficulty bomb, still there but delayed for now. will hit hard in 2021 according to Vitalik. So I still think they plan on forking regardless before POS.
 
Last edited:
this could explain why the difficulty shot up a few weeks back. maybe some of these came online.
 
trick0502 said:
this could explain why the difficulty shot up a few weeks back. maybe some of these came online.
Click to expand...

nah, it didn't shoot up but was gradual increase that coincides with ETH going up in price from $300 range to $800 range in mid Dec. So just natural gpu hash moving back over to ETH from mining cryptonight or other algos because of ETH price increase.
 
Orddie said:
this is bull shit.

though whats the point if it's going POS
Click to expand...

We've been hearing "isn't it going POS soon" for eons now. There's something VB hinted at long ago: the threat of ETH going POS has really been about serving as an existential threat to keep ASICs away. He didn't want the Bitmain mafia and its miner's union moving into ETH like it did BTC/LTC and assuming a controlling position by default ("He who controls the hash.. controls the universe").

People tend to assume Bitmain's business model and long game is actually about selling computier hardware. No, their boxes are a means to an end. Jihan Wu is all about being first to controlling interest in an evolving paradigm shift in global currency while everyone is asleep or just barely beginning to figure it out.
 
Last edited:
odditory said:
We've been hearing "isn't it going POS soon" for eons now. I'll let you in on something that VB hinted at long ago. The threat of ETH going POS was really mainly about an existential threat to keep ASICs away. He didn't want the Bitmain mafia and its miner's union moving into ETH like it did BTC/LTC and assuming a controlling position by default ("He who controls the hash.. controls the universe").

People tend to assume Bitmain's business model and long game is actually about selling computing hardware. No, their boxes are a means to an end. Jihan Wu is all about being first to controlling interest in an evolving paradigm shift in global currency while everyone is asleep or just barely beginning to figure it out.
Click to expand...

in the case of ether with its very active development team wouldnt it pe possible to simply switch the algo behind ether rendering the asic useless and discouraging anyone from trying again?
 
cdabc123 said:
in the case of ether with its very active development team wouldnt it pe possible to simply switch the algo behind ether rendering the asic useless and discouraging anyone from trying again?
Click to expand...

It's not that simple unfortunately, the idea that you just fork whenever you don't like something that someone else is doing with your coin/protocol/platform. There's always a cost to forking. The Sia team faced this dilemma when Bitmain released their Sia miner - understandably they weren't happy since they had their own Obelisk ASIC miner in the works but not set to release for 4-5 more months. Their immediate reaction was "let's just fork" but they quickly realized they'd be shooting themselves in the foot and it actually made more sense not to, angry as they were.

Obviously each situation is different and I'm sure a fork is being discussed. This will be interesting to watch unfold - a clash of titans that was inevitable and I'm somewhat surprised it took this long.
 
Yeah, that's what I'd guess. Not to mention, isn't it memory bandwidth / latency that's paramount with ETH?

So are they making an ASIC with 32 DDR3 channels? Wonder if that makes up for it not being fast GDDR5?
 
odditory said:
People tend to assume Bitmain's business model and long game is actually about selling computier hardware $3000 at a time. No, their boxes are a means to an end. Jihan Wu is all about being first to controlling interest in an evolving paradigm shift in global currency while everyone is asleep or just barely beginning to figure it out.
Click to expand...

You give that little shit way to much credit. He is selling boxes to other people not in his control.

cdabc123 said:
in the case of ether with its very active development team wouldnt it pe possible to simply switch the algo behind ether rendering the asic useless and discouraging anyone from trying again?
Click to expand...

switching is forking and is already planned for example whenever switch comes to POS they are changing the algo behind it.
 
xrealm20 said:
Yeah, that's what I'd guess. Not to mention, isn't it memory bandwidth / latency that's paramount with ETH?

So are they making an ASIC with 32 DDR3 channels? Wonder if that makes up for it not being fast GDDR5?
Click to expand...

That's 32x 1 gigabit chips, so only one 32-bit "channel" per ASIC. A desktop uses 64-bit channels.
 
MrSneis said:
Does this mean the end for small time ETH miners?
Click to expand...
It would --- yes.

For example.

I bought one of the Bitmain A3. It is my first ASIC.
The A3 computes siacoins' algorithm at 815GH/s at stock speeds and cost $2,400.
A 1080TI computes siacoin's algorithm at 3GH/s at stock speeds and costs $800. ($700 MSRP - but can't buy them for that)

That's a 270:1 ratio to the fastest current gen CPU, at only a 3:1 price difference.

Most people seem to overclock their A3's and push the needle towards 900GH/s.

The A3 uses 1300 watts. If you undervolt GPUs you could run 6, 1080TI ~ 1300 watts. So even from a power perspective an ASIC blows GPU's out of the water. With 1300 watts total draw you can get - 815GH/s or 18GH/s with 6, 1080TI.

So yes -- everything you mined on a small operation for the last year, might be outdone in one day by a single ASIC. Difficulty will skyrocket, and a GPU will make basically nothing subsequently.
 
Last edited:
This may be good news, given how ridiculous it is to try to get GPUs right now. I have a feeling it's not going to be cheap though... but I could see big farms buying them and then offloading their cards. Good news for small miners and gamers. On the other hand, it may also increase difficulty much faster now.
 
so quick back of envelope calc based on 3 boards each with 6 asics with 32bit DDR3-2600 each.

64 bit DDR3-2600 (desktop) is 21 GB/s. So at 32-bit per asic is 10.5 GB/s. with 6 asics per board, x3 boards = 10.5 * 6 * 3 = 189 GB/s memory bandwidth.

so less bandwidth for the entire antminer than an rx 470. rx 470 is 211 GB/s stock.

cas of DDR3 is 10.3ns, cas of GDDR5 is 10.6ns so not much room for any "latency" improvement in algo.

edit: maybe the angle is power reduction as compared to gpu?
edit2: also this give each asic only 4GB so there will eventually be DAG issues fitting into memory.
edit3: think this is just FUD, this miner makes since if it was 2015, not 2018.
 
Last edited:
Archaea said:
It would --- yes.

For example.

I bought one of the Bitmain A3. It is my first ASIC.
The A3 computes siacoins' algorithm at 815GH/s at stock speeds and cost $2,400.
A 1080TI computes siacoin's algorithm at 3GH/s at stock speeds and costs $800. ($700 MSRP - but can't buy them for that)

That's a 270:1 ratio to the fastest current gen CPU, at only a 3:1 price difference.

Most people seem to overclock their A3's and push the needle towards 900GH/s.

The A3 uses 1300 watts. If you undervolt GPUs you could run 6, 1080TI ~ 1300 watts. So even from a power perspective an ASIC blows GPU's out of the water. With 1300 watts total draw you can get - 815GH/s or 18GH/s with 6, 1080TI.

So yes -- everything you mined on a small operation for the last year, might be outdone in one day by a single ASIC. Difficulty will skyrocket, and a GPU will make basically nothing subsequently.
Click to expand...

ethereum is much more mem intencive so it wont be a similar ratio. for example if performance hypothetically scaled directly with mem i cant see a asic offering too muhc performance increase.
 
After some more research into this, it looks like just more of the usual Fake China News™

Wouldn't go panic dumping those Radeons and Vega's just yet...
 
Last edited:
This is all FUD. There will be no ETH ASIC. If such a thing ever does hit the market, it will be nothing but a bunch of MacGyver'd together video cards that will not be any faster or more efficient than home built setups.

We've seen this before. It is called the AM200
 
doublejack said:
This is all FUD. There will be no ETH ASIC. If such a thing ever does hit the market, it will be nothing but a bunch of MacGyver'd together video cards that will not be any faster or more efficient than home built setups.

We've seen this before. It is called the AM200
Click to expand...
Correction

You haven’t seen it from bitmain.

Bitmain is the same company who snuck production units of the a3 through design, testing, and manufacturing stage without one peep of news getting out to the greater public. I recently watched a video of a 45 minute interview with the lead sia coin developer. He said straight up the sia team had NO idea bitmain was launching those miners till the day the rest of the public knew — which was basically when they came up for mass sale.

Bitmain is the 800 lb gorilla. They do what they please.
 
Archaea said:
Correction

You haven’t seen it from bitmain.

Bitmain is the same company who snuck production units of the a3 through design, testing, and manufacturing stage without one peep of news getting out to the greater public. I recently watched a video of a 45 minute interview with the lead sia coin developer. He said straight up the sia team had NO idea bitmain was launching those miners till the day the rest of the public knew — which was basically when they came up for mass sale.

Bitmain is the 800 lb gorilla. They do what they please.
Click to expand...

I get what you're saying. If Bitmain wants to keep something quiet, then nobody will hear about it until they are ready to sell it. And yes, that was not a Bitmain product.

The larger point, though, is that Ethash is by nature highly resistant to ASICs. I don't believe Bitmain is attempting to create an ASIC for ETH, but even if they are it is extremely unlikely they will succeed at creating anything more efficient and/or faster than already available GPUs. I don't fear Bitmain announcing some breakthrough Ethereum mining ASIC that will suddenly render GPUs obsolete. If they do have something in the works, it will just be more of the same that we saw with the AM200.

You can't really compare the situation with the A3. The Sia devs were indeed surprised, but how surprised were they really? They were working on their own ASIC, no? The only thing that happened there was a business with far more resources and more experience beat them to it.

Bitmain may do what they please, but they can't defy the laws of physics or mathematics.
 
I'll believe this when I actually see it. Anyone dumping AMD cards over this "news" is a moron.
 
doublejack

I don’t think anyone should hastily dismiss what bitmain can do with an asic of their own design —- and assume it can’t spank current GPUs.

Every asic they’ve released in the last couple years has outright spanked GPUs. What GPU now mines anything that bitmain asics mine? Nothing...

As I mentioned. One A3 is the equivalent to roughly two hundred and seventy 1080tis at mining Blake2b. No longer do GPUs mine Litecoin, dash, or bitcoin. Bitmain is getting bigger and stronger every quarter. Hiring more talent, and upping their capability.

ASICS are end game in crypto and GPUs will move on to the next CPU mined coin that gets popular enough to have some devs wrote code to mine it.


CPU -> GPU -> ASIC
 
Last edited:
Watch this video when you have the change. Or rather listen to it. It’s an interview with the lead dev from sia coins and the lead dev from vert coin. It’s the video i referenced above.


They talk another their views on crypto, mining, bitmain, asic, the future etc.
 
Archaea said:
doublejack

I don’t think anyone should hastily dismiss what bitmain can do with an asic of their own design —- and assume it can’t spank current GPUs.

Every asic they’ve released in the last couple years has outright spanked GPUs. What GPU now mines anything that bitmain asics mine? Nothing...

As I mentioned. One A3 is the equivalent to roughly two hundred and seventy 1080tis at mining Blake2b. No longer do GPUs mine Litecoin, dash, or bitcoin. Bitmain is getting bigger and stronger every quarter. Hiring more talent, and upping their capability.

ASICS are end game in crypto and GPUs will move on to the next CPU mined coin that gets popular enough to have some devs wrote code to mine it on an asic


CPU -> GPU -> ASIC
Click to expand...

its not a matter of talent or time its simpally not practical to do it. on high mem intensive coins often performance scaled with memory bandwidth assuming the core is 100% effeciant. its pretty easy to program a fpga or asic to handle all the instruction any of these cryptos need but when they need a massive memory bandwidth the asic providers now need to buy a significant amount of mem they cannot produce them selfs. Often they are completely unabalivle to get most mem in a reasonable time-frame (hbm) as the stuff they could get can often be cheaper when just bought on a gpu. Unless they manage to find a cheap sourse of VERY high bandwidth memory you wont see asics for some of thease coins. ethereum would also be one of the last coins to target for a asic as #1 it can pretty much fork whenever avoiding too much opposition #2 it is EXTREMLY mem intencive
 
The very start of the video (only did the first few minutes, get to rest later when I have time) points out that SIA algo was purposely chosen to be good for doing an asic with intent of doing so while ETH algo was purposely chosen to be asic resistant. So a bit apples to orange comparison just because Bitmain ate SIA's lunch before SIA could to say Bitmain can do something similar in performance with ETH algo.

ETH does seem to be dropping a bit more than the usual suspects so fake news may have had the intended effect for slow news week given this week ends in Chinese New Year.
 
Last edited:
You must log in or register to reply here.
Back
Top