Topic: Ultra-Low-Voltage Energy-Efficient Bitcoin Mining ASIC from Intel (23 Feb 2022) (Read 1281 times)

Yes more than one time with intel based gear I have had to have extra gear to update gear they sent with old firmware.

They had an issue with eth ports that was really hard to fix.

I have had issues where I had to use old keyboard with old wire designs . ie ps/2 port to alter firmware to allow for usb key boards.

Alas from my (limited) experience, Intel is no better at this.
I've bought the parts and built myself, 2 servers - all up about $40k - (all intel: MB, CPU, Case, PSU but other party RAM) and had some interesting problems with them.
Specifically, the Intel motherboards:

First one the base firmware didn't spin the fans - at all - lots of overheating and hard shutdowns.
After discussing with them online for a while, it was indeed the MB needed a software update to fix it.
So while you could say it's my fault for not having done that at the start, I will point out that they had known about it for over a year, but they still didn't update the MB at the factory to fix this rather serious problem.

Second one couple years later, I bought the R version of the same MB due to getting a Xeon Gold 6258R (equivalent fastest 2nd gen CPU they sell)
Oddly enough though, it did not work at all.
It gave some weird light code that no one on the Intel board could identify (and not in the PDF manual)
Turns out that the firmware in the R version of the MB (that exists purely to support the 2nd gen R CPUs) didn't support the top of the line 6258R.
So since I fortunately had 2 Xeon Gold 6148 in the other MB, I took one out, put it in the new MB, booted fine, updated the MB, and now it works with the 6258R.
Had I not had a non R CPU on hand to update the MB, I would have been screwed and had to send it back to Intel to update it as it should have been to start with.

So, yeah, don't expect much from Intel either.
They do quite literally expect people to update faulty firmware themselves, and not ship it updated even if they know its a major problem.

Not for a while. From what was discussed elsewhere, they are going to the big buyers 1st then to the smaller bulk then to distributors.
Till they trickle down to us will probably be months.
It's probably a smart move on their part. 1000s and 1000s of end users can be a nightmare to support, putting them in the hands of professionals 1st to find issues that can be dealt with in a data center environment with full time support engineers is probably better.

Unlike other manufacturers that let us do the beta testing of the products.

-Dave

I've not seen any exact release dates and possible distributor partners...

Do we know when those Intel miners gonna be available for sale?

Just saw that. A bit less efficient then the S19j XP BUT if they are true to the marketing and the chips and miner are 1/2 the price then who cares.
And it's Intel. Might not be the best at all times, but their chips and miners are probably a lot better then everyone else.
Lets see how long till they come to market and we can actually buy them.

-Dave

Some more details on the chip and it seems the Bonanza2 is behind S19 XP. Good news for existing miners!

https://www.tomshardware.com/news/intels-second-gen-bitcoin-miners-performance-and-pricing-listed

Toms Hardware has a write up:
https://www.tomshardware.com/news/intel-details-its-bitcoin-mining-bonanza-mine-chips-and-systems

The 1st gen are essentially useless unless they are so dirt cheap as not to matter. If you can get the miner cheap enough then it might be worth it.

The 2nd gen looks like it might be a competitive product since there are big names buying it, but until more info is released it's all a guess.

-Dave

Being sha-256 chips does not lock them into being for BTC. Most of the logistics tracking blockchain applications I've seen from IBM are all using sha-256.

That said they can certainly be used for mining BTC but until specs are released we have no way of knowing how they compare to existing chips from Bitmain, Canaan, et al. So far all Intel is saying is that "they are 1000x more efficient than GPU's"

Umm.. but as far as I've read,
BMZ1 and BMZ2 are confirmed as btc (sha-256) asics.

I heard that this info leaked a bit earlier than Intel intended,
and that they would have official launch this month (which means next week ?).

Usual order quantities are said to be in thousands (I don't know exact numbers, e.g. for MOQ ).

Folks here should realize that Intel is talking about Blockchain Technology in it's myriad of current and possible applications - NOT just Bitcoin.
From the news release:GPU arrays are of course not used to mine BTC but they are used in blockchain-based logistics tracking applications, large scale Public Utility assets tracking, etc. that use blockchain tech for the records keeping.

Well, like you said previously, there isn't a full product out, or even a commercially available chip with specs published, so it's all just speculation at this point.

I don't doubt that below 20w/th is possible with the S19pro, but not every S19pro. Could be only one out of 10 can achieve that. Same for S17pros hitting 30w/th.

---

A couple of interesting tidbits of info in there. They've created a whole new group for doing projects like this.

With custom firmware the S19 pro can easily do below 20w/th according to one of the devs at vnish, so efficiency wise i highly doubt intel chips will beat that. So the cost of final products will be the judge.

As far as quality is concerned, I have to agree with you, although most issues are beyond the chip level, so we do not really need new chips to lower that failure rate, we just need better "everything else".

Microbt seems to be the only manufaturer with proper quality control, even the new Avalons have high failure rate, it is annoying given how expensive these gears sell for.

I'm expecting and hoping their end product is the chip, not a full miner. Let multiple integrators compete in the marketplace to build the best miner. But I can't really imagine they'd do a worse job than current miners out there if they did decide to market the end product. The failure rate on new gen miners although better than the 17 series still seems pretty awful. And the firmware sucks. Really, I just don't see how Intel could possibly do worse.

Bitmain has a working product that intel does not have yet, personally, I can't take the above table any seriously until there is a finished product all of this is unreal, the chip level is the most important but there are many other factors, the production capacity and final cost are two important things that we can't ignore.

Well, Bitmain's 1397 chip was their 2nd crack at 7nm and is about 25mm² and best efficiency of 30w/th when running ~280GH/sec. So hashrate per die area is pretty close between the two.

We also don't know what the high end of the hashrate might be on Intel's chip, what is quoted is most likely just the most efficient operating conditions, so it can likely run much faster at lower efficiency.

So I'd say Intel put Bitmain to shame, destroying their efficiency # at the same node size right out of the gate.

Here's a bit more information about this chip:


Here's a comparison table with some of the other miners, showing Intel's BZM1 as the most efficient chip:



There are also indications that there will be a second chip from Intel, BZM2, which will be released at some point in the future.

Admist all this talk, has anyone actually looked at the size of the BZ1 chip?

14.1 mm<sup>2 and only 137 GH? At 4000 per wafer each chip would cost a customer around $3 - it's got to be the worst mining deal for years,  $23 per TH versus a standard cell implementation of around $5 ?

I'm really disappointed in Intel, perhaps the second iteration will do much better.</sup>

Well for chip making, shift from Asia and specifically, Taiwan (not China). AFAIK Russia has next to no cutting, much less bleeding-edge fabs that use EUV litho processes. China definitely does not 'cause the West will not sell them the EUV tech and for at least the foreseeable future they cannot just copy it because they have no access to or ability to copy the underlying tech involved. PCB mfg and assembly of miners is a different matter.

That said, it's about freakin' time!

maybe the shift of mining away from China and Russia is coming.

Could be interesting for us all.

Wonder if Intel is planing doing their mining chips at the new Fabs to be built in Ohio? https://www.cnet.com/tech/computing/intels-100b-ohio-megafab-could-become-worlds-largest-chip-plant/

It makes sense for mega farms to once again be able to secure mining chips and just have their own custom miners built from them. Virtually every major data center does that with custom server mobo's and storage solutions tailored to their specific needs that they designed together with the actual hardware vendors that build them for the likes of Facebook, Amazon, Google, Hulu, Netflix etc. For new build outs I foresee most being liquid cooled setups rather like BM's ANTRACKS. Gives the maximum power density per cu/ft with none of the problems of scaling up immersion tank setups. Air cooling should be right-out.

Whenever I see farms like this I always have to ask  -- why in the hell...? Noise be damned, maint of the fans and PSU's has to employ over a dozen techs and other support people just to keep up with expected failure rates...
And by today's standards that farm in the above pic was a fairly small operation...

eg...
 

Way back around 2013 there was a 'monster' farm in Oregon that rolled their own miners using BitFury chip and their own hashboard designs. Somewhere I have an article saying their were pulling over $250k a day in BTC rewards...

So, they have a client:


and that client got  a month later a $525M Credit Facility from Blockchain.com

From a miner's point of view, oh sh***!
I'm scared to start updating a table on how much gear and how much money these companies are putting into gear, guess pretty soon you either mine for fun and to support the network or you don't mine anymore, profits will be counted in cents.

Good read.


and an interesting conflict in the story

in that 55 watts a th vs  18 watts a th are both listed As the power  used.

a board with 25 chips burning 2.5 watts = 62.5 watts does 3.425 th if 25 x .137 is correct.

62.5/3.425 = 18.24 watts a th.

now we all know  fans and a controller and other shit = some power.

But I can say 3 boards = 187.5 watts

maybe a six board rig pulling 375 watts + other parts say 75 watts = a 450 watt miner that does 21 th is possible

that makes it 21 or 22 watts a th which is better than all gear other than the s19xp

And if set to be a 450 watt unit it would  be a nice item.

Well if 500 watts and 20th are possible at a good price the small guy will buy it.

Update on the expected performance specs https://www.anandtech.com/show/17218/intels-next-gen-bitcoin-asic-called-bzm2-built-on-7nm-137-gigahashsec-at-25-w

😄 As I said earlier, that is pretty much what I expected to hear for them. 7nm lets them get REAL experience using their EUV litho processes and in theory by not shooting for 5nm makes it easier to hit those performance targets.
More to the point, of course since they already have their own Foundries using EUV there are no other customers (or at least very few) vying for a place in the production lines.

AFAIK that's what they're going to sell. That's their main business anyway, selling chips. Sometimes they do sell end products like the NUC, but that's the minority.

It would be really interesting to have a small unit at home that doesn't produce a lot of noise and heat and yet still gets you a bit of bitcoin every day through a pool.

From a working demo to actually producing this is a long way with the mandatory hop of it actually being that efficient compared to the current gear. But, what I would love to see some small miners, not USB sticks but some really low-power miners on pair with what futurebit offers, it would always be better to have 1 million home miners doing 2-5TH/s than a single entity controlling 5Exa, plus with really small consumption one wouldn't really care about electricity prices, 100-200 watts is nothing.


@philipma1957 rather than mobo chips I would rather see standalone cards, but ... we're a bit daydreaming right now.

I wonder if they would ever make these commercially given the bad press crypto mining has been getting recently.

Intel are a global household name, and getting involved in the "dirty eco-killing" mining business might a PR step in the wrong direction.

Well it would be nice to give a home owner a quality miner with good efficiency .

Under 800 watts full power
and compete with 30 watts a th.

Just think:

 25th unit that uses 750 watts = Good
 25th unit that uses 625 watts = Better
 25th unit that uses 500 watts = Best

I think it could sell well. It can't be a lot more efficient that the beast miners.

I picked that th and those watts for specific reasons.

If they really were smart they could sell a mobo using up to 2 mining chips

and let the mobo be able to be upgrade to better mining chips. I e much like an intel server mobo

Well it looks like I am going to eat my words from what I have been saying around here for a while.
I REALLY thought that Intel would never enter such a niche market. They have for years only been about CPUs, memory, things that can be sold to everyone everywhere.
Yes, they always had some smaller product lines, but that was usually obtained from when they bought other businesses. Something like this developed in house has been rare from them as of late.

As others have said, lets see how it pans out in terms of efficiency & cost & availability.

-Dave

Let's just hope they decide to sell the chips like they do with their CPUs and let multiple 3rd parties develop the end products. That way the home miner might be able to get a quiet and efficient 1KW miner, and the industrial guys can get their 10kw 415V 3 phase monster... or whatever is coming next...

As I posted on Discord Considering that Intel is one of the handful of Foundries that have the EUV litho equipment required to produce chips with 7nm and smaller gate sizes it makes sense for them to put it to good use by practicing on mining chips.

It will be interesting to see if it is able to perform J/Hash like current miners,
or if it is just a 'working demo' with poor performance.
Intel's CPUs still aren't even under 10nm yet ...

Interesting! It would be a game changer for sure, if Intel decides to hop in to the btc asic game.

Intel will be showing a working demo of their paper: Bonanza Mine: An Ultra-Low-Voltage Energy-Efficient Bitcoin Mining ASIC at the ISSCC Conference on Wednesday February 23rd, 7:00 AM PST.

This is the International Solid-State Circuits Conference, where most of the best new chips are presented. This conference will be held in San Francisco and online because of the spread of omicron.

This is the program of the conference in pdf. You can find the Intel demo in page 27:



This is probably going to be related to a previous Intel patent from 2018:

Optimized sha-256 datapath for energy-efficient high-performance bitcoin mining