Intel Corporation Patent for High-Performance Energy ...
Intel Corporation Patent for High-Performance Energy ...
Intel® SHA Extensions Intel® Developer Zone : Bitcoin
Intel® SHA Extensions
Intel Awarded a Patent for an Energy Efficient Bitcoin ...
Intel Files Patent for More Efficient Bitcoin (BTC) Mining ...
The Economics of "ASICs" for RandomX
Regarding RandomX and ASICs, people seem to mis-understand the interaction between hardware/software, the costs of development/manufacture of world class semiconductors, and the economics of ASIC mining. These on-chip, bare metal processing units that RandomX requires are highly advanced, and required to make chips "smart." An "ASIC" takes simple dumb functions and iterates them on bare silicon, rather than as a software abstraction (a layer higher) on top of the underlying hardware. There are a few things that RandomX didn't include, meaning that a fully optimized "ASIC" processor could eliminate some unused logic units to make room for the others particular to RandomX. At most, the designers of RandomX estimate a 2x improvement over current CPUs. 2x. That's almost nothing when you consider the economics of what it takes to make one of these things. As someone who spent 5 years as a process engineer in a world class semicon fab, I can tell you the design, testing, manufacture and turnaround time for advanced CPUs is enormous. Nothing like dumb SHA256 ASICs or simple memory chips. You better be gettting a huge return on investment to go to all that trouble. 2x is nothing. Bitcoin ASICs arose because there was something on the order of 1000x speed advantage over GPUs. That's the kind of economics you need to recoup the costs involved. Not to mention that you are still competing against the ever progressing AMD and Intel advancements (and probably ARM and RISC-V soon enough). Bitmain has very little competition. They have the patents and the process for new ASICs. So no, it's unlikely, for years, that anyone will develop an economically viable "ASIC" (lol) for RandomX. What we might see are CPU manufacturers use RandomX as a benchmark for their new units, since you will have the promise of a sort of "rebate" by mining Monero. In fact that could be an awesome marketing opportunity for Monero.
CrowdScrape 1.2 - Now available in Chrome Web Store
The CrowdStrike Intelligence Team is proud to announce the release of CrowdScrape version 1.2.0 available in the Chrome Web Store. CrowdScrape is a Chrome Plugin designed to allow users to scrape indicators from various websites and in-browser documents such as PDF reports while matching the data up against CrowdStrike Intelligence. This release adds integration with the CrowdStrike Indicator Graph. This tool produces indicator lists that collect:
Domain, IP addresses, URLs, hashes (MD5, SHA1, SHA256) and Bitcoin addresses
On-Screen Tagging of CrowdStrike known indicators, and links to Indicator Search
Matches to CrowdStrike Intelligence with links to reports in CrowdStrike Intel Portal
Integrates with the CrowdStrike Indicator Graph to visualize intelligence
In addition, you can use CrowdScrape to copy any indicators to clipboard making it easier to pull OSINT from different sources and converting this to a text file for implementation into your systems. Try it out with a trial of Falcon X and drop us some feedback in this thread!
Cross posted from Monero Regarding RandomX and ASICs, people seem to mis-understand the interaction between hardware/software, the costs of development/manufacture of world class semiconductors, and the economics of ASIC mining. These on-chip, bare metal processing units that RandomX requires are highly advanced, and required to make chips "smart." An "ASIC" takes simple dumb functions and iterates them on bare silicon, rather than as a software abstraction (a layer higher) on top of the underlying hardware. There are a few things that RandomX didn't include, meaning that a fully optimized "ASIC" processor could eliminate some unused logic units to make room for the others particular to RandomX. At most, the designers of RandomX estimate a 2x improvement over current CPUs. 2x. That's almost nothing when you consider the economics of what it takes to make one of these things. As someone who spent 5 years as a process engineer in a world class semicon fab, I can tell you the design, testing, manufacture and turnaround time for advanced CPUs is enormous. Nothing like dumb SHA256 ASICs or simple memory chips. You better be gettting a huge return on investment to go to all that trouble. 2x is nothing. Bitcoin ASICs arose because there was something on the order of 1000x speed advantage over GPUs. That's the kind of economics you need to recoup the costs involved. Not to mention that you are still competing against the ever progressing AMD and Intel advancements (and probably ARM and RISC-V soon enough). Bitmain has very little competition. They have the patents and the process for new ASICs. So no, it's unlikely, for years, that anyone will develop an economically viable "ASIC" (lol) for RandomX. What we might see are CPU manufacturers use RandomX as a benchmark for their new units, since you will have the promise of a sort of "rebate" by mining Monero. In fact that could be an awesome marketing opportunity for Monero.
Transcript of discussion between an ASIC designer and several proof-of-work designers from #monero-pow channel on Freenode this morning
[08:07:01] lukminer contains precompiled cn/r math sequences for some blocks: https://lukminer.org/2019/03/09/oh-kay-v4r-here-we-come/ [08:07:11] try that with RandomX :P [08:09:00] tevador: are you ready for some RandomX feedback? it looks like the CNv4 is slowly stabilizing, hashrate comes down... [08:09:07] how does it even make sense to precompile it? [08:09:14] mine 1% faster for 2 minutes? [08:09:35] naturally we think the entire asic-resistance strategy is doomed to fail :) but that's a high-level thing, who knows. people may think it's great. [08:09:49] about RandomX: looks like the cache size was chosen to make it GPU-hard [08:09:56] looking forward to more docs [08:11:38] after initial skimming, I would think it's possible to make a 10x asic for RandomX. But at least for us, we will only make an ASIC if there is not a total ASIC hostility there in the first place. That's better for the secret miners then. [08:13:12] What I propose is this: we are working on an Ethash ASIC right now, and once we have that working, we would invite tevador or whoever wants to come to HK/Shenzhen and we walk you guys through how we would make a RandomX ASIC. You can then process this input in any way you like. Something like that. [08:13:49] unless asics (or other accelerators) re-emerge on XMR faster than expected, it looks like there is a little bit of time before RandomX rollout [08:14:22] 10x in what measure? $/hash or watt/hash? [08:14:46] watt/hash [08:15:19] so you can make 10 times more efficient double precisio FPU? [08:16:02] like I said let's try to be productive. You are having me here, let's work together! [08:16:15] continue with RandomX, publish more docs. that's always helpful. [08:16:37] I'm trying to understand how it's possible at all. Why AMD/Intel are so inefficient at running FP calculations? [08:18:05] midipoet ([email protected]/web/irccloud.com/x-vszshqqxwybvtsjm) has joined #monero-pow [08:18:17] hardware development works the other way round. We start with 1) math then 2) optimization priority 3) hw/sw boundary 4) IP selection 5) physical implementation [08:22:32] This still doesn't explain at which point you get 10x [08:23:07] Weren't you the ones claiming "We can accelerate ProgPoW by a factor of 3x to 8x." ? I find it hard to believe too. [08:30:20] sure [08:30:26] so my idea: first we finish our current chip [08:30:35] from simulation to silicon :) [08:30:40] we love this stuff... we do it anyway [08:30:59] now we have a communication channel, and we don't call each other names immediately anymore: big progress! [08:31:06] you know, we russians have a saying "it was smooth on paper, but they forgot about ravines" [08:31:12] So I need a bit more details [08:31:16] ha ha. good! [08:31:31] that's why I want to avoid to just make claims [08:31:34] let's work [08:31:40] RandomX comes in Sep/Oct, right? [08:31:45] Maybe [08:32:20] We need to audit it first [08:32:31] ok [08:32:59] we don't make chips to prove sw devs that their assumptions about hardware are wrong. especially not if these guys then promptly hardfork and move to the next wrong assumption :) [08:33:10] from the outside, this only means that hw & sw are devaluing each other [08:33:24] neither of us should do this [08:33:47] we are making chips that can hopefully accelerate more crypto ops in the future [08:33:52] signing, verifying, proving, etc. [08:34:02] PoW is just a feature like others [08:34:18] sech1: is it easy for you to come to Hong Kong? (visa-wise) [08:34:20] or difficult? [08:34:33] or are you there sometimes? [08:34:41] It's kind of far away [08:35:13] we are looking forward to more RandomX docs. that's the first step. [08:35:31] I want to avoid that we have some meme "Linzhi says they can accelerate XYZ by factor x" .... "ha ha ha" [08:35:37] right? we don't want that :) [08:35:39] doc is almost finished [08:35:40] What docs do you need? It's described pretty good [08:35:41] so I better say nothing now [08:35:50] we focus on our Ethash chip [08:36:05] then based on that, we are happy to walk interested people through the design and what else it can do [08:36:22] that's a better approach from my view than making claims that are laughed away (rightfully so, because no silicon...) [08:36:37] ethash ASIC is basically a glorified memory controller [08:36:39] sech1: tevador said something more is coming (he just did it again) [08:37:03] yes, some parts of RandomX are not described well [08:37:10] like dataset access logic [08:37:37] RandomX looks like progpow for CPU [08:37:54] yes [08:38:03] it is designed to reflect CPU [08:38:34] so any ASIC for it = CPU in essence [08:39:04] of course there are still some things in regular CPU that can be thrown away for RandomX [08:40:20] uncore parts are not used, but those will use very little power [08:40:37] except for memory controller [08:41:09] I'm just surprised sometimes, ok? let me ask: have you designed or taped out an asic before? isn't it risky to make assumptions about things that are largely unknown? [08:41:23] I would worry [08:41:31] that I get something wrong... [08:41:44] but I also worry like crazy that CNv4 will blow up, where you guys seem to be relaxed [08:42:06] I didn't want to bring up anything RandomX because CNv4 is such a nailbiter... :) [08:42:15] how do you guys know you don't have asics in a week or two? [08:42:38] we don't have experience with ASIC design, but RandomX is simply designed to exactly fit CPU capabilities, which is the best you can do anyways [08:43:09] similar as ProgPoW did with GPUs [08:43:14] some people say they want to do asic-resistance only until the vast majority of coins has been issued [08:43:21] that's at least reasonable [08:43:43] yeah but progpow totally will not work as advertised :) [08:44:08] yeah, I've seen that comment about progpow a few times already [08:44:11] which is no surprise if you know it's just a random sales story to sell a few more GPUs [08:44:13] RandomX is not permanent, we are expecting to switch to ASIC friendly in a few years if possible [08:44:18] yes [08:44:21] that makes sense [08:44:40] linzhi-sonia: how so? will it break or will it be asic-able with decent performance gains? [08:44:41] are you happy with CNv4 so far? [08:45:10] ah, long story. progpow is a masterpiece of deception, let's not get into it here. [08:45:21] if you know chip marketing it makes more sense [08:45:24] linzhi-sonia: So far? lol! a bit early to tell, don't you think? [08:45:35] the diff is coming down [08:45:41] first few hours looked scary [08:45:43] I remain skeptical: I only see ASICs being reasonable if they are already as ubiquitous as smartphones [08:45:46] yes, so far so good [08:46:01] we kbew the diff would not come down ubtil affter block 75 [08:46:10] yes [08:46:22] but first few hours it looks like only 5% hashrate left [08:46:27] looked [08:46:29] now it's better [08:46:51] the next worry is: when will "unexplainable" hashrate come back? [08:47:00] you hope 2-3 months? more? [08:47:05] so give it another couple of days. will probably overshoot to the downside, and then rise a bit as miners get updated and return [08:47:22] 3 months minimum turnaround, yes [08:47:28] nah [08:47:36] don't underestimate asicmakers :) [08:47:54] you guys don't get #1 priority on chip fabs [08:47:56] 3 months = 90 days. do you know what is happening in those 90 days exactly? I'm pretty sure you don't. same thing as before. [08:48:13] we don't do any secret chips btw [08:48:21] 3 months assumes they had a complete design ready to go, and added the last minute change in 1 day [08:48:24] do you know who is behind the hashrate that is now bricked? [08:48:27] innosilicon? [08:48:34] hyc: no no, and no. :) [08:48:44] hyc: have you designed or taped out a chip before? [08:48:51] yes, many years ago [08:49:10] then you should know that 90 days is not a fixed number [08:49:35] sure, but like I said, other makers have greater demand [08:49:35] especially not if you can prepare, if you just have to modify something, or you have more programmability in the chip than some people assume [08:50:07] we are chipmakers, we would never dare to do what you guys are doing with CNv4 :) but maybe that just means you are cooler! [08:50:07] and yes, programmability makes some aspect of turnaround easier [08:50:10] all fine [08:50:10] I hope it works! [08:50:28] do you know who is behind the hashrate that is now bricked? [08:50:29] inno? [08:50:41] we suspect so, but have no evidence [08:50:44] maybe we can try to find them, but we cannot spend too much time on this [08:50:53] it's probably not so much of a secret [08:51:01] why should it be, right? [08:51:10] devs want this cat-and-mouse game? devs get it... [08:51:35] there was one leak saying it's innosilicon [08:51:36] so you think 3 months, ok [08:51:43] inno is cool [08:51:46] good team [08:51:49] IP design house [08:51:54] in Wuhan [08:52:06] they send their people to conferences with fake biz cards :) [08:52:19] pretending to be other companies? [08:52:26] sure [08:52:28] ha ha [08:52:39] so when we see them, we look at whatever card they carry and laugh :) [08:52:52] they are perfectly suited for secret mining games [08:52:59] they made at most $6 million in 2 months of mining, so I wonder if it was worth it [08:53:10] yeah. no way to know [08:53:15] but it's good that you calculate! [08:53:24] this is all about cost/benefit [08:53:25] then you also understand - imagine the value of XMR goes up 5x, 10x [08:53:34] that whole "asic resistance" thing will come down like a house of cards [08:53:41] I would imagine they sell immediately [08:53:53] the investor may fully understand the risk [08:53:57] the buyer [08:54:13] it's not healthy, but that's another discussion [08:54:23] so mid-June [08:54:27] let's see [08:54:49] I would be susprised if CNv4 ASICs show up at all [08:54:56] surprised* [08:54:56] why? [08:55:05] is only an economic question [08:55:12] yeah should be interesting. FPGAs will be near their limits as well [08:55:16] unless XMR goes up a lot [08:55:19] no, not *only*. it's also a technology question [08:55:44] you believe CNv4 is "asic resistant"? which feature? [08:55:53] it's not [08:55:59] cnv4 = Rabdomx ? [08:56:03] no [08:56:07] cnv4=cryptinight/r [08:56:11] ah [08:56:18] CNv4 is the one we have now, I think [08:56:21] since yesterday [08:56:30] it's plenty enough resistant for current XMR price [08:56:45] that may be, yes! [08:56:55] I look at daily payouts. XMR = ca. 100k USD / day [08:57:03] it can hold until October, but it's not asic resistant [08:57:23] well, last 24h only 22,442 USD :) [08:57:32] I think 80 h/s per watt ASICs are possible for CNv4 [08:57:38] linzhi-sonia where do you produce your chips? TSMC? [08:57:44] I'm cruious how you would expect to build a randomX ASIC that outperforms ARM cores for efficiency, or Intel cores for raw speed [08:57:48] curious [08:58:01] yes, tsmc [08:58:21] Our team did the world's first bitcoin asic, Avalon [08:58:25] and upcoming 2nd gen Ryzens (64-core EPYC) will be a blast at RandomX [08:58:28] designed and manufactured [08:58:53] still being marketed? [08:59:03] linzhi-sonia: do you understand what xmr wants to achieve, community-wise? [08:59:14] Avalon? as part of Canaan Creative, yes I think so. [08:59:25] there's not much interesting oing on in SHA256 [08:59:29] Inge-: I would think so, but please speak [08:59:32] hyc: yes [09:00:28] linzhi-sonia: i am curious to hear your thoughts. I am fairly new to this space myself... [09:00:51] oh [09:00:56] we are grandpas, and grandmas [09:01:36] yet I have no problem understanding why ASICS are currently reviled. [09:01:48] xmr's main differentiators to, let's say btc, are anonymity and fungibility [09:01:58] I find the client terribly slow btw [09:02:21] and I think the asic-forking since last may is wrong, doesn't create value and doesn't help with the project objectives [09:02:25] which "the client" ? [09:02:52] Monero GUI client maybe [09:03:12] MacOS, yes [09:03:28] What exactly is slow? [09:03:30] linzhi-sonia: I run my own node, and use the CLI and Monerujo. Have not had issues. [09:03:49] staying in sync [09:03:49] linzhi-sonia: decentralization is also a key principle [09:03:56] one that Bitcoin has failed to maintain [09:04:39] hmm [09:05:00] looks fairly decentralized to me. decentralization is the result of 3 goals imo: resilient, trustless, permissionless [09:05:28] don't ask a hardware maker about physical decentralization. that's too ideological. we focus on logical decentralization. [09:06:11] physical decentralization is important. with bulk of bitnoin mining centered on Chinese hydroelectric dams [09:06:19] have you thought about including block data in the PoW? [09:06:41] yes, of course. [09:07:39] is that already in an algo? [09:08:10] hyc: about "centered on chinese hydro" - what is your source? the best paper I know is this: https://coinshares.co.uk/wp-content/uploads/2018/11/Mining-Whitepaper-Final.pdf [09:09:01] linzhi-sonia: do you mine on your ASICs before you sell them? [09:09:13] besides testing of course [09:09:45] that paper puts Chinese btc miners at 60% max [09:10:05] tevador: I think everybody learned that that is not healthy long-term! [09:10:16] because it gives the chipmaker a cost advantage over its own customers [09:10:33] and cost advantage leads to centralization (physical and logical) [09:10:51] you guys should know who finances progpow and why :) [09:11:05] but let's not get into this, ha ha. want to keep the channel civilized. right OhGodAGirl ? :) [09:11:34] tevador: so the answer is no! 100% and definitely no [09:11:54] that "self-mining" disease was one of the problems we have now with asics, and their bad reputation (rightfully so) [09:13:08] I plan to write a nice short 2-page paper or so on our chip design process. maybe it's interesting to some people here. [09:13:15] basically the 5 steps I mentioned before, from math to physical [09:13:32] linzhi-sonia: the paper you linked puts 48% of bitcoin mining in Sichuan. the total in China is much more than 60% [09:13:38] need to run it by a few people to fix bugs, will post it here when published [09:14:06] hyc: ok! I am just sharing the "best" document I know today. it definitely may be wrong and there may be a better one now. [09:14:18] hyc: if you see some reports, please share [09:14:51] hey I am really curious about this: where is a PoW algo that puts block data into the PoW? [09:15:02] the previous paper I read is from here http://hackingdistributed.com/2018/01/15/decentralization-bitcoin-ethereum/ [09:15:38] hyc: you said that already exists? (block data in PoW) [09:15:45] it would make verification harder [09:15:49] linzhi-sonia: https://the-eye.eu/public/Books/campdivision.com/PDF/Computers%20General/Privacy/bitcoin/meh/hashimoto.pdf [09:15:51] but for chips it would be interesting [09:15:52] we discussed the possibility about a year ago https://www.reddit.com/Monero/comments/8bshrx/what_we_need_to_know_about_proof_of_work_pow/ [09:16:05] oh good links! thanks! need to read... [09:16:06] I think that paper by dryja was original [09:17:53] since we have a nice flow - second question I'm very curious about: has anyone thought about in-protocol rewards for other functions? [09:18:55] we've discussed micropayments for wallets to use remote nodes [09:18:55] you know there is a lot of work in other coins about STARK provers, zero-knowledge, etc. many of those things very compute intense, or need to be outsourced to a service (zether). For chipmakers, in-protocol rewards create an economic incentive to accelerate those things. [09:19:50] whenever there is an in-protocol reward, you may get the power of ASICs doing something you actually want to happen [09:19:52] it would be nice if there was some economic reward for running a fullnode, but no one has come up with much more than that afaik [09:19:54] instead of fighting them off [09:20:29] you need to use asics, not fight them. that's an obvious thing to say for an asicmaker... [09:20:41] in-protocol rewards can be very powerful [09:20:50] like I said before - unless the ASICs are so useful they're embedded in every smartphone, I dont see them being a positive for decentralization [09:21:17] if they're a separate product, the average consumer is not going to buy them [09:21:20] now I was talking about speedup of verifying, signing, proving, etc. [09:21:23] they won't even know what they are [09:22:07] if anybody wants to talk about or design in-protocol rewards, please come talk to us [09:22:08] the average consumer also doesn't use general purpose hardware to secure blockchains either [09:22:14] not just for PoW, in fact *NOT* for PoW [09:22:32] it requires sw/hw co-design [09:23:10] we are in long-term discussions/collaboration over this with Ethereum, Bitcoin Cash. just talk right now. [09:23:16] this was recently published though suggesting more uptake though I guess https://btcmanager.com/college-students-are-the-second-biggest-miners-of-cryptocurrency/ [09:23:29] I find it pretty hard to believe their numbers [09:24:03] well [09:24:09] sorry, original article: https://www.pcmag.com/news/366952/college-kids-are-using-campus-electricity-to-mine-crypto [09:24:11] just talk, no? rumors [09:24:18] college students are already more educated than the average consumer [09:24:29] we are not seeing many such customers anymore [09:24:30] it's data from cisco monitoring network traffic