All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
CPU Crypto Mining and power efficiency.
randvegeta
Member, Host Rep
The amount of interest I've seen from people looking to mine things like Monero has gotten me interested.
Now obviously I cannot see any way to make money renting a server for the purpose of mining. If mining was more profitable than the cost of the server rental, it would make more sense to just mine directly. Which got me thinking....
As a provider with our own hardware, we literally have dozens of machines sitting in the DC unused. Some machines are completely useless for mining (1st, 2nd and 3rd gen i3 for example don't have AES-IN), but surprisingly many CPUs are actually viable for mining, and I've recruited about 20 odd machines to start mining. No matter what, the amount of money that can be earned will be tiny, but it is still kind of interesting.
I'm wondering if anyone here has done any tests on making CPUs more efficient for mining. Specifically, has anyone tried to underclock a rig to see if the performance per-watt is improved? This may sound counter-intuitive as the hash-rate WILL fall if you lower the clock speed. BUT, if you reduce the clock speed, you lower power consumption. Has anyone tried and measured this? For many CPUs, reducing the clock speed by 25% can reduce power consumption by 50%. If a 25% reduction in clock speed decreases hash rate proportionally, then that in theory increases the power effectiveness by 50%. As in, there is a 50% greater yield for the same amount of power. Now assuming the HDD/RAM/Motherboard and other stuff doesnt use much power at all, this could potentially make any AES-IN enabled CPU a viable miner, which might be nice for us providers with spare gear laying around. Anyone tested this?
Comments
Cache (and cache speed) matters for most coins, not AES-NI/AVX/AVX2/AVX512. For now.
This means, aside of new CPUs, Opterons and first/second Gen E5 (and especially E7, 38-60MB per CPU, 8S scalable) deliver the highest performance - obviously running Opterons is not cost effective anywhere really and E7 is expensive.
NOTABLY if someone releases a coin based on AVX2 or AVX512 you WILL see INSANE power draws - AVX2/512 can double the TDP in pretty much any part that has it and the Xeons with 2 instruction pipelines can even go higher.
Not necessarily, this depends on exact algo used/coin. Same as both exists on GPU mining and why some AMD cards with HBM (Fury, Vega, Pro Duo) yield MUCH higher ETH hashrates than Nvidia (even on GDDR5X).
You cannot do this on most modern CPUs, you can essentially only disable the HT cores or any other than the first physical core - multiplier settings are locked on server/consumer SKUs and FSB overclocking is dead (in fact, trying this will crash on a 8xxx CPU after barely 50Mhz gain, and in reverse you have problems as this clock signal on most boards is also used for PCIe and various other things).
In Highend and Workstation CPUs, yea, sure - but a binned part like an i3-4xxx (which is a binned E3-12x5 with 2 disabled cores) that lost the silicon lottery already is not really going to run more efficient by reducing to even less than fused.
In most generations you can estimate the savings somewhat by Vcore voltage on full load; if high you might be able to reduce total consumption, if low your CPU did get binned by eg. needs for more i3s and not by defect or has other defects than the cores (or just very minor).
Every post by @William is a different college-level seminar. Thanks man!
@william seems to have encyclopedic general knowledge... scary actually. That's just not natural.
That's interesting. But the i3s have pretty poor yield in absolute terms any way. i7s get more than double the hash rate with the same clock speed. Anything with less than 4 cores seems not worth while.
I'm not sure what exactly I should be taking away from your wall of text... Should I try and underclock or not? Most of my current Xeon E3, E5 and i7s all can have the CPU clocked down. My i7s outperform my E3s and my older i7s and E3s all perform better than the new ones (strangely). So it makes almost no sense to mine with new CPUs considering they are easier to rent out as servers. It's mostly older gen gear I have laying about.
tl;dr on his previous post would be nice lol
IF your board has options for it, sure go ahead - can't hurt to try. Have BIOS reset jumper known before, some boards flat out refuse to boot on too high (and possibly low) clock changes.
Also try to disable HT and lastly all cores and HT and re-measure.
You sort of play silicon lottery 2.0 here - Consumer CPUs are primarily binned from server/enterprise for either over or underperforming (because, in the base, an E3-1230 is absolutely the same as a 1240, just that the 1240 within it's TDP performs higher so it is a 1270 now, and not a 1230 or 1250) so just like with overclocking hit or miss.
You also, if you use FSB (old) or non-multiplier clock (FSB 2.0 in practice), want to either see if the board has a second clock generator for PCIe or not use the PCIe slots.
See:
https://www.tweaktown.com/guides/8342/amd-x399-tr4-threadripper-motherboard-buyers-guide/index5.html (ctrl+f "clock", Threadripper article but principle the same on everything).
Well, this makes sense - i3 (until recently) is dual core, i7 is quad, i7 is the same base as the E3 comes from which forms the i3 on lower yield - i7 forms i5 on lower yield.
Essentially you have this (applies, mostly, 1:1 until Skylake/v5 and Kaby Lake/v6):
E5 and E7 (usually -EP for enterprise) is separated by features and other DIE ultimately than any of the above. Sometimes you see the tech used there appear in the next generation of consumer CPUs then (like PCIe 3.0). Larger socket/more pins used due to QPI (now UPI) for multi socket.
HE desktop CPUs, eg. the 2011 i7, are castrated Xeons on lower end and custom DIEs on higher end - you can watch some delidding and especially die-sanding videos on YT which shows the CPU PCB layers, some have ECC controllers (that obviously do not work) and some do not (== was not a Xeon).
Celeron parts are produced as stock primarily, i have not yet seen eg. a G4400 (a dual core pentium part) that would be fused out of an i5.
Workstation parts - last was the W36xx series as E5-16xx is a normal E5 with some minor changes - are Xeons with native ECC and broken QPI/UPI that prevents multi socket usage or high performing on 2/4 cores so end up binned in the (back then) very expensive W3680/W3690 (X5680 and X5690 respectively for 2S).
This is hardly new, AMD does the same and GPUs as well - this is why the so called ASIC quality can massively range on the same GPU range, it's also not really a secret that a 1070 is just a castrated 1080 and so on.
Why binning happens has pretty much 3 reasons:
Give me some CPU models (and RAM speed you use) and i might be able to tell you why, but a LOT changed to v5/v6 so this only works on older easy.
Underclocking might work but cause issues, Cache matters most followed by AES-NI, AVX costs insane power/thermal
They would be way better if CF would not complain every time and block the entire page so i have to copy around tabs with inspect source. Utter junk.
https://prnt.li/f/76b87a47eed89eb9b74da986d5f8a8f2-legheepea2.png
Funny enough, by now i am mostly limited by expressional english skills...
Generally I remember this things (they feed from RSS and work, Youtube and so on in background) however cannot really recall them in a useful way but just as block, then filter out what i need from it which takes some thinking time.
The price for this is practically non existing short term memory and the usual ADHD/ADD problems of not being able to actually use this in a written test or similar for various reasons.
LET's horrible content blocker is the single most irritating thing on the Internet.
"Here is an detailed discussion of the technical problem you're having with examples and code and...oh frack, captcha..."
For a while I considered putting all my posts on pastebin and just replying here with links to the unmolested content...
That'd be one way to keep the CF LuftwaffeSSNazis from keeping you from typing etchosts.
Concerning Monero, i've found this interesting thread:
https://forums.servethehome.com/index.php?threads/monero-mining-performance.12116/
Some post give the hashrate / watt - some CPU seems really decent, better than most GPUs. Depends of what is the other hardware connected of course. PoW mining is an awful waste of energy if you want my opinion, though. I hope this crap won't last for too long. But it doesn't seem like the hype is going down. Will there be a real, mainstream use in the future? Hard to tell, seems quite inefficient for what it does (if you compare power consumption of btc transaction vs visa for example).
Monero mining is not efficient on GPUs, that is why. This was designed intentionally.
Generating basically cash is a mainstream use...
Depends on the relevant algorithm. Usually it's mainly some type of hash. vector support (avx etc.) depends. They cost power but can (much depending on algo and implementation) very considerably speed up things.
One can pretty mich differentiate between 3 main groups (lowest first):
With a normal processor, pretty much no matter which, you're limited to the hobby corner. Generally speaking, unless electricity happens to be free you are highly likely losing money.
With GPUs you are in the amateur league. If you play it big (100s or more) you are a large one in the class of mid size fish.
To really earn money you need ASIC based systems and many of them.
Side note: Knowing about programming and crypto (profoundly well) quite probably helps. Reason: Most mining software (and other crypto) currency) software is of mediocre quality.
As you said this depends on the coin, the ones designed explicitly to not support any ASIC or GPU mining (yea, Monero) do not really have advantages on GPUs - they are often worse considering an i7 maxes at 95W TDP while a RX480 will use 150W+.
Your list also lacks FPGA; they beat GPU/CPU but are a stage below ASIC.
@William
Yes, I left out fpga to keep it simple; fpga can be considered as a weaker small brother of asic.
As for monero, i.e. cryptonight, careful what you assert. Reason: The one thing that denies the gpu advantage is a hash ratchet/kdf (~"scrypt") with 2 MB state space (which a GPU doesn't have for each subunit (of which it typically has many)).
From a cryptographical view that's a classical kdf and unfortunately one with a rather low memory cost. I personally (like many others in the field) prefer a min. of 64 MB and up to 512 MB (depending on the hardware boundaries of a target market, the low end of which nowadays typically is smartphones with gigabytes of memory).
2 MB is absolutely not out of reach and a quite poor choice. Looking closer, cryptonight very strongly favours the biggest of players who can afford to get/have asics and fast (but very expensive) sram. As the algorithms used require but a very simple set of operations (which btw is positive) one could, provided a player has the big money, create chips with, say 64 of those simple cores, each with a relatively small register set plus 2 512 bit registers and some standard crypto hw (e.g. montgomery) and a very fast scenario optimized memory bus to 128 MB sram.
Yes, that would cost some millions, but: it would pay for itself through the computational advantage plus: the secret service would happily fund that, too, as it would happen to give them a major advantage, too.
As for the power consumption, yes , gpus are hogs but you must look at relevant performance per watt. After all a gpu has way more "cores" than an amd64 cpu.
Well, not always but rather currently due to design and needs - See Xeon Phi, it can be done.
What you imply seems to be rather Perf/Core than Perf/W, because if a CPU produces more or the same amount of money on less power usage (which is the case on a few) the Perf/W figure cannot be better for the GPU, by pure laws of physics.
And yes, sure, ASICs cannot be entirely prevented but the cost for them makes it not realistic to be funded at this time - same as preventing GPU mining can, and has been, affected by HBM/HBM2 being introduced rather unexpectedly, same as the Vega increases on ETH.
More. BTC ASIC development is estimated at medium 2 digit millions, and that was comparatively simple.
Yeaaaa: if you have access to this kind of hardware on larger scale without raising questions or political issues - obtaining millions of HW crypto modules from a presumably US based company as Chinese could, same as with CPUs, end not so well.
Same as quantum computers this limits "able" players even more than just money does, which yea on a gov scale is not too relevant - a single large building costs more after all.
@William
2 digit millions, OK, but you earn that back plus gain a lot. Most importantly though, the "equality brake" is poorly parameterized, so the "invitation" is out. 2 MB state space keeps todays GPUs at bay but not much more.And btw. with more than 1 processor design open sourced it doesn't take that much to build at least considerably optimized processor asics.
@William,
Since I'm not in HK to actually monitor the actual power usage, but I do have some interesting sensor data.
So I've got a 3 Xeon E3s mining (probably not very efficiently.. but just testing here) monero.
Server A: Xeon E3-1245v2 (Running at 3.5Ghz) = 263H/s
Server B: Xeon E3-1245v2 (Running at 2.6Ghz) = 195H/s
Server C: Xeon E3-1270v3 (Running at 3.6Ghz) = 264H/s
A reduction in the clock speed by exactly 25% has resulted in a hashrate exactly 25% less. Server A's CPU is running at about 58 degrees while Server B's CPU is running at just 46 degrees.
Interestingly, Server C, even though it is newer and faster, seems to be almost exactly the same performance as the v2 model, but is likley using more power. CPU running much hotter at 64 degrees!
I have another server, also running a Xeon E3-1270v3, but that is mostly sitting idle. Sensors indicate CPU temperature at just 32 degrees.
What this means for actual power consumption is not 100% clear, but it seems that Server C may be burning more power for the same hash rate at server A, and server B may be offering the best performance per watt of energy.
If using linux, make sure you increase largepages. XMR-stak-cpu offers this as a tweak to increase performance
@William an I7 won't have enough cache. A good E5 would be better than a RX (h/w) but way more costly (if you buy your RX at the right price, ie not the insane price they have been selling on ebays (or here ;-) not so long ago...)
Not really.. all of that can be googled in 15 minutes and it just shows that he either spends a lot of time creating the reply or happens to fiddle with these subjects (or around them) daily and therefore having no life.
Either way, it's actually useful for this forum
His replies are often well written and quite deep and he is quite knowledgeable, he's an appreciable contributor to let, indeed!
For a start: How about knowing many things in his (and related) fields is his (professional) life? What's bad about that?
Moreover, frankly, your post outed yourself as, uhm, not the brightest bulb. Reason: One can read all day long at wikipedia or elsewhere and it's not worth a lot without a solid foundation of knowledge. @William obviously has that foundation and if he sometimes quickly looks up a detail that's simply normal, but it works only because the foundation is there.
Similar thing with @Maounique. One might like or not like him but it's obvious that that man has touched a fucking lot of actual data center hardware with his (I guess) 5 tentacles. And it's that kind of practical had-his-hands-on knowledge that's the most valuable one.
Ahh, a debate with the pro crimea annexion guy. Theres no better way to spend a sunday night than to talk with a putin troll
First off, IT field is rather small when compared to medicine for example. The chances are that, when you know about CPUs, CPU instructions, GPUs, and cryptomining in general, one can easily construct a 5 paragraph presentation for the previously discussed topic of mining efficiency. You don't need to be smart to explain about server hardware either if you know desktop hardware and can build a basic PC. All you need is additional info to build around your core message, and that can be easily googled or spoken from your previous experience from related fields.
Im not saying its bad, its the opposite. He does great contribution, unlike me as im a lazy slacker. but for @randvegeta be so amazed of such generic info is rather strange.
@stefeman
Funny. Crimea became ukrainian because khrushchev gave it away as a gift - illegally, of course, but at that time (short after Stalin) nobody dared to challenge the "great leader" khrushchev. So, for a start, the act of rewinding a criminal act is hardly annexation. Moreover (look it up) annexation implies an act against the will of the people; The Crimeans, however, wanted to return back home to Russia as the referendum clearly shows. So, no matter how you turn it, it was not an annexation. But hey, don't let facts disturb your views ...
(P.S. Just remembered: Even the carnegie foundation, hardly suspected of being pro Putin, after making polls in Crimea to show that the evil Russian had lied, had to (very quietly, of course) confess that "somewhat more than 80%" of the Crimeans wanted to seperate from nazi-ukrostan and return to Russia)
Funny that you say that. I myself always explain the complexity of our field by comparing it to medicine.
Pardon me, I don't mean to attack you, honestly, but what you wrote there just shows how clueless you are.
How infamous! How can @randvegeta possibly dare to have his own view! If that man had any manners he would know that, before speaking his majesty "lazy slacker" stefeman is to be asked first.
To put it bluntly: @William knows and you don't; in fact you do not even know that you don't know.
Proposition: You may continue to enjoy the liberty of saying what you feel like saying - but: randvegeta and others are free to speak their mind, too, and chances are that they at least have some vague idea what they are talking about. Fair enough?
To be fair, I am easily amazed. For instance, I'm amazed at how this thread is turning out. Fascinating stuff!
So Its OK to forcefully reclaim other countrys recognized land 100 years after.. sure man.. Russia logic. And nobody wanted to join Russia except the green invaders.. what are you smoking? You had two options, vote yes or be subject to cleanising after the area is annexed forcefully anyway.. If the other option is to be death on a ditch, I would sing praises of god putin everyday as long as his gunmen patrols outside my house/neighbourhood.
The vote was a good joke at best. No 3rd party observers and those that were invited were highly pro russian like Johan Bäckman from finland for example. This same guy claimed that Russian tourist kids are kidnapped by our state and given to finnish families so they can be teached to hate their former homeland. Some quality people they had in there to observe the vote.
Even the Russian newspaper passed it as tongue in a cheek article a day later.
How so? You make a claim, yet you fail to counter my argument.
Im just stating my opinion on his strange reply and questioning the depth of the earlier content when compared to the reaction.
Im not denying anyones freedom of will, but thats probly you ppl can not relate to out there in Moscow. You like to force your words on people.
It seems like you don't appear to know what you assume about other people for being intentionally so small worded. Based on this convo so far, I bet you can write 8 page essay about some subject without touching the actual subject/issue at all.
Whos denying their freedom of speech?
Crimea was given as a gift in 1954. Evidently you don't have the slightest clue what you are talking about.
It's amazing how stupid people can be.
So you seriously believe the fairy tales of Putins KGB henchmen having pointed guns at the heads of Crimeans? (And don't you worry about the KGB not anymore being in existence. Those facts just disturb the propaganda).
It will shock you and certainly make me look like the ultimate Putin troll in your eyes but even western observers of that referendum had to admit that it was done properly. The green men you love to mention were indeed there - to protect the Crimeans from the ukro-nazis.
Those nazis burnt hundreds of ethnic russian people alive, they raped ethnic russian women, killed them and put them into mass graves, and they paraded openly showing nazi regalia, flags, and symbols. And here comes the best part: all of that is documented and publicly available. Unlike Putins KGB henchmen or Putins invisible batallions in ukrostan for whom there was never any proof.
Btw, my main point is not "Putin is great*. My main point is that I hate dirty lies and utterly dumb false assertions, no matter regarding whom.
Ok, so 60, not 100 years. But all that point is moot as long as russia guaranteed ukraine borders in an international treaty in exchange for it's nuclear weapons. Once it invaded it, it is in a breach of international law, and dont tell me the big green men were locals or volunteers helping the poor oppressed tartars to break free from the fascists. Referendum or whatever local ocupation forces and traitors did is irrelevant past that point. Until the legal situation is not re-established, the law states that everything that came after the invasion whcih broke the treaty obligations is null and void. You are too bright to believe your own propaganda.
The sad thing is NOBODY benefits from it except Putin and his regime. Not the Ukraininans, not the tartars, not even the Russians there or in Russia, all suffer the mafia state, the police and occupation army abuses, forced conscriptions etc, same as in eastern Ukraine under Russian control, albeit not outright anexation yet. You probably do not know about the abuses in all russian puppet regimes, from transistria to chechenia. Mafia states much worse than in Russia where is only corruption and oligarchy for now, but will eventually become a full mistico-militaristic mafia regime.
The people which believe and act on propaganda today, will be the victims of the wars and famine tomorrow. Whether that will happen in 5 or 10 years is irrelevant.
I know many people from Crimea. I know of none who are unhappy about the result. Crimea is predominantly ethnic Russians. Language, culture and heritage is mostly Russian.
Western media likes to portray it as an annexation but the vote was overwhelmingly pro-Russia. Where is your evidence of Russian coercion on election results?
Generally speaking, the people in Crimea are economically better off. And they are as free as any Russians to move and travel. And I repeat, every Crimean I know (ok.. not really that many, but about a dozen or so) are all happy about being part of Russia.
Who are you to deny them this when it was voted on peacefully?