• Forum has been upgraded, all links, images, etc are as they were. Please see Official Announcements for more information

Why does X11 run cooler?

jakecrow

New member
I've seen a lot of forum threads lately where people seem to think that the only reason X11 runs cooler is that the miner for it is crap and not optimized well. Many of these people also feel there is an optimized miner out there somewhere. I'm not convinced of that, but the point is this type of thing can hurt adoption by new miners.

So can anyone with technical knowledge chime in here? Why is it cooler? Is there a way to optimize it for more hash but at the cost of higher temps?

This is an honest question, I love DRK and want it to succeed and this is one area I think needs some work. I hate to see new people scared away because they read all this stuff going around about secret mining software.
 
I don't have technical knowledge, but I've been wondering about this as well, and I *think* it runs cooler (yet with up to 3X the hash rate of script) because it's more of a time consuming process than an energy consuming process.
 
Thanks for the replies guys, but these are just guesses right? Don't get me wrong I'm not trying to sound rude, I'd just like to REALLY know what's going on. Maybe no one knows but I feel like there's gotta be someone out there who does. A technical explanation could be very beneficial to the X11 community imo. I'd -love- to be able to tell people "It's not because the miner is unoptimized - it's because of XYZ" and actually be able to back it up with facts.

I just want us to get out to a bigger audience and I see this as a barrier which is, maybe, easy to break down.
 
The answer is that some kinds of processing are extremely easy to scale up on a GPU to 100% utilisation. In distributed computing parlance, these are called "embarrasingly parallel" problems, because they are trivial to scale up on parallel compute hardware. X11 is relatively more difficult (compared to scrypt) to parallelise on a GPU with perfect efficiency. This is because the code has to coordinate 11 different hashing algos, each of which is computed in a different way and in a different timeframe. So hypothetically we might have optimised an X11 miner as well as humanly possible, but it still only manages 50% utilisation of a GPU.

Does that sort of make sense?
 
So ironically, X11 is not very gpu friendly!
If asics ever got out, they should crush gpus more so than ever was the case with sha256 or scrypt.
But that's OK because if we ever get to that point, that can only mean that drk has been a wild success
 
And we could switch to yet another algorithm if they started building ASiCs for X-11. Actually, with everyone jumping the script ship, I think script asics may be a failure, killing that industry. I kind of hope not, as it's good to see industry build up around crypto currency. This is another reason why I'm not so sure exchanges aren't a good thing. They help establish the price of coins as well as act as an intermediary between local currency (fiat) and crypto.
 
^ I agree, in fact I've sold my LTC to BTC. Since LTC is pinned to BTC price, and LTC doesn't really have much innovation and has the potential to be forked, I don't see the point in holding it.

I think ASIC's will not be a failure for the companies that make them, but for the miners...
 
This is another reason why I'm not so sure exchanges aren't a good thing. They help establish the price of coins as well as act as an intermediary between local currency (fiat) and crypto.

Agreed, The more exchanges (good ones, not GOX, a trustless, secured by a block chain decentralized one being the best?) the better. The more people the better. Traders and Arbitrageurs and Consumers actually using a coin make a coin less volatile. At the end of the day coins are only going to have any value if people use them.

Sorry for going so off topic.

Cooler GPU's === AWESOME.
 
I believe it was originally designed as a CPU miner targeted algorithm, so makes sense that it is not GPU efficient.
 
Sorry been away from here a couple days, but thanks for the answers. LimLims your answer makes sense I suppose - basically the miner is as optimized as we can make it, is what you're saying?

My main reason for asking this is because of threads like this: https://bitcointalk.org/index.php?topic=556277.0 . Yeah, I don't believe that there's a secret miner out there doing better than the everyman miner we're all using. That said I think it's safe to say that people will read threads like that and never even bother with X11 because they might feel cheated.

So, is anyone out there trying to optimize it further? Has anyone tried? If it simply can't be done then some legit technical explanations in threads like the one I linked could be beneficial to the community imo. As it stands now you just have trolls and scared people spreading rumors which can't help at all.
 
Sorry been away from here a couple days, but thanks for the answers. LimLims your answer makes sense I suppose - basically the miner is as optimized as we can make it, is what you're saying?

My main reason for asking this is because of threads like this: https://bitcointalk.org/index.php?topic=556277.0 . Yeah, I don't believe that there's a secret miner out there doing better than the everyman miner we're all using. That said I think it's safe to say that people will read threads like that and never even bother with X11 because they might feel cheated.

So, is anyone out there trying to optimize it further? Has anyone tried? If it simply can't be done then some legit technical explanations in threads like the one I linked could be beneficial to the community imo. As it stands now you just have trolls and scared people spreading rumors which can't help at all.

Well, the x11 is basically the algo of qubit, with the algos of quark, but without the randomness of quark, so being extremely naive I'd expect the hashrates of x11 to be somewhere between the two (but closer to quark since it has more hashing algos). So if we look at the hashrates of quark with sph-sgminer, it gets less than x11 (for equivalent scrypt hashing), and x11 gets less than qubitcoin. Also, if you look at both quark and qubit, there are reports that it runs at lower temps too. So it isn't something that just affects x11, but seems to affect the other coins that use some of the algos that x11 uses.

https://bitcointalk.org/index.php?topic=475795

http://forum.qrk.cc/thread/1732/quark-gpu-mining
 
Well, the x11 is basically the algo of qubit, with the algos of quark, but without the randomness of quark, so being extremely naive I'd expect the hashrates of x11 to be somewhere between the two (but closer to quark since it has more hashing algos). So if we look at the hashrates of quark with sph-sgminer, it gets less than x11 (for equivalent scrypt hashing), and x11 gets less than qubitcoin. Also, if you look at both quark and qubit, there are reports that it runs at lower temps too. So it isn't something that just affects x11, but seems to affect the other coins that use some of the algos that x11 uses.

https://bitcointalk.org/index.php?topic=475795

http://forum.qrk.cc/thread/1732/quark-gpu-mining


I have to share, which is my opinion for now, that your logic can be misinterpreted.

Quark has 9 rounds of hashing, with three of them being random. This is chosen by an instantaneous choice depending on the result of the previous hash -- i'll assume it adds 0 time in comparison to the amount of hashing done due to it being an if/else argument.

Qubit has 5 rounds of hashing.

x11 has 11 rounds of hashing.

Assuming more work must be done to complete 11 hashing functions than both 9 or 5, you would expect the time taken to complete the x11 hash to be lower than what it would take to hash either quark or qubit.

This should manifest in an overall lower hashrate . . not something between. This is not the case, as Quark has a lower hashrate than x11.

The only reason for this I can think of is that blake, bmw, groestl, jh, keccak and skein collectively take much more time to compute than luffa, cubehash, shavite, simd and echo . . otherwise those three extra random functions chosen from the pool of the same blake, bmw, groestl, jh, keccak and skein would result in a higher hashrate than darkcoin.

Without data pertaining to the time each function takes individually, I can't really take this much further. What I can point you to is that these functions are not designed to be memory intensive.

Without the intense rewriting of RAM you get in Scrypt, your card is effectively doing much less work. Note that VRM temps with Scrypt are generally higher than core temps . . while VRM temps with these algos are generally lower than core temps.

That is where your power saving is coming into play - you're not using the extra power that Scrypt burns in using RAM . . not that the miner isn't optimized . . just that the hash function itself isn't designed to ever fully utilize your GPU's capabilities . . and never will be because that's what Sha 256 and these chained algorithms are designed to do.
 
I have to share, which is my opinion for now, that your logic can be misinterpreted.

That is where your power saving is coming into play - you're not using the extra power that Scrypt burns in using RAM . . not that the miner isn't optimized . . just that the hash function itself isn't designed to ever fully utilize your GPU's capabilities . . and never will be because that's what Sha 256 and these chained algorithms are designed to do.

I am glad my incorrect and poorly thought out answer lead to a correct and very clear answer :).
 
Well, the x11 is basically the algo of qubit, with the algos of quark, but without the randomness of quark, so being extremely naive I'd expect the hashrates of x11 to be somewhere between the two (but closer to quark since it has more hashing algos). So if we look at the hashrates of quark with sph-sgminer, it gets less than x11 (for equivalent scrypt hashing), and x11 gets less than qubitcoin. Also, if you look at both quark and qubit, there are reports that it runs at lower temps too. So it isn't something that just affects x11, but seems to affect the other coins that use some of the algos that x11 uses.

https://bitcointalk.org/index.php?topic=475795

http://forum.qrk.cc/thread/1732/quark-gpu-mining

Bumping this to correct it - X11 actually uses far more memory than it needs to, which causes it to be a lot easier on the core than it should - it's waiting for memory half the time. While it's nowhere near as memory-intensive as Scrypt, it's still slow for partially this reason. That's why it runs cooler - the compute capability isn't being used very much.

Another reason it runs cool is a lack of occupancy, causing there to be less concurrency. AMD calls them "concurrent waves" or "waves in-flight." If you'll see here (nsfw): https://ottrbutt.com/tmp/skein-analysis.png - you see Skein is using 73 registers, limiting it to three waves in flight. Many of the hashes in X11 have the same level or worse occupancy because of their VGPR usage. An analysis of a rewritten Skein that hasn't been optimized well yet you can see here (nsfw): https://ottrbutt.com/tmp/skein-analysis2.png - 51 registers, allowing another wave in flight, which is a big win. Additionally, drop two more registers from that, and you get yet another. This uses a lot more of the card's compute capabilities.

EDIT: I mean, drop three more regs from that - apparently I can't count.
 
Last edited by a moderator:
Back
Top