[XMR] Monero Mining

[grouper fish]

Thinking about adding a few 290x cards to my rig.

Do you think there will be a big difference in hash with the upcoming 390x cards?

As I understand it they are a rebrand of 290x but with 8gig memory instead of 4gig. Not sure how the memory differance will affect the mining output?

[GingerAle]

Thinking about adding a few 290x cards to my rig.

Do you think there will be a big difference in hash with the upcoming 390x cards?

As I understand it they are a rebrand of 290x but with 8gig memory instead of 4gig. Not sure how the memory differance will affect the mining output?

Honestly, I don't know. From what I've gathered from various inquiries into nvidia owners, the hash rate doesn't scale well with.... pretty much anything.

a 750ti gets 250 - 280 hashrate. From this I assumed a 980 (or .... i forget what it was.. something that shoulda been ridiculous) should have gotten..

well lets see... 750ti has 640 cores, 980 has 2048 cores.

So yeah. If it scaled with cores it would be 800 h/s our of that card. The highest I've ever seen reported is 400 h/s. That coulda been with a 970. Regardless, it was way less than expected.

And meanwhile, this guy:

https://bitcointalk.org/index.php?topic=1026287.0

is reporting 220 h/s on a GTX 570, which has 480 cores.

My hunch is that cryptonight on GPUs has more to do with the speed of each processor that can access 2 MB of fast memory. So perhaps with the increased memory on the new card, you can push more cores through?

I dunno though, and I've only researched nvidia cards - no idea how the ATI architecture differs. I've found it odd that my 750ti only uses 1 gig of its RAM (it has 2 gigs), and I can't seem to push it farther with any thread setting.

[5w00p]

It's closely related to L3 cache.

750Ti has 2MB L3 and the 970 has 1.75MB, IIRC.

smooth or someone posted about it somewhere; I remember reading it.

Because the cryptonote algo needs ~2MB to operate, so if it can be kept in on-die memory, versus (V)RAM, it is significantly faster.

It applies to CPUs and GPUs.
That's why smooth and them always say to use L3 cache size divided by 2 to determine the optimal number of threads to utilize when mining on a CPU.

[smooth]

It's closely related to L3 cache.

750Ti has 2MB L3 and the 970 has like 1.5MB.

smooth or someone posted about it somewhere; I remember reading it.

Because the cryptonote algo needs ~2MB to operate, so if it can be kept in on-die memory, versus (V)RAM, it is significantly faster.

It applies to CPUs and GPUs.
That's why smooth and them always say to use L3 cache size divided by 2 to determine the optimal number of threads to utilize when mining on a CPU.

For CPUs that is true (cache per core). For GPUs, there are so many instances of the algorithm operating at once (each randomly accessing 2 MB of data) that caches of that scale are largely irrelevant. What matters is the pipelining and other low level details of the memory architecture, so it is hard to say how a different GPU will perform without trying it (or very careful low level analysis of the architecture, which in many cases isn't even publicly available).

[turkandjaydee]

If I have 8 Core processor, is it better to use "-t 8" or not using "-t 8"?

so turns out, i got better on 4 thread that 8 thread, is it just visual bug?

[nioc]

AFAIK a limiting factor is the L3 cache and runs best when each core has 2 MB of cache.  This is for more than 2 cores as a dual core always seems faster running 2 cores.

With a 4 core i5 whose specs I don't recall, running 2, 3 or 4 cores gave the same HR with the difference obviously being the CPU and power usage.

[grouper fish]

Guess I'll have to try my way forward then.

Will problably keep of buying the 290x untill some more details about the 300 series are out.

One card that cought my attention is the R9 nano, 175w and faster then the 290x sounds a bit to good to be true. ( http://videocardz.com/56609/amd-radeon-r9-fury-x-r9-fury-x2-and-r9-nano-detailed-some-more )

[turkandjaydee]

AFAIK a limiting factor is the L3 cache and runs best when each core has 2 MB of cache.  This is for more than 2 cores as a dual core always seems faster running 2 cores.

With a 4 core i5 whose specs I don't recall, running 2, 3 or 4 cores gave the same HR with the difference obviously being the CPU and power usage.

huh, that actually make sense.
I should try one by one thread count.

[Eastwind]

AFAIK a limiting factor is the L3 cache and runs best when each core has 2 MB of cache.  This is for more than 2 cores as a dual core always seems faster running 2 cores.

With a 4 core i5 whose specs I don't recall, running 2, 3 or 4 cores gave the same HR with the difference obviously being the CPU and power usage.

I had the same observation. There is no difference between 2, 3 and 4 cores.

[smooth]

AFAIK a limiting factor is the L3 cache and runs best when each core has 2 MB of cache.  This is for more than 2 cores as a dual core always seems faster running 2 cores.

With a 4 core i5 whose specs I don't recall, running 2, 3 or 4 cores gave the same HR with the difference obviously being the CPU and power usage.

huh, that actually make sense.
I should try one by one thread count.

Yes, and measure power usage in addition to hash rate, if that is something you care about. Some configurations may produce the same or slightly less hash with much less power.

[superresistant]

 
Just for you to know, the mixing at monero.crypto-pool.fr is working and online.
We also added a faster payout for everyone by default.

Happy mining.

[smooth]

Just for you to know, the mixing at monero.crypto-pool.fr is working and online.
We also added a faster payout for everyone by default.

Happy mining.

This looks nicely done. Are you sweeping the coinbase payments to an intermediate wallet then mixing from there?

[superresistant]

Just for you to know, the mixing at monero.crypto-pool.fr is working and online.
We also added a faster payout for everyone by default.
Happy mining.

This looks nicely done. Are you sweeping the coinbase payments to an intermediate wallet then mixing from there?

Exactly.

[smooth]

Just for you to know, the mixing at monero.crypto-pool.fr is working and online.
We also added a faster payout for everyone by default.
Happy mining.

This looks nicely done. Are you sweeping the coinbase payments to an intermediate wallet then mixing from there?

Exactly.

Fantastic! Thanks for adding this.

[5w00p]

Just for you to know, the mixing at monero.crypto-pool.fr is working and online.
We also added a faster payout for everyone by default.

Happy mining.

Enfin, ce sont d'excellentes nouvelles !

[DirtyUniverse]

Just for you to know, the mixing at monero.crypto-pool.fr is working and online.
We also added a faster payout for everyone by default.

Happy mining.

Best Monero pool on the World. <3

[GingerAle]

After a quick 'n dirty analysis of the network distribution of hash on monerohash.com (divide hashrate of each pool by the number of miners in each pool), the pool monero.net has a per miner hashrate (137.7 kh/s) that is 40.4X that of the average of all other miners (3.4kh/s).

I can think of 3 reasons for this:

1) Someone is investing significant resources into mining monero in that pool

2) Someone has developed an optimized miner (GPU, etc)

3) And, of course, bonnet

If the reason is #1 - great. Whoever you are, thanks for investing 40.4X the average contributors resources to support the network.

if the reason is #2, congrats to whoever you are for tweaking the code. However, being that the security of the network depends upon the network hashrate, I think it would be beneficial to share this code, if it exists. Granted, if this code and individual exists, they are enjoying wonderful profits right now, so we would need to make the release of this code financially rewarding.

The pool reports finding a block on average every 7 minutes, therefore this pool gets a rough estimate of 2k XMR per day. So if this is an optimized miner, it will probably cost a lot.  

Upon further examination, this pool might be the source of the network flucation (I think someone else may have discovered this before), as evident here: http://monero.net/#pool_stats





However, even at the lowest points, this pool still hashes 600 kh/s, which puts the hashrate / miner at 29X the rest of the entire network.


Edited: interesting. With my new numbers (must have missed something in the copy and paste. woops!) when the low end of the wave is used, the average hashrate / user for that pool goes to 46.16, which is still 13X the average for all other users.

Thus, I propose a bounty dependent upon release of mining software that provides 29X increase in hashrate for my mining rig. I don't have much, but I planned on building a new rig soon. Thus, I will pay 500 XMR for this software (that might not exist), or contribute 200 XMR for a bounty for releasing the software (that might not exist) to the entire network (because if that happens, then the whole hashrate goes up and I'm just burning 200 XMR). Actually, to be honest, if I got it privately I would release it anyway because network, so we should probably try to get the public-release bounty going.

If reason #3, well... then.... people should really install antivirus software.

EDITED some numbers due to copy paste error. Edit time 10:45 AM EST.

[jwinterm]

...
If reason #3, well... then.... people should really install antivirus software learn how to computer.

I don't think AV software is necessarily very helpful in this case, since the miner is likely installed with the user's permission, either bundled with another program that they just click through too fast (a la utorrent), or installed surreptitiously when another program is installed. This has been a topic of discussion lately in the vertcoin thread, as one pool (ipominer) has apparently had a big botnet mining at it's pool that is a substantial part/majority of the network ( https://bitcointalk.org/index.php?topic=895020.1100 ). The pool in question tries to deny that it's actually a bot net, arguing semantics, since apparently people are installing it just by clicking through the installation options too fast when installing some other program.

Anyway, I think there's a few other options that kind of fall somewhere in between option (1) and option (3). Like, if someone works with/owns a large server farm, and their only running it at full blast in off hours, or maybe even a gaming focused internet cafe or something, where they have a bunch of comps equipped with gpus that they can run in off hours. Who knows...

With that being said, I pledge about 3.50 XMR to the bounty proposed in this thread

[GingerAle]

edited some numbers due to copy and paste error. Numbers of that pool are still extreme, but less so.

[newb4now]

edited some numbers due to copy and paste error. Numbers of that pool are still extreme, but less so.

Interesting analysis! Thanks for posting this