what
:grin:
New miner code "works". At half the rate of the old one and produces nothing but invalid shares.
So I guess rate is really 0% of the old one
lol good work duck
Thanks
Whatever asshole made this pow was not worried about ease of programming
didnt you suggest he may have been drunk?
no no, the explorer was written by a drunk
the pow just a jerk
what is the POW algo?
6 passes of reading and hashing
spinning around 3 times really fast in your rolly chair.
would have made it more but each pass has an associated proof in the block header
and was trying to keep those reasonable
spinning too much makes you dizzy
which is why 128-bit hashes in those proofs
does GPT vs MBR for formatting a drive have any bearing on snowblossom performance?
Amost certainly not
woo! produces valid shares now
ok, it is same speed when not involving ram
wtf
ram threads freaking out 24x7 problem tax cpu too much
you know things have gone off the rails when you are typing Thread.yield() and hoping for better results
i got a Plextor M8Pe 1TB Nvme M.2 and this adapter to put it in https://www.amazon.co.uk/Akasa-Internal-ak-pccm2p-01-M-2-Card-Interface-interface/dp/B01LZMIBVP but it hashes only 25 kh/s with CPU at 100% (i7 6700), 40, 60 or 80 threads
I don't suppose there's a setting I could change that would yeild a better performance?
fuck me, yield() is giving better results
@THX 1138 4EB don't ask me, I've had terrible luck getting any sort of hardware to do what it should
:disappointed:
If the specs are correct, 85kh/s is the theoritical max for that drive
how do you determine that?
data rate (2000mb/s according to amazon)
so 2000*1024*1024 / 4096 / 6 = hash rate per second
4096 because it takes an entire 4k page to read a word
and 6 passs per PoW
i think this says the random read speed is 91 MB/s http://ssd.userbenchmark.com/SpeedTest/175630/PLEXTOR-PX-1TM8PeG which is so much less it must be a different value? UserBenchmark: Plextor M8Pe NVMe PCIe M.2 1TB PX-1TM8PeG
91 MB/s is terrible
well on user benchmarks it's one of the best scored for random read, losing only to intel optanes which top out at 176 MB/s
if that were true, you'd only get 3.8kh/s
I get 1800MB/s from a five year old intel nvme
I don't know what sort of crack that site is smoking
what is this data in the 4k read column?
no idea
but you can get 80MB/s from a decent hard drive
not random, but whatever
ssd doesn't have seek time
unless they are doing something super stupid like read, wait for data, read next chunk, etc
you need to parallelize to saturate ssd on random read
That is the only way those numbers make any sense to me
i see
anyways, try cranking your threads way up
like 200 or 500
ok, wilco
i know some people have said to wait two hours but in my experience it seems to level out pretty quickly (now at 24K with 200 threads and not going up any more; will try 500 threads soon)
I'd say about 2 minutes gives you a good idea in this setup
if you are using memfield or depending on cache hotness that is a different story
but you aren't expecting any cache hits really
oh ok
alas, 500 is no better than 80 :disappointed:
hopefully someone can advise what makes my system depend so much on CPU to get the IO from my drive. maybe something isn't implemented well (Asus Pro Gaming Z170 motherboard)
What is the CPU?
i7 6700
quad core, 3.4 ghz?
uhh, quad core for sure but i don't know what it's frequency is
that's fine
should be plenty
there's only one 6700, right? it's not the K version
if anyone wants a fun time, the new miner is testable
in the jeanluc branch, called Arktika
1800MB/s?!
I tried, but can't run if anyone wants a fun time, the new miner is testable
can you share config example with new params?
#snow_path_list=mem_4,/var/nvme_intel/snow/chunk #thread_list=16,256 snow_path_list=/var/nvme_intel/snow/chunk thread_list=64
you give it a list of paths and threads for each location
mem_#gb means load that many gb in ram
also, you need the snow file broken in 1gb chunks
if I want only in memory ?
it won't be faster
Sad