@Clueless @Fireduck should I be getting 0h/s while hybrid mining loads cache into mem?
I don't remember, but I think so
once it's loaded, what do you get?
it never seems to fully load
and it stays at 0
Usually what happens is out of memory error and then it restarts. Not ideal. On bug list.
I think @Tyler Boone is in charge of that
@Tyler Boone has joined the channel
People like being volunteered for things, right,?
So now I'm trying to use the prebuilt release. It only uses 170gb compared to 210 with the other miner that I built, but it doesn't seem to be fully utilizing the cores. Its only using about half of my cpu and the hr is only half of what it should be. Threads are set to 96 on a 48core
my hash too low only 1.48M Am I setting something wrong?
@ppp is mining in memory and is getting about 1.48M/s I think this is normal, can anyone verify if that looks right?
@Shoots was speaking of that 2 weeks ago, i think its because of threading, each thread take a part ram to ping pong through hash and mine, more thread > more ram used , prebuilt dont use much ram but not enough precached to ba able to maximize core use, build from source use core at max but at the cost of more ram by thread, but probably to much depending on the freq of you core.
I suggest you try this, given that you are running into a thread limit. 1. Turn off your miners. 2. Create a "Ram Disk" using a windows utility (a ram disk make a *temporary* disk in ram, that shows up like *D:/*) 3. *copy* (do not delete/move) *copy* the snow folder to the ram disk 4. Modify your miner.conf to use the new disk, eg `snow_path=D:/snow` 4. Run *multiple miners* You'll be able to use more than 64 threads
---
But, that might not help at all. You might already be saturating your ram IO
@Clueless you’re just adding syscall overhead per seek there
@ppp that’s not bad - how many memory channels?
network hashrate is piling back on
`15.453 minutes per share ` wow
@stoner19 at what diff?
29
Did the pool upgrade to the latest check-in?
how’d the diff creep up that high for you?
no idea
Makes diff increase for pool shares less aggressive
@Fireduck is it the pool or the miner?
still would be nice to know who is running vauxhall. still no stats coming from them other than on the explorer page about what blocks they get
@Fireduck bumped my pool yesterday as i saw diff 30 on a miner of mine - now it drops miners from stats if their diff creeps up too high
damn vauxhall just got 7 blocks in 15 minutes
I have a feeling they have a lot of hr
@Shoots seems like the early whale - also pumped diff at will
yeah 5yg at it again
he was with us on protopool but it went down last night, ran out of mem again. Protovist has it back up and running and I think he has it setup to restart now
seems like the pool has a memory leak or something
ate through 24gb of memory like nothing
mine has yet to exceed 300MB, but i have a managed jvm where i slot that in
@Rotonen two
@ppp sounds implausibly high for a dual channel system - which processor?
2686v3
@ppp that’s a 4 channel processor https://ark.intel.com/products/81055/Intel-Xeon-Processor-E5-2683-v3-35M-Cache-2_00-GHz i’m getting about 850k per socket on haswell E5s - you have two sockets? Intel® Xeon® Processor E5-2683 v3 (35M Cache, 2.00 GHz) quick reference guide including specifications, features, pricing, compatibility, design documentation, ordering codes, spec codes and more.
@Rotonen The pool drops miners from stats if they get too high? That sounds like correct or reasonable at all. Is that something you can reproduce?
@Rotonen How many memory sticks can build four channels
x10dai
@Shoots the pool really shouldn't have any memory problems. Not impossible since it has to keep a bunch of pending work units and share information in memory but it really shouldn't happen.
@Fireduck ’toi high’ as in something like over 1h per share
@ppp you are confused over terms and what is what - go read up on motherboard block diagrams and also on how memory channels work
@Fireduck at this point i find it more likely the operator gave the pool a high heap size and then the oom reaper did its thing
@Rotonen ok, cool. That all makes sense.
@Fireduck Jun 26, 2018 5:14:24 AM snowblossom.miner.MiningPoolServiceAgent recordShare INFO: Share recorded for snow:vx8pglccl7nepzre9h3fraetreqn924yw7mlww4s - (32 sha res) (134217728 hashes) OpenJDK 64-Bit Server VM warning: INFO: os::commit_memory(0x0000000723880000, 24 641536, 0) failed; error='Cannot allocate memory' (errno=12)
32GB VM
@Protovist can you tell me what jvm args that was run with?
default
2895 ubuntu 20 0 13.784g 1.379g 24160 S 0.3 4.5 0:56.76 java
Will check it later to see if it's grown. :slightly_smiling_face:
it should be no where near that large
looking into it
@Protovist is it possible someone is mining with a huge fleet (thousands of clients) ?
It was around 500 at the peak IIRC.
hum, that should be no problem
currently 159 workers
I don't see a way where the memory would go out of control
If you were to capture a heap dump while running, that would be helpful
jmap -dump:format=b,file=dump.bin PID
and send me that stupid file
or open it up in mat and tell me what the hell is has way too many of: https://www.eclipse.org/mat/ The Eclipse Foundation - home to a global community, the Eclipse IDE, Jakarta EE and over 350 open source projects, including runtimes, tools and frameworks.
Basically, the the pool any object we have more than 100,000 of is wrong
here's an example of something which will probably not be great for snowblossom, but will lure people with price: https://www.anandtech.com/print/13010/realtek-nvme-pcie-rts5762-rts5763dl-controllers New Realtek PCIe 3.0 x4 NVMe Controllers: RTS5762 & RTS5763DL up to 3.5 GB/s
basing on the fact random reads are almost hidden in all materials
I remember back when realtek was the bane of any networked pc
and the enterprise connector space is getting confusing: https://www.anandtech.com/print/13015/aic-shows-dual-nf1-to-u2-ssd-adapter AIC Shows Dual NF1 to U.2 SSD Adapter
but hey, more passive small stuff (seasonic does not do sfx so silverstore reigns there): https://www.anandtech.com/print/13019/silverstone-launches-nj450sxl-psu-fanless-sfx-l-450w Silverstone Launches NJ450-SXL Power Supply: Fanless SFX-L, 80Plus Platinum, 450W
very cool
probably not, but robust :stuck_out_tongue:
right
one probably wants that at the top of a case and nothing else on top of it
what if the cat likes to sleep there?
cats are self directing in that, they'll just move away if it's not pleasant
the psu should not care, unless the cat goes into the case and into the air inlet
but then the psu is probably not the biggest issue at that point
I usually have a least a few systems with no case
wall mounting is an art form
power 9 and epyc look pretty damn good for snowblossom based on the pg and 7zip benchmarks https://www.phoronix.com/scan.php?page=article&item=power9-talos-2&num=4 Phoronix is the leading technology website for Linux hardware reviews, open-source news, Linux benchmarks, open-source benchmarks, and computer hardware tests.