cuda error out of memory nicehash что делать
Nicehash Miner 2.0.1.1 CUDA error ‘out of memory’ in func ‘cuda_eq_run’ #1294
Comments
Hitu63 commented Sep 30, 2017
Hey,
i’m using nicehash miner 2.0.1.1 with Win 10 pro 64bit.
My Rig is build with 4GB RAM and 8x GTX1080ti.
Sometimes, nicehash miner is freezing after a couple of hours mining, and showing:
CUDA error ‘out of memory’ in func ‘cuda_eq_run’ line
I already set up the virtual memory to:
Startsize: 500MB
Maximumsize: 90.000MB
Someone knows a solution or reason for this problem?
The text was updated successfully, but these errors were encountered:
Timoulis commented Oct 1, 2017 •
Really. 4GB RAM and 8X GTX 1080ti?
Only Windows need the half of it.
OS can not handle the virtual memory because there is no left of it.
Add more RAM and you will be ok.
j2inet commented Nov 1, 2017
I get the same error consistently at startup when it’s trying to test the mining algorithms.
GTX 1080 with 8 gigs video RAM
16 gigs system memory
Core-i7 6700 @ 3.4 GHz
It works fine on another system I have with a GTX1080 and a GTX1060.
ksze commented Nov 3, 2017
Pretty similar issue as @j2inet.
GTX 1080 with 8 GiB video RAM
32 GiB RAM
Core i7-6700
Windows 10 Pro 64-bit Fall Creator Update
GeForce driver version 388.13
CUDA 9.0.176
mitch619911 commented Dec 3, 2017
I am running 2 x GTX 1060 3GB cards with 4GB ram and am getting this error. still need more ram.
I have been running for almost a year without errors and now I get it?
mitch619911 commented Dec 3, 2017
I’m using 2.0.1.5 Beta.
increasing virtual memory?
mitch619911 commented Dec 3, 2017
mitch619911 commented Dec 3, 2017
Starting to happen hourly 🙁
mitch619911 commented Dec 3, 2017
mitch619911 commented Dec 3, 2017
Followed that post but it looks like its smaller now?
mitch619911 commented Dec 3, 2017
I tried this:
https://mitchs.tech/i/188623.png
Gonna run it for a few hours and see what happens.
mitch619911 commented Dec 4, 2017
mitch619911 commented Dec 4, 2017
Please note that we will fix this issue with the next release. Till then we suggest you disable NeoScrypt in your miner. Thank you for understanding.
Kind regards,
NiceHash Team»
Johnb21 commented Dec 23, 2017
10ktoday commented Dec 28, 2017
westy666 commented Dec 29, 2017
Same issue here as well with over the past couple of days with 2.0.1.5 beta.
Have 8gb systems ram running 2×1070’s and 1×1060 with Windows 10 and latest Nvidia drivers.
Just seen it happen in front of my eye’s. Seems to get the error then keep restarting the miner window then eventually freezes.
gofres commented Dec 30, 2017
I have identical set up to OP and getting the same issue using both the latest NH and NHML. It ran stable all over the Christmas period but has since been crashing a few times a day.
Virtual memory set to 32gb which should be plenty.
I’ve sat and watched it run for ages to catch an error. It seems like each algo switch adds more load to RAM until it crashes. Haven’t tried setting a single algo yet to test.
Don’t get any issues using other miners.
Anyone found a solution yet?
CyrptoCraig commented Jan 5, 2018
I just had this same issue and went to google for help but everyone was just on about the VM and i didn’t think it could have been that.
I then looked at all my cards and thought i’ll get rid of any OC settings on afterburner and it worked fine.
I have 1 1060 and when i turned down the memory clock on afterburner it worked fine so might be worth checking overclocking settings.
Hope that helps others.
ThePhoneGuy65 commented Jan 10, 2018
ThePhoneGuy65 commented Jan 10, 2018
ThePhoneGuy65 commented Jan 10, 2018
just tried to open Google chrome, while mining w/ 3 cards. got an error. » not enough memory to open »
tried again. display was flashing until I stopped the miner. Immediately opened.
so. virtual memory.
morphk3t commented Jan 11, 2018
I have found myself in the similar situation that people have above. I currently have a 6x 1070 Asus Turbo that was setup with Nicehash and made a few changes by changing the VM to 16GB and also adding the following environmental variables:
Name: GPU_FORCE_64BIT_PTR
Value: 1
Name: GPU_MAX_ALLOC_PERCENT
Value: 100
Name: GPU_SINGLE_ALLOC_PERCENT
Value: 100
Name: GPU_MAX_HEAP_SIZE
Value: 100
Name: GPU_USE_SYNC_OBJECTS
Value: 1
After doing so it seems to have fixed the issue, or so I thought. Seems that when it runs neoscrypt is where it crashes and restarts the miner. Have not seeing it do it on any other algo. If I recall correctly, one can disable certain algos from mining, no?
morphk3t commented Jan 11, 2018 •
NVM, seems I found the issue listed here, if issue is related with neoscrypt: https://github.com/nicehash/excavator/issues/78
duracoin commented Jan 11, 2018
i had the same issue with a 6 card rig. Solution is to set pci slots to gen1 in bios and also set virtual memery in windows performance settings to the following: min:4000 max:45000
after this it works fine. if you have more cards keep increasing virtual memory untill it is stable. i had mine at 35000 before i made the changes now its fine.
ThePhoneGuy65 commented Jan 12, 2018
I do have my GTX cards set to Gen2. so I will try that for sure. I adjusted the virtual Memory already.. but will double check & report back. Thank you » Duracoin «.
mitch619911 commented Jan 18, 2018
Seems to be Nicehash this time.
DIsable Neo, https://mitchs.tech/i/2073cd.gif
ellensp commented Jun 13, 2018
I just struck this error. I upgraded from 5 * 1060 6gb rig to a 6 *1060 6gb ie added 1 more GPU.
PhoenixMiner and Claymore’s Dual Ethereum also reported memory allocation issues in cuda.
The issue for me was the swap size. I had it set to 16000mb but it needed more. Is now working with 24000MB swap. Why? I dont know but I suspect the cuda drivers generate the DAG in real memory for all cards then move it into the GPU’s memory. Since the dag at present is
slonuh commented Jul 4, 2018
Elensp, brilliant solution! It helps me.
So everybody, you should set minimum Windows Virtual memory swap according summ memory of your GPU’s.
Ошибка «out of memory» при алгоритме NeoScrypt
Stee1Arm
Новичок
Добрый вечер.
Подскажите пожалуйста, в чем может быть причина:
Майню на Nicehash, при переключение на алгоритм NeoScrypt вылетает ошибка «out of memory».
Майнер перезагружается и с ново выдает ту же ошибку, так по кругу пока не поменяет алгоритм.
Когда выставляю NeoScrypt только на 2 картах, все работает хорошо, больше 2-х вылетает с ошибкой.
Ферма:
8x Palit GTX 1080 Ti JetStream 11GB
1x Corsair 1000W 80+ Gold RM1000i
3x Corsair 850W 80+ Gold RM850
Asus Prime Z270-A + G4400 3.3Ghz 3MB
8GB DDR4 2400Mhz
Kingston SSD 120GB
Файл подкачки ставил 16000 и 24000 не помогает.
Бывалый
Добрый вечер.
Подскажите пожалуйста, в чем может быть причина:
Майню на Nicehash, при переключение на алгоритм NeoScrypt вылетает ошибка «out of memory».
Майнер перезагружается и с ново выдает ту же ошибку, так по кругу пока не поменяет алгоритм.
Когда выставляю NeoScrypt только на 2 картах, все работает хорошо, больше 2-х вылетает с ошибкой.
Ферма:
8x Palit GTX 1080 Ti JetStream 11GB
1x Corsair 1000W 80+ Gold RM1000i
3x Corsair 850W 80+ Gold RM850
Asus Prime Z270-A + G4400 3.3Ghz 3MB
8GB DDR4 2400Mhz
Kingston SSD 120GB
Файл подкачки ставил 16000 и 24000 не помогает.
pilat200
Свой человек
Добрый вечер.
Подскажите пожалуйста, в чем может быть причина:
Майню на Nicehash, при переключение на алгоритм NeoScrypt вылетает ошибка «out of memory».
Майнер перезагружается и с ново выдает ту же ошибку, так по кругу пока не поменяет алгоритм.
Когда выставляю NeoScrypt только на 2 картах, все работает хорошо, больше 2-х вылетает с ошибкой.
Ферма:
8x Palit GTX 1080 Ti JetStream 11GB
1x Corsair 1000W 80+ Gold RM1000i
3x Corsair 850W 80+ Gold RM850
Asus Prime Z270-A + G4400 3.3Ghz 3MB
8GB DDR4 2400Mhz
Kingston SSD 120GB
Файл подкачки ставил 16000 и 24000 не помогает.
CUDA error in CudaProgram.cu:373 : out of memory (2) #1857
Comments
WillianSalceda commented Nov 7, 2019 •
CUDA version: 10.0, CUDA runtime: 8.0
Available GPUs for mining:
GPU1: GeForce GTX 1050 Ti (pcie 2), CUDA cap. 6.1, 4 GB VRAM, 6 CUs
Nvidia driver version: 441.12
Eth: the pool list contains 1 pool (1 from command-line)
Eth: primary pool: daggerhashimoto.br.nicehash.com:3353
Starting GPU mining
Eth: Connecting to ethash pool daggerhashimoto.br.nicehash.com:3353 (proto: Nicehash)
GPU1: 30C 30% 36W
GPUs power: 35.6 W
Eth: Connected to ethash pool daggerhashimoto.br.nicehash.com:3353 (172.65.195.159)
Listening for CDM remote manager at port 4000 in read-only mode
Eth: Subscribed to ethash pool
Eth: Worker 3KbfgWhzLi4QrM6METY3aAzAA1NBhWj7j9$0-FfPFLHLcjVWIwstHKh8jHw authorized
Eth: New job #79ff35a2 from daggerhashimoto.br.nicehash.com:3353; diff: 8590MH
GPU1: Starting up. (0)
GPU1: Generating ethash light cache for epoch #296
Eth: New job #b158f306 from daggerhashimoto.br.nicehash.com:3353; diff: 8590MH
Eth: New job #7f4f2f7d from daggerhashimoto.br.nicehash.com:3353; diff: 8590MH
Light cache generated in 2.6 s (20.7 MB/s)
GPU1: Allocating DAG (3.33) GB; good for epoch up to #298
CUDA error in CudaProgram.cu:373 : out of memory (2)
GPU1: CUDA memory: 4.00 GB total, 3.30 GB free
GPU1 initMiner error: out of memory
Eth speed: 0.000 MH/s, shares: 0/0/0, time: 0:00
Eth speed: 0.000 MH/s, shares: 0/0/0, time: 0:00
Eth: New job #831b4fb4 from daggerhashimoto.br.nicehash.com:3353; diff: 8590MH
Eth: New job #e68b5bc2 from daggerhashimoto.br.nicehash.com:3353; diff: 8590MH
Eth speed: 0.000 MH/s, shares: 0/0/0, time: 0:00
Eth speed: 0.000 MH/s, shares: 0/0/0, time: 0:00
Eth: New job #f7f9bd07 from daggerhashimoto.br.nicehash.com:3353; diff: 8590MH
Eth: New job #be9671bf from daggerhashimoto.br.nicehash.com:3353; diff: 8590MH
Eth speed: 0.000 MH/s, shares: 0/0/0, time: 0:00
GPU1: 30C 30% 36W
GPUs power: 35.6 W
Eth speed: 0.000 MH/s, shares: 0/0/0, time: 0:00
Eth speed: 0.000 MH/s, shares: 0/0/0, time: 0:00
Eth: New job #40399f1d from daggerhashimoto.br.nicehash.com:3353; diff: 8590MH
Eth: New job #c947cfff from daggerhashimoto.br.nicehash.com:3353; diff: 8590MH
Eth: New job #9a1e3f7f from daggerhashimoto.br.nicehash.com:3353; diff: 8590MH
Eth speed: 0.000 MH/s, shares: 0/0/0, time: 0:00
*** 0:00 *** 11/7 17:13 **************************************
Eth: Mining ETH on daggerhashimoto.br.nicehash.com:3353 for 0:00
Eth: Accepted shares 0 (0 stales), rejected shares 0 (0 stales)
Eth: Incorrect shares 0 (0.00%), est. stales percentage 0.00%
Eth: Average speed (5 min): 0.000 MH/s
Eth speed: 0.000 MH/s, shares: 0/0/0, time: 0:00
The text was updated successfully, but these errors were encountered:
Исправляем ошибку при майнинге Ethereum на GeForce 1050
Все кто занимается добычей криптовалюты, рано или поздно сталкиваются с определенными проблемами в работе оборудования. Часто, решаются они быстро, а в большинстве случаев, вовсе проходят сами, после автоматического обновления ПО. 
Сегодня же, поговорим о ошибке «CUDA error», а если быть точнее, то о том, как её победить. Я лично столкнулся с подобной проблемой, по-этому знаю о чем говорю. Не буду мучить вас предысториями о том, как у меня все начиналось и как я от этого страдал — это лишнее. Давайте сразу приступим к сути вопроса и его решению.
Если вы занимаетесь добычей Ethereum (эфириума), и у вас возникает такая ошибка (ниже выложу скрин), значит мой метод решения именно для вас: 
Решение ошибки CUDA error
Шаг 1. На рабочем столе находим «Мой компьютер» нажимаем на него правой кнопкой, и выбираем последний пункт «Свойства»
Шаг 2. Идем по пути: «Дополнительные параметры системы» — «Дополнительно» — «Параметры» — опять «Дополнительно»
Шаг 3. Нажимаем кнопку «Изменить» возле надписи «Виртуальная память»
Шаг 4. Вводим нужную цифру (напомним, что она не должна быть ниже 16 000 Мб)
Собственно все. После таких манипуляций, мой риг сново ожил, а майнинг вернулся на прежний уровень дохода. Для тех кто не понял, что нужно делать, ниже оставлю видео, как это выглядит у меня, а там уже разберетесь. Собственно вот, как и обещал!
[creativ_media type=»youtube» url=»www.youtube.com/watch?v=Y-F-VcsUUpI»]
Не знаю, поможет ли вам этот способ, но мне помог. На самом деле, ошибки связанные с файлом подкачки, довольно таки частые, а данный метод поможет их решить. Всем хорошего майнинга и меньше ошибок при добыче крипты!
CUDA_ERROR_OUT_OF_MEMORY: out of memory #201
Comments
vans163 commented Nov 29, 2019 •
Is this normal 440.36 driver on Ubuntu 19.10. Randomly happens. GTX 1660 Turing cards. Before I used 1060s and never had this error, could even run 20 decoding/encoding sessions.
The text was updated successfully, but these errors were encountered:
paulhothersall commented Nov 29, 2019
19.10 seems to use like 1GB of Ram just using the desktop.
Each transcode even on a different GPU seems for me to be taking a 55MB resource hit on GPU0 thats additional to the actual encode, so in my case with 2 x Quadro 2GB cards its very easy to hit a ram limit.
I have also found if you are doign lots of short transcodes, that the ram Lingers before being released. So you might be hitting that issue?
I charitably believe this might be why Nvidia might want to limit the number of sessions. They for sure don;t seem to overly care on the drivers so long as it «works» for the amount specified on the matrix.
Are you doing some extra steps?
Snawoot commented Nov 29, 2019
Please verify you have enough VRAM to fit your encoding session. In order to do that:
vans163 commented Nov 29, 2019 •
The VRAM is 6GB, I am only running 1 FFMPEG process per card with at most 2-3 multiplexed encodes.
All those ffmpeg processes are stuck on transcoding, they have been running 6hr+ and the encoder is at 100% utilization.
I do have a lot of short transcodes for certain use cases, maybe I should sleep some time between them? But overall it seems to just crap out randomly, even though the cards have 4gb+ free VRAM available.
This is how the command looks like, I am thinking maybe its due to the scale_npp?
Another random error.
Everything worked perfectly smooth and fine with ubuntu 16.04 and 6GB 1060s, same driver.
Snawoot commented Nov 29, 2019 •
Everything worked perfectly smooth and fine with ubuntu 16.04 and 6GB 1060s, same driver.
As far as I know 1660 is based on Turing architecture and it’s NVENC engine pretty much different from 1060 counterpart. It may have different memory requirements from Pascal ones.
Here is some things to try:
This one is clearly related to decoder. It complains about incorrect gpu number. If this error describes situation correctly, it means:
vans163 commented Nov 29, 2019
vans163 commented Nov 29, 2019
So I was just testing a few things manually and randomly starting a session got
Then immediatly tried again the same ffmpeg command and everything worked. So weird and inconsistent.
Snawoot commented Nov 29, 2019
That’s clearly a memory allocation error. Without regard to memory-related error code, practically nvEncCreateBitstreamBuffer does nothing more than a memory allocation. What is your stream resolution?
vans163 commented Nov 29, 2019
Its weird I run it again and everything works. Like a 1 time error, even tho the card has 0 processes on it and 0 used vram.
Snawoot commented Nov 29, 2019
It’s worth trying to modify ffmpeg source (I guess you built from source for NVENC support) and log amount of memory requested by nvEncCreateBitstreamBuffer. I think this number may vary.
vans163 commented Nov 30, 2019 •
So the CUDA_ERROR_OUT_OF_MEMORY: out of memory seems sporadic, the fix is just to restart the job if it appears. Everything works fine there is enough memory, maybe my system is unstable / bad ACPI / bad PCIE slot?
Snawoot commented Nov 30, 2019
Can’t say for sure. You can try files with larger and smaller resolution and if issue never reproduces with smaller files and always reproduces with larger files, it’s not that sporadic.
1265578519 commented Nov 30, 2019
vans163 commented Nov 30, 2019 •
Can’t say for sure. You can try files with larger and smaller resolution and if issue never reproduces with smaller files and always reproduces with larger files, it’s not that sporadic.
Like the same exact ffmepg commandline (job), on the same GPU, if it gets the out of cuda memory error, and I start the job again 100ms after, it works fine, nothing changed on the GPU itself in regards to memory. Its like a random bug which makes me think its related to shoddy hardware.
If this was the case, then job would never work. So having too large a res is not the problem. Also these jobs are below the threshold for h264. (4096×4096)
1265578519 commented Nov 30, 2019 •
Max # of concurrent sessions 2
你补丁未使用成功，就会提示这种错误，达到并发数量2，你重新安装patch sh试一下
如果是随机性发生，可能是补丁有一个BUG问题，等待作者查看
这个错误和内存使用无关，是达到了最大数量限制2，和输出错误的width or height也没有关系，这个仅仅是触发了驱动并发限制的固定提示
vans163 commented Jan 27, 2020
@1265578519 It seems to be the case, randomly triggered, I just make it restart the job, and its fine the error is gone. Maybe some kind of race condition with the patch and driver interaction?
1265578519 commented Jan 27, 2020
vans163 commented Jan 27, 2020
It only started happening for me on Turing cards, Pascal cards did not have this problem.
Snawoot commented Jan 27, 2020 •
@vans163 Is there a chanсe you have two drivers installed in your system? Like system packaged driver from repos and driver installed by run-file from Nvidia site?
@1265578519 Could you please use English? I guess no one will be happy if I’ll speak exclusively in my native Russian.
vans163 commented Jan 28, 2020
Maybe something is funky, I installed the official ubuntu package driver, noticed the issue, removed it placed NVIDIA blob driver, same issue (theres like a set of 3 different errors, but they all are non-critical). I could keep an eye on and eventually report if I build another rig, it might be some hardware issue?
Snawoot commented Jan 28, 2020
Well, I had hardware issues only with power supply unit and it peeped out when encoding stops. It seems sometimes it caused one of PSU voltages rapidly go back to normal level and system hang. But no hardware issues I had were causing clear software error code CUDA_ERROR_OUT_OF_MEMORY (which, I admit, is indeed inherent to session limit).
However, there are few notable points in problems you described:
So, yes, I believe your issue is specific to your platform, but I can’t rectify reason.
But I’ve got some fresh idea for you: best patch is no patch at all. You said before normally you have one or so video encoding session per GPU. I guess only reason why you have to use this patch is multiple GeForce GPUs still have total limit in 2 encoding sessions (also, this situation effectively invalidates hypothesis that there is any real reason for 2 sessions restriction). But if your system is VT-x capable, you can create few virtual machines, each for one GPU, and assign GPUs to VMs with pci-passthrough. You’ll have separate systems with distinct limits for each card in one physical box. BTW, pci-passthrough into KVM virtual machine is how I do tests: I have Linux host system on my PC, and assign GTX 1080 to Win10 or Ubuntu VM for tests. Here is resources which might be helpful: