Miner.farm Crypto Currency Community Forum

Browse and post your favorite coin/crypto news, miner.farm and PiMP OS updates and announcements, mining guides, overclocking tips, and more...

Quick Links

"Eth-Death" - Understanding Ethash GPU VRAM Requirements

  • Support

    Are you suffering from miner crashes and/or receiving DAG errors when trying to mine Ethash?

    • Your GPU's may be suffering from "Eth-Death."

    Before we can understand what "Eth-Death" actually is, we need to first cover one of the key components in the mining of Ethash:

    The DAG File

    DAG stands for "Directed Acyclic Graph". It is used in all Ethash coins, like Ethereum, Ethereum Classic, Metaverse, Ubiq and other coins to provide a proof of work. The DAG file is generated every mining epoch and it increases from epoch to epoch. It is very important to know current and future size of DAG file because it has a huge impact in relation to usable mining hardware (i.e. GPU's).

    This DAG file is loaded (and stored) directly to your GPU memory (VRAM) during the mining process. If, however, the DAG file size happens to be larger than your GPU's VRAM capacity, then the file cannot be loaded and mining cannot happen. This usually results in DAG errors being reported by the mining software or (worst case) complete crashing of the miner and/or rig.

    Since increasing a GPU's VRAM capacity is not an option, those GPU's, who do not have the necessary VRAM capacity, are technically "dead" to mining the Ethash algorithm. (i.e. They have suffered "Eth-Death")

    Want to check out what the current DAG file size is? There are plenty of calculators on the internet that will work out the value for you.

    "Wait... the current DAG file size is only 3.91GB. My GPU's are 4GB..."

    • Let me stop you right there. Your "4GB" GPU's are most likely not rocking the 4GB you think they are.

    When it comes to computer hardware, end-item manufacturers "round up" when it comes to marketing. Don't confuse this with the whole "1000 is actually 1024" thing. This has more to do with a manufacturing practice known as "product binning".

    In our specific case, we have two major players involved: 1. The chip manufacturer and 2. The GPU manufacturer.

    Chip Manufacturers (Samsung, Micron, Elpida, Hynix, etc...)

    Memory chips, due to their complexity, are not able to be made to exact specs/sizes. After the chips are made, the manufacturer will machine test them to see how much is addressable and usable (along with other quality checks). After testing, the chips are then sorted into groups based upon those results. (Any chips that fail to meet the manufacturer's thresholds/standards are usually recycled back into the process to be made into new chips.)

    The chip manufacturer prices out these chip groups according to their quality and sells them to the end-item manufacturers in bulk.

    GPU Manufacturers (Asus, EVGA, MSI, PNY, etc...)

    As already mentioned above, these manufacturers buy the chips in bulk from the chip manufacturers.

    Ever notice how these guys don't ever release just a single RX570 or GTX1070?... Also, ever notice that the different variations of the same base GPU have different specs and prices?

    That's because they're using better parts... more specifically, better chips. The lower priced GPU's end up with lower quality chips while the higher priced GPU's end up with the higher quality chips.

    At this point, the marketing departments take over and work their "magic" with number rounding. Since nobody would want to buy a 3.92GB GPU, they just round it up and market it as a 4GB GPU. Basically, they just say "Close enough!" and ship them out.

    "Okay, but if the GPU has 3.92GB and the DAG is only 3.91GB... I should still be good to go, right?"

    • Maybe... it's a bit more complicated than that...

    You still need to account for the GPU's resource overhead. Although it's not a lot, it still exists. A GPU still needs resources even if it's just sitting idle.

    Do you have a monitor connected to this GPU? If so, producing an output to a monitor requires even more resources.

    Most of the mining software's "work" code is also loaded into the GPU's VRAM for faster execution... blah, blah, blah.

    The main point here, is to know that even the basic operation of the GPU uses up resources - resources that could otherwise be put towards mining. Without knowing exactly how much this "total overhead" is, it starts to get really hard to tell how much impact, if any, it may have towards mining.

    However small it may be, when we're dealing with the difference between 3.91GB and 3.92GB, the GPU's overhead could be just large enough to push the GPU over the threshold and cause an earlier than expected/calculated "Eth-Death."

    AMD Ethash Miners: If you'd like to know exactly how much VRAM you have available, fire up your Eth miner and wait for it to start mining, then just run this command:

    card=0; echo "$(($(cat /sys/class/drm/card"$card"/device/mem_info_vram_total) - $(cat /sys/class/drm/card"$card"/device/mem_info_vram_used)))" | numfmt --to iec --format "%8.4f"

    To check any additional GPU's you may have running, simply adjust the card=0 section in the beginning of the command to whatever number was assigned to your actual AMD GPU.
    Note: GPU's are numbered by the system by whole numbers starting with 0. Most onboard video hardware will be given 0, so if the above command doesn't work, try starting with 1.


    If you really want to get specific output... down to the exact byte, just remove the | numfmt --to iec --format "%8.4f" part from the command like so:ca9f8555-f163-4e3f-ae85-fc3b9ec933c0-image.png

    nVidia Ethash Miners: If you'd like to know exactly how much VRAM you have available, fire up your Eth miner and wait for it to start mining, then just run this command: nvidia-smi
    Memory usage and totals are right there in the middle. You can also see exactly how much VRAM is being used by your miners in the bottom section.

    "So I'm stuck and all is lost?!"

    • For Ethash, Yes.

    There is no fix for a GPU that has suffered from "Eth-Death". Unless something changes within the Ethash algorithm in the future (very unlikely), "Eth-Death" is permanent.

    Some of the Ethash mining software developers have been actively trying to delay "Eth-Death" as long as physically possible by optimizing their code. Although these tweaks are really "hit-or-miss" and only work for some, they are worth trying if you're trying to squeeze as much ETH from your GPU's as possible.

    One that seems to have worked for a few is Phoenixminer with the following additional flags: -altinit -rvram 1

    Understand that these are not cures... they're just a delay, at best - attempts at delaying the inevitable.

    • For mining in general, No.

    The "sun may have set" on your GPU's dreams of forever mining Ethash, but it doesn't have to mean the end of mining. There are many other algorithms/coins out there that your "4GB" GPU can still work on.

  • Beta Team

    Thanks for this, Doc. It should be noted that if your GPU has already paid for itself, you can happily mine away at any algorithm whose return exceeds your power cost because it's pure profit. AMD is doing reasonably well on ProgPow, Equihash 192 and 144, Beam III, one or two others; even some of those "ASIC" coins can be profitable on certain AMD cards. Nvidia is doing well at most other GPU mineable algorithms.

    Also, older cards still make good prospectors for speculative mining new or promising coins. Test-nets always need some hash. And there are lots of worthy projects on BOINC or Folding@Home. You can even earn some GridCoin or BiblePay for "donating" your GPU power to some of these projects; and get a tax-deduction at the same time!

  • Support

    Sorta... just don't get too caught up in the whole "now profitability" misconception.

    So many miners get caught up in the "now profitability"... mining only when the coin's current value is worth more than the electric needed to mine it. Only a few number of rare-case miners/farms need to deal with this (i.e. those who must sell/trade their earned coin immediately in order to pay the bills each month).

    For everyone else, profitability can only be calculated at the exact time of sale/trade of the mined coin (current sale/trade value vs. cost of electricity used to mine it). The misconception happens when miners attempt to calculate profit based upon the value of the coin at the exact moment it is being mined.

    Here is the reality: because the crypto's value comes from the market (traders):

    • It is very possible to mine a "now profitable" (worth more than the cost of electric) coin and two months later, sell/trade the coin for an actual loss.
    • Likewise, it is very possible to mine a "now not profitable" (worth less than the cost of electric) coin and two months later, sell/trade the coin for an actual profit.

    This same misconception is what leads some miners into believing that "auto switching" coins/alogs results in higher profits, which just isn't the case (unless, of course, you fall into that small category of miners already mentioned above).

    Bottom Line: Profit can only be calculated at the time of sale/trade of the crypto. Period.

  • Beta Team

    Mine --> HODL

    is a totally valid strategy. :)


Want 10% more hash from your rigs?

We promise to keep your email safe and never spam you.

© 2014-2020 Miner.farm | By Miners, For Miners | Portable Instant Mining Platform, LLC