Ethereum GPU mining?

[Feld]

^Thanks Elfear!

Oh and in case anyone was wondering, the EthArmbian OS has both geth and parity installed as services, and you can choose which one to run.
I chose Parity to take it a bit easier on the SSD, and also since there are so many geth nodes already.

Awesome! Nice job!

 

[Charlie22911]

Just a heads up, you don’t want to heat sink the NAND; only the controller. NAND actually lasts longer at elevated temperatures, and wears faster at lower temperatures.

 

[IntelUser2000]

Just a heads up, you don’t want to heat sink the NAND; only the controller. NAND actually lasts longer at elevated temperatures, and wears faster at lower temperatures.

Yea. This is also true if you are using NAND SSD as long term storage. Don't keep them too cool. 😛

thilanliyan: Nice setup. Custom mods are really fun. I cut open a DVD writer case and used the aluminum panel as the heatsink for the memory on my XFX RX 470. Makes it run more stable at higher mem clocks. My MSI RX 470 has a very short heatsink I bought from eBay on the RAM chips. That works well too.

There's also a Windows-based alternative like the LattePanda: https://www.dfrobot.com/product-1404.html

 

[DrMrLordX]

I think that's the smallest Ethereum node I've ever seen. Cool.

 

[thilanliyan]

Just a heads up, you don’t want to heat sink the NAND; only the controller. NAND actually lasts longer at elevated temperatures, and wears faster at lower temperatures.

What? Is this true? What kind of temps are we talking?
I didn't put heatsinks directly on the NAND, but the SSD already came with a thin aluminum heatsink, and I just added several more on top, along the length of it (the green coloured ones in the pics). Before the fan and heatsinks it was blazing hot, I couldn't even leave my finger on for more than like 1/2 second.

EDIT: This is an article about temps and NAND I found:
https://www.eeweb.com/profile/eli-t...al-temperature-and-nand-flash-in-ssd-products
Basically says that data retention at low temps is better but actually write/erase is better at higher temps. Not sure what to do lol.

What kind of operating temps should I be shooting for?

 

[thilanliyan]

There's also a Windows-based alternative like the LattePanda: https://www.dfrobot.com/product-1404.html

I did look at the LattePanda but decided to stick with the NanoPC-T4 to keep the wattage down. Having said that, I thought I could run the NanoPC-T4 passively, but I was uncomfortable with the temps. Seeing as I had to add active cooling anyway (although the fans are silent at 5v), it might have been just as effective to use the LattePanda. Linux support is a bit lacking for the LattePanda from I've read though.

 

[thilanliyan]

Looks like there are 6.33TB written already while running the node 😱
On the bright side it's operating at 35C lol 😀

 

[IntelUser2000]

I did look at the LattePanda but decided to stick with the NanoPC-T4 to keep the wattage down.

The Atom x7 on the first LattePanda should be just as power sipping. Since it only needs the 5V 2A connector it shouldn't use more than 10W either. They do have the newer one based on the Geminilake-coming but not yet available. They already have the Core m3 ones which are much more expensive. A review on the m3 version says running at Prime 95 peaks at 19W but steadies down to 12W, probably when Turbo disengages. The newer ones may be better with Linux since the x7 Atoms were really meant for Smartphones and Tablets.

Looks like there are 6.33TB written already while running the node 😱

So how long did it take to finish the sync? Do you think the writes will reduce to a much more manageable level as the sync is finished?

What kind of operating temps should I be shooting for?

If its running blazing hot as you say, then its worth putting a heatsink regardless, since it'll be running all the time.

Here's more info: https://www.anandtech.com/show/9248/the-truth-about-ssd-data-retention

As the table shows, the data retention is proportional to active temperature and inversely proportional to power off temperature, meaning that a higher power off temperature will result in decreased retention.

It is actually different than what I believed. I thought higher temperatures were always better.

 

[thilanliyan]

So how long did it take to finish the sync? Do you think the writes will reduce to a much more manageable level as the sync is finished?

Well, I used the warp mode in Parity which does an initial sync up to the more recent blocks, then syncs the ancient blocks. I have had it running for 3 days and don't think it's finished yet.

What worries me is the "Data Units Written" value in the smart attributes is going up by about 100gb/hour, so the total is near 7TB right now, compared to when I posted that screenshot earlier today.

Anyone have an idea whether that smart value is the actual amount of data written to the drive? Or does it mean something different?

 

[IntelUser2000]

Yea, it should be the actual amount written. Seems similar to my experience.

Tell us if the rate changes drastically once it finishes. It should anyways.

I guess the data written doesn't come solely from downloads. Unless your effective, average internet speed is 240Mbit/s. It would also mean in 3 days you have used 7TB of data. So once it downloads it uses that to shuffle data around basically?

 

[thilanliyan]

Tell us if the rate changes drastically once it finishes. It should anyways.

I guess the data written doesn't come solely from downloads. Unless your effective, average internet speed is 240Mbit/s. It would also mean in 3 days you have used 7TB of data. So once it downloads it uses that to shuffle data around basically?

Yes, will do (the ancient block sync is around 6.1 million out of ~6.5 million). I hope it does lol. I should have thought about importing the chain from another source, maybe would have saved some of the write cycles. This morning it's near 8TB. If it gets to ~15TB and it still has not finished I might just shut it off.

My connection is only 50mbit, and I doubt it has downloaded nearly 8TB of data.

 

[thilanliyan]

Writing on the SSD seems to have settled down around ~9TB after fully synching. I'll keep an eye on it in the coming weeks for any spikes. I think the drive is rated for 300TB of writes.

 

[IntelUser2000]

Writing on the SSD seems to have settled down around ~9TB after fully synching. I'll keep an eye on it in the coming weeks for any spikes. I think the drive is rated for 300TB of writes.

So its done? How long did it take? 4 days? That's like 3 times as fast as geth.

The 480GB Adata SX8200NP has a rated lifespan of 320TBW.

 

[thilanliyan]

So its done? How long did it take? 4 days? That's like 3 times as fast as geth.

The 480GB Adata SX8200NP has a rated lifespan of 320TBW.

Yes, it took about 4 days.
However, it seems to have got stuck with the sync after I rebooted the node. I may have to redo from scratch lol. I'm going to try exporting then importing chain, see if that works.

 

[Charlie22911]

It’s easier to understand NAND retention with relation to temperature if you consider the following:


1. Think of temperature as a measure of kinetic energy (vibrations) in the constituent particles of a system (NAND).


2. NAND is like an array of capacitors (containers that hold electrons) that store electrons. 0 or 1 depends on if electrons are present.


With that in mind, vibrating particles (electrons) are more likely to escape their “container” as they they vibrate harder (increased temperature); this is bad for retention. This also makes them easier to erase however (good for writes, less stress on the “container”).


This is very simplified obviously, and off the top of my head; the concept should be sound however.

 

[thilanliyan]

This is very simplified obviously, and off the top of my head; the concept should be sound however.

Yes, it's easy enough to understand, but I have not been able to figure out what a "good" operating temp is. If I am just storing then yes, low temps are better, but what about a drive in use? You need it to be able to store data but also read/write.

 

[Charlie22911]

Yes, it's easy enough to understand, but I have not been able to figure out what a "good" operating temp is. If I am just storing then yes, low temps are better, but what about a drive in use? You need it to be able to store data but also read/write.

The controller will sink heat into the PCB and warm the NAND naturally. The simple answer is that you shouldn’t have to do anything, I personally wouldn’t even bother heatsinking the SSD; you aren't likely to push the controller hard enough for it to throttle with an SBC.

 

[IntelUser2000]

Yes, it's easy enough to understand, but I have not been able to figure out what a "good" operating temp is. If I am just storing then yes, low temps are better, but what about a drive in use? You need it to be able to store data but also read/write.

Well, in use, little bit higher temps are better. But it shouldn't have a problem doing that.

 

[thilanliyan]

PSA: run the "--db-compaction ssd" flag if running Parity on an SSD as that seems to have helped the actual space used and the amount of writes immensely.

 

[IntelUser2000]

@thilanliyan So give us an update on the node. How is it going? Will you keep it running?

 

[thilanliyan]

@thilanliyan So give us an update on the node. How is it going? Will you keep it running?

Well, it's been running steady as a parity node. I think it's on pace for about 15 TB written per month, but probably will go up I suppose as the blocksize increases. It's currently at 38 TB (about 25 of that was before I had to re-sync) and I think once it gets to about 250TB written I will stop it and maybe put in a bigger drive with better endurance. This kind of storage requirement is not sustainable in the near future unless some new type of fast drive comes out with much better endurance.

 

[Red Squirrel]

Wow you must have a serious storage setup!

Actually how well does the data compress, is it all text? I suppose it would be viable to just do a giant raid 0 with 10TB drives, and then have a backup. Can then keep adding more drives to the array. 10TB drives are roughly $500 so not cheap to build an array big enough even going raid 0 though.

 

[Elfear]

So for you guys who follow the crypto world closely, any updates on what we can expect from Ethereum? PoS any time soon?

 

[thilanliyan]

Wow you must have a serious storage setup!

Actually how well does the data compress, is it all text? I suppose it would be viable to just do a giant raid 0 with 10TB drives, and then have a backup. Can then keep adding more drives to the array. 10TB drives are roughly $500 so not cheap to build an array big enough even going raid 0 though.

Haha sorry I should have clarified that is 38TB written, not 38TB total storage. I'm using a measly 500GB NVME drive lol. This Adata drive I'm using is rated for 350TB written endurance, and I'm at 38 TB. At current write rates the drive would last maybe a year and a half, which is pretty pathetic tbh (from the Eth/parity standpoint). Something needs to be done about the storage requirements for running nodes, otherwise that will also become an expensive arms race.

 

[Red Squirrel]

Ohhhh like wear on the SSD. I thought it was actual data downloaded from the chain so far.

So on a spindle I guess this would not be too much of an issue. I kinda want to do a raid array and try to download the chain again. My NFS storage was too slow but if I do a dedicated local array it will probably be able to keep up I imagine.