Building an efficient rig
I want to open up a discussion so I can learn more about all the different hardware that goes into a Burstcoin mining rig.
I have read a lot about bottle-neck issues when there are many HDD hooked up to one PC. Is there a formula, or way to calculate if a CPU or GPU will be rock-star at meeting dead-lines for all hard-drives on a system?
From what I've read, for bigger set-ups, it sounds worth it to spring for a GPU. What level of GPU would be required to speedily scan ~180+TB? (I am basing this number of having 22 8TB drives connected through USB3 hubs.)
Is system RAM important for BURST mining? If so, how much would a bigger system demand?
What size PSU would be appropriate for a system like this? I am intending to use powered USB hubs, with external hard drives. . . but will there be any extra draw on my MB PSU? Or would a dinky 350w be appropriate?
Getting more excited about BURST everyday, and looking forward to building my first full-sized rig!
ZapbuzZ last edited by ZapbuzZ
I can recommend server rack equipment because the cost of desktop use wouldn't be very profitable considering block difficulty. That would include enterprise class SAS Disks. I can also recommend a minimum disk capacity set of minimum 65TB on NINJA pools and probably over 100 if mining solo. I wouldn't take my opinion for granted i'm just 1 person but consider I stopped mining at 22tb 1 year after starting with 1gb that would be my current target solution.
vaxman last edited by
Have you at least read the preceding posts ?
Your question is so broad, there is lots of info to be found here, you have to read yourself, though.
@zapbuzz Thank you Zap! This is awesome information~ Checking into SAS Disks now~ Do you have any servers dedicated to BURST?
@vaxman Tons of awesome posts on this forum. I've been reading through a lot. But I haven't found specific information on RAM requirements, or how 20+ hard drives effects PSU considerations. External drives power themselves. . . I believe? But does it create any power draw on the MB?
And while there are some posts about achieve adequate DLs, there really isn't much technical information I've read about how to pair a CPU or GPU (in terms of it's performance specs) to a specific #-of-TB array to make sure those DLs are being met in time.
I'm hoping to get a thread together to help people building a dedicated Burst rig understand all the peripheral hardware considerations~
rds last edited by
I have three rigs all work well. One is a $250 laptop with 4 threads, 12GB of ram that reads 90TB of externals connected through 1 USB 3.0 port in 70 seconds. The second is a desktop type with 8 threads 16GB ram, 4 USB 3.0 cards and an AMD 270 GPU. Scans 216TB of externals in 75 seconds CPU miner. Also running background mining on miningpoolhub, CPU and GPU. The third is a 32 thread server 96GB, scans 288TB in 32 seconds CPU mining. Also running background mining on miningpoolhub, CPU.
The takeaway, you don't need a whole lot of memory to mine or plot. You definitely need USB 3.0 or internal SATA/SAS.
Get the most TB for the least money.
ZapbuzZ last edited by ZapbuzZ
I haven't been mining recently due to lack of capacity and power line unreliability. I believe I shall return to burst when conditions are better. I believe that UPS would be a good idea but like many things in my country at the moment the price is staggering. My rig is not a rack server its a full tower server built with legacy gaming hardware. My disks are second hand but you can get reliable performance from those that once plotted the disks only read data without writing. The power and data cables have metal tongues that lock the cables into place to prevent failure from vibration. I have to say that external WD 8TB USB3 MYBOOK drives are a good alternative to SAS disks. Note SAS disks are not a necessity but an optimal choice for massive capacity.
Aweome information RDS and ZapBuzZ. This type of experience really helps me a lot in determining what hardware combos might work.
This is a great video to include on this thread too:
rds, do you mind if I ask you how much your server rig is currently earning? Sounds like a pretty beefy set-up! What type of hard-drives do you have in that rig? Do you run Windows 10 on a virtual server in the machine? How does the software end work mining Burst on a server? (I'm inexperienced with servers, but beginning to learn since it seems useful for Burst~)
I really love the concept behind Burst. I run a small mining company with GPU and ASIC, but I'm getting my investors behind the idea of hard-drive mining, so we can support the burst network in a larger way.
I appreciate your help!
rds last edited by rds
The sever looks just like a desktop platform to me, although it is headless. I run Win 10 pro with Teamviewer or Chrome remote desktop to access it. No virtual machines doing work, but I do run a few on there for development, sandboxing etc. The server runs pool software (CRK code) that all my miners are pointed to. The server earns about 1 block per day. The other 2 machines about 1 block per day. ~2500B/day total. BTW, the server uses ~700W, the desktop unit 430W and the laptop rig, 90 watts. I run background mining for Monero to cover the electric bill, but I actually can get about 2.5x the electric bill running xmr-cryptonight miners on the server and the desktop platform.
@meni-menindorf I run several servers for Burst Mining - 2 x 24 Bay servers, and a server with a 16 Bay external SAS chassis. Look for Supermicro servers, they're traditionally the best priced.
Working with servers is different from PCs, but done properly provide great performance. You can find plenty of dual/quad socket servers on eBay for great prices - just ensure whatever ones you look at have SAS2 backplanes, SAS1 is significantly slower.
For your drives, use Xplotter to create a single drive capacity plot.
For drives, you might want to consider these: https://burstforum.net/topic/8122/more-8tb-drives-from-viixim
One of my servers is running 24 of the above drives - all internal. 173TB Mined in 48 seconds, and has the capability of doing CPU and/or GPU mining.
vaxman last edited by vaxman
@meni-menindorf I run a single server with 3 SAS JBODS, 45 slots each, attached with 2x LSI SAS9202-16e. The HBAs are PCIe 16x and @100$, all cases are Supermicro. These JBODs sell for about 500-900 $, but today none are available on ebay. Must be all these miners upgrading to industrial scale..
I'm strictly CPU-mining, as my OS has no opencl-support.
I get about 90 MB/s with AVX2 per 2.1 GHz core (E5-2650v3 ES).
20 cores (no HT) scan 1.8-1.9 GB/s.
Speed diff SSE4 to AVX2 about 2x (dcct, mddct).
One might translate that to scanning 185 GB of plotfile per second per GHz per core using AVX2 or half that with SSE4 on up to 4 year old intel CPUs.
If you have to buy and can't rely on existing hardware AND have lots of plotted space, jminer (opencl on GPU) is most likely the most economic path.
I'm running freebsd with zfs, which is rock stable - mining uninterrupted and wihout any manual intervention for months, reboots are because of kernel updates only.
If you are going big (100+ disks), I'd suggest plotting w/GPU to a stage disk (array), and optimize from there to the target disk/stripe/raidset. The stage area should be more than a single disk, as a recent GPU easily pushes 350+MB/s. Having two stage-arrays allows for parallel plotting and optimizing.
Using ZFS or any copy-on-write filesystem on the target allows shingled (SMR) disks to keep their maximum possible write-throughput. The gain here is they are cheaper than PMR disks and need 0.5-0.7 of the energy of faster spinning PMRs.
Memory footprint for mining is minimal (~3 GiB for me, for 400+ TiB, 20 threads), but you can configure read-ahead/buffering at compile time (dcct, mdcct). You have to have enough PCIe lanes for HBAs and possibly a GPU, so a dual-socket E5 or an Epyc is needed here.
Memory footprint for plotting is up to your design and depends on many factors (I use gpuPlotGenerator in buffered mode, 4 GPUs, utilizing 16 GiB main memory, for 8 GiB chunks, stagger 32768). YMMV a lot.
If you're scaling to 100+ disks, keep in mind that (even new) disks will fail. One failure shouldn't take down half your rig. And you need many logical spindles to feed 1, 2, 3 ? Gbytes/second into the miner, running on 10+ CPU- or 1000+ GPU -cores, so you need to read many files in parallel. A single filesystem would'nt cope here.
There have been 500 TB builds with USB3 here;
So even this is possible. I have no idea how stable this is, though. As I want it maintenance-free, I went the JBOD route.