Building Cheap Miners : My "Secret"

[PharmEcis]

I will caveat that the box I had running w/ 4 1080s did use 2 PSUs.  I was using risers but but all cards were powered from internal w/ no external PSU and breakout board.  (I raped the SAS connector and used 12v to 5v step downs).

However, due to heat, I have changed that config to pull all cards out of the box along w/ the 2 that weren't in it so that all GPUs are on breakout boards and external supplies.  I never checked power draw though.

[1713277242]

1713277242

[1713277242]

1713277242

[1713277242]

1713277242

[senseless]

It's out now, so.. This is my real secret for mining and why I never scaled cpu hardware... You can buy these and replace your gpus with them. They'll work in risers... I'll never publicly release firmwares like he's doing, but if he's truly going to support it and release his firmwares... This is where the mining is at...

https://bitcointalk.org/index.php?topic=3459858.0

Be careful, it is possible for someone to create a firmware which would cause the FPGA to be physically destroyed and possibly take the host PC with it.

[MinersRus]

It's out now, so.. This is my real secret for mining and why I never scaled cpu hardware... You can buy these and replace your gpus with them. They'll work in risers... I'll never publicly release firmwares like he's doing, but if he's truly going to support it and release his firmwares... This is where the mining is at...

https://bitcointalk.org/index.php?topic=3459858.0

Be careful, it is possible for someone to create a firmware which would cause the FPGA to be physically destroyed and possibly take the host PC with it.

Funny how you posted this in the "Building Cheap Miners" thread.

"Each VCU1525 card costs $4000, or $32K for the whole rig." they are anything but cheap.

[PharmEcis]

It's out now, so.. This is my real secret for mining and why I never scaled cpu hardware... You can buy these and replace your gpus with them. They'll work in risers... I'll never publicly release firmwares like he's doing, but if he's truly going to support it and release his firmwares... This is where the mining is at...

https://bitcointalk.org/index.php?topic=3459858.0

Be careful, it is possible for someone to create a firmware which would cause the FPGA to be physically destroyed and possibly take the host PC with it.

Funny how you posted this in the "Building Cheap Miners" thread.

"Each VCU1525 card costs $4000, or $32K for the whole rig." they are anything but cheap.

While true, they are the future and I'm speaking with someone right now about investing $50k with me so I can load a bunch up in my farm.

With the new tech coming, everything we have done in this thread is going to be obsolete.  My electricity bill was $1,000 last month.  If I sold all my hardware and only had 8 of those cards I would product 5x as much for 1/10 of the power draw, a much smaller thermal envelope to deal with and a much higher return after ROI.  Efficiency is the name of the game and eventually being cheap just shoots yourself in the foot.

[senseless]

While true, they are the future and I'm speaking with someone right now about investing $50k with me so I can load a bunch up in my farm.

With the new tech coming, everything we have done in this thread is going to be obsolete.  My electricity bill was $1,000 last month.  If I sold all my hardware and only had 8 of those cards I would product 5x as much for 1/10 of the power draw, a much smaller thermal envelope to deal with and a much higher return after ROI.  Efficiency is the name of the game and eventually being cheap just shoots yourself in the foot.

Toss it in my pile. I'll be working with next gen chips while everyone is fighting over the 9p. I'm currently raising with a $1M target.

Funny how you posted this in the "Building Cheap Miners" thread.

"Each VCU1525 card costs $4000, or $32K for the whole rig." they are anything but cheap.

I'm not really sure this thread is about cheap miners other than the physical CPU / host machine and being able to get the lowest cost, most pcie ports, enough power supply. You have people in this thread buying 1080s which are $800 cards. They're paying $800 to generate $5/day of revenue. If they buy 5 they've spent $4,000 to generate $25/day of revenue. Or, they could spend $4,000 use 1/10th the amount of power and generate $40+/day. There were periods during the last bull run where it was possible to make $100/day per fpga. In terms of ROI and efficiency increases, these are far superior than any GPU on the market. GPUs will be completely useless for mining in 2019.

[MinersRus]

It's out now, so.. This is my real secret for mining and why I never scaled cpu hardware... You can buy these and replace your gpus with them. They'll work in risers... I'll never publicly release firmwares like he's doing, but if he's truly going to support it and release his firmwares... This is where the mining is at...

https://bitcointalk.org/index.php?topic=3459858.0

Be careful, it is possible for someone to create a firmware which would cause the FPGA to be physically destroyed and possibly take the host PC with it.

Funny how you posted this in the "Building Cheap Miners" thread.

"Each VCU1525 card costs $4000, or $32K for the whole rig." they are anything but cheap.

While true, they are the future and I'm speaking with someone right now about investing $50k with me so I can load a bunch up in my farm.

With the new tech coming, everything we have done in this thread is going to be obsolete.  My electricity bill was $1,000 last month.  If I sold all my hardware and only had 8 of those cards I would product 5x as much for 1/10 of the power draw, a much smaller thermal envelope to deal with and a much higher return after ROI.  Efficiency is the name of the game and eventually being cheap just shoots yourself in the foot.

With the new tech coming, everything we have done in this thread is going to be obsolete.

Not at that price. Now if the price drops to $1000 and is universally available then that would be a different story and would be welcomed by the community.

I hope you do realize that Monero will hard fork again in about 5 months. The Monero developers were actively discussing ways to brick even FPGA Monero miners before the last hard fork. With the short time they had they focused on only killing the ASIC's which they did. Now with 6 months passing from April to the next hard fork they will have a much better handle on how to stop FPGA miners.

[MinersRus]

While true, they are the future and I'm speaking with someone right now about investing $50k with me so I can load a bunch up in my farm.

With the new tech coming, everything we have done in this thread is going to be obsolete.  My electricity bill was $1,000 last month.  If I sold all my hardware and only had 8 of those cards I would product 5x as much for 1/10 of the power draw, a much smaller thermal envelope to deal with and a much higher return after ROI.  Efficiency is the name of the game and eventually being cheap just shoots yourself in the foot.

Toss it in my pile. I'll be working with next gen chips while everyone is fighting over the 9p. I'm currently raising with a $1M target.

Funny how you posted this in the "Building Cheap Miners" thread.

"Each VCU1525 card costs $4000, or $32K for the whole rig." they are anything but cheap.

I'm not really sure this thread is about cheap miners other than the physical CPU / host machine and being able to get the lowest cost, most pcie ports, enough power supply.

It is for me. I found this thread via Google Search and have used the knowledge to put together a mining farm at lower cost.

You have people in this thread buying 1080s which are $800 cards. They're paying $800 to generate $5/day of revenue. If they buy 5 they've spent $4,000 to generate $20/day of revenue. Or, they could spend $4,000 use 1/10th the amount of power and generate $40+/day. There were periods during the last bull run where it was possible to make $100/day per fpga. In terms of ROI and efficiency increases, these are far superior than any GPU on the market. GPUs will be completely useless for mining in 2019.

Yes some people here do have very large budgets but their ideas like using the HP Z400, or the HP DL580 G7's with Xeon E7-8837's, or the Dell R815's with AMD 6200/6300 Opterons are very good ideas in how to build a farm at a much lower cost. My primary GPUs that I mine with are GTX 750's (Nvidia Maxwell) that I buy for less than $40.

GPUs will be completely useless for mining in 2019.

I cringe every time I see statements like that. GPUs will progress this year with Nvidia's new release and AMD will have a new solution next year. Both will do better so that is a moving target.

[senseless]

Both will do better so that is a moving target.

The MINIMUM efficiency increase over GPUs I've ever had was 3x (and that was for a completely unoptimized hastily written code). The largest I've had so far? Around 100x. Average seemed to be in the 30-50x range. They offer a greater level of configuration that is not possible on GPUs. It'll be a good time soon to short nvidia and amd stock. 2019 is going to be a bad year for both of them. Not only are they going to lose mining market share to FPGA; they will also be losing scientific (uni, govt, corporate) research into AI, genomics, etc. Xilinx and Altera both are starting to reverse their archaic sales practices and mentality. They've realized they're losing out on a market segment that is multiples larger than the ones they've historically sold to. Intel with altera is planning on putting FPGA chips directly on the motherboard and having a fiberoptic path on the motherboard between the FPGA and CPU. The bus width will be a high multiple of the PCI-E bus speed. Oh ya, and people can compile their existing opencl gpu code to the fpga. This allows them to get up and running quickly with their existing code until they migrate over to a full RTL design. Welcome to the future.

My trade, what i've done my entire life, has primarily been sys admin, network admin, coder, etc. Once someone offloads apache, mysql, nginx, etc processing to a FPGA co-processor... It will allow applications that had historically needed huge clusters to operate; to operate with a fraction of those servers and offloading tasks to the FPGA. Power consumption for application processing will drop by 10x or more. (with the exception of data storage servers and clusters). FPGA are also significantly more stable than GPUs. The only thing anyone should really be using a gpu is for video, and graphic acceleration. Ya, they can do other stuff, but they're not really that suited to it. RTL development for fpga coprocessing is going to be a big business in the future (5-10 years).

[greyday]

Both will do better so that is a moving target.

The MINIMUM efficiency increase over GPUs I've ever had was 3x (and that was for a completely unoptimized hastily written code). The largest I've had so far? Around 100x. Average seemed to be in the 30-50x range. They offer a greater level of configuration that is not possible on GPUs. It'll be a good time soon to short nvidia and amd stock. 2019 is going to be a bad year for both of them. Not only are they going to lose mining market share to FPGA; they will also be losing scientific (uni, govt, corporate) research into AI, genomics, etc. Xilinx and Altera both are starting to reverse their archaic sales practices and mentality. They've realized they're losing out on a market segment that is multiples larger than the ones they've historically sold to. Intel with altera is planning on putting FPGA chips directly on the motherboard and having a fiberoptic path on the motherboard between the FPGA and CPU. The bus width will be a high multiple of the PCI-E bus speed. Oh ya, and people can compile their existing opencl gpu code to the fpga. This allows them to get up and running quickly with their existing code until they migrate over to a full RTL design. Welcome to the future.

My trade, what i've done my entire life, has primarily been sys admin, network admin, coder, etc. Once someone offloads apache, mysql, nginx, etc processing to a FPGA co-processor... It will allow applications that had historically needed huge clusters to operate; to operate with a fraction of those servers and offloading tasks to the FPGA. Power consumption for application processing will drop by 10x or more. (with the exception of data storage servers and clusters). FPGA are also significantly more stable than GPUs. The only thing anyone should really be using a gpu is for video, and graphic acceleration. Ya, they can do other stuff, but they're not really that suited to it. RTL development for fpga coprocessing is going to be a big business in the future (5-10 years).

I don't necessarily disagree with this forecast, but Nvidia and AMD are big companies with a huge growth in income the past few years. They've already announced mining-focused gpus (which are, granted, just slightly modified versions of their higher mid range cards), but I sincerely doubt at least one of them doesn't have a plan to compete, whether that's entering the AI market or shifting to/adding a line of FPGAs to straight up releasing ASICs themselves. I wouldn't count wither of them out from a business standpoint (though if they don't adapt their model, I 100% agree).

I own a small amount of stock in both, by the way. Just for full disclosure.

[Surfer123]

Hey Guys,


Huge fan of this thread. I'm building a rig with a HP XW4600 and I have a question about the PSU:

I'm looking at running a 4 RX570 rig and I'm wondering do I need a PSU with 8 PCIe ports to power the RX570's and risers? Or is it possible to power the RX570 and riser from the same PCIe cable? Or possible another way of doing it?

[sundownz]

Both will do better so that is a moving target.

The MINIMUM efficiency increase over GPUs I've ever had was 3x (and that was for a completely unoptimized hastily written code). The largest I've had so far? Around 100x. Average seemed to be in the 30-50x range. They offer a greater level of configuration that is not possible on GPUs. It'll be a good time soon to short nvidia and amd stock. 2019 is going to be a bad year for both of them. Not only are they going to lose mining market share to FPGA; they will also be losing scientific (uni, govt, corporate) research into AI, genomics, etc. Xilinx and Altera both are starting to reverse their archaic sales practices and mentality. They've realized they're losing out on a market segment that is multiples larger than the ones they've historically sold to. Intel with altera is planning on putting FPGA chips directly on the motherboard and having a fiberoptic path on the motherboard between the FPGA and CPU. The bus width will be a high multiple of the PCI-E bus speed. Oh ya, and people can compile their existing opencl gpu code to the fpga. This allows them to get up and running quickly with their existing code until they migrate over to a full RTL design. Welcome to the future.

My trade, what i've done my entire life, has primarily been sys admin, network admin, coder, etc. Once someone offloads apache, mysql, nginx, etc processing to a FPGA co-processor... It will allow applications that had historically needed huge clusters to operate; to operate with a fraction of those servers and offloading tasks to the FPGA. Power consumption for application processing will drop by 10x or more. (with the exception of data storage servers and clusters). FPGA are also significantly more stable than GPUs. The only thing anyone should really be using a gpu is for video, and graphic acceleration. Ya, they can do other stuff, but they're not really that suited to it. RTL development for fpga coprocessing is going to be a big business in the future (5-10 years).

I'm quite interested in this if you decide you want to sell your software.

[senseless]

I'm quite interested in this if you decide you want to sell your software.

I've reached out to you a couple of times thinking you might have some interest in being pitched. But I don't plan on releasing any firmwares or selling devices.

[zer0k]

Tried reading through the whole thread...but it's really long

Has anyone found any "cheap" 4U+ rack mount systems that can run 4-7 cards internally?

[dhouse]

Yes some people here do have very large budgets but their ideas like using the HP Z400, or the HP DL580 G7's with Xeon E7-8837's, or the Dell R815's with AMD 6200/6300 Opterons are very good ideas in how to build a farm at a much lower cost. My primary GPUs that I mine with are GTX 750's (Nvidia Maxwell) that I buy for less than $40.

Whoa how are 750's profitable? I guess at that low price it would make sense, but seems like it would take forever for ROI just the computer costs.

edit: don't mean to challenge or disagree, i am actually very intrigued by low cost mining options and just curious how you make it profitable.

[sundownz]

I'm quite interested in this if you decide you want to sell your software.

I've reached out to you a couple of times thinking you might have some interest in being pitched. But I don't plan on releasing any firmwares or selling devices.

Damn... must be in my PMs... I almost never look there. Will check it =)

[MinersRus]

Tried reading through the whole thread...but it's really long

Has anyone found any "cheap" 4U+ rack mount systems that can run 4-7 cards internally?

Not sure what your idea of cheap is but my idea of cheap is the following:

I would recommend the HP DL580 G7 with the PCIe expansion board installed. Make sure you get it with the PCIe expansion board since that adds six more PCIe slots two of which are x16 slots. Two x16 slots are on the main board. I was lucky enough to purchase one for $330 and the other four at $487.

One of these systems currently has four double slot sized GPUs installed inside. These systems can take full size GPUs.

Purchase the least expensive HP DL580 G7 and replace the processors with four Xeon E7-8837's ($44 on eBay). With these processors you can get about 1560 H/s mining Monero on the CPUs alone.

To power the GPUs you will need some of these cables as the power connectors in the DL580 G7 are proprietary.

See this post: https://bitcointalk.org/index.php?topic=1955358.msg33071070#msg33071070

---------------------------------

I recommend that you go back to page 31 and start reading from there as much more details about the HP DL580 G7 are provided.

---------------------------------

EDIT: 6/2/2018

Be sure to first update the BIOS to the latest version before upgrading the processors to the E7-8837's

Here is the latest BIOS for the DL580 G7 2018.02.22(A)(22 Mar 2018)
https://support.hpe.com/hpsc/swd/public/detail?sp4ts.oid=4142793&swItemId=MTX_0b8f3842a4cd4835b6dd193ee9&swEnvOid=4184

[MinersRus]

I had to do the following:

1) Type "sudo sysctl -w vm.nr_hugepages=128"
2) Edit "/etc/security/limits.conf"
3) Add "* soft memlock 262144" and "* hard memlock 262144"
4) Change in XMR-STAK "config.txt" to "use_slow_memory : never"

This resolved the issue.

So... it seems I have to do #1 every time I re-boot as well now.

The other steps are saved but it seems to forget #1 upon re-boot.

SOLVED:

Replace Step #1 as follows:

1) In console go to "/etc" -- type "sudo gedit sysctl.conf" -- add "vm.nr_hugepages=128"

This saves it permanently.

So... not a huge fan of Linux over here, haha... such a PITA.

Thank you for this. I made sure that these steps were done for my R815 Ubuntu miners.

[MinersRus]

Yes some people here do have very large budgets but their ideas like using the HP Z400, or the HP DL580 G7's with Xeon E7-8837's, or the Dell R815's with AMD 6200/6300 Opterons are very good ideas in how to build a farm at a much lower cost. My primary GPUs that I mine with are GTX 750's (Nvidia Maxwell) that I buy for less than $40.

Whoa how are 750's profitable? I guess at that low price it would make sense, but seems like it would take forever for ROI just the computer costs.

edit: don't mean to challenge or disagree, i am actually very intrigued by low cost mining options and just curious how you make it profitable.

I have 37 GTX 750's installed in my various systems. On Monero using XMR-Stak they hash 230 H/s on average. Low of 220 H/s high of 250 H/s if they can overclock.

That gives me 8510 H/s for just those 37 GTX 750's.

They use about 38 watts each while mining. Power usage is 1406 watts internally. With 85% efficient power supplies that means 1654 watts at the wall.

Monero Mining Profitability for the above cards
https://www.cryptocompare.com/mining/calculator/xmr?HashingPower=8510&HashingUnit=H%2Fs&PowerConsumption=1654&CostPerkWh=0.071&MiningPoolFee=1

Yearly I will mine about 24 XMR with a profit of about $4500 after power costs on those 37 GTX 750's.

Also realize that the systems I purchased, like the HP DL580 G7's that have 11 PCIe slots, are also mining on the processors which adds 1560 H/s to each system.


-------------------------------


If you just look at a single GTX 750 purchased at $38.50 (current eBay price) and use these numbers:

220 H/s - lowest that they do
38 watts = 44.7 watts at the wall
7.1 cents per KWh - my rate
No mining fee - I set donation level to 0.0% in XMR-Stak
1% Pool Fee

https://www.cryptocompare.com/mining/calculator/xmr?HashingPower=220&HashingUnit=H%2Fs&PowerConsumption=44.7&CostPerkWh=0.071&MiningPoolFee=1

Shows a profit of $115 yearly so at $38.50 each one is paid for in four months.

-------------------------------

On the subject of ROI on computer costs here is a breakdown on the three HP DL580 G7's with 20 GTX 750's running in them.

Costs: HP DL580 G7 one at $330, two at $487 for a total of $1304. Twenty GTX 750's at an average cost of $40 totals an additional $800 for a Grand Total of $2104.

There are some other miscellaneous costs (Mining rack, 60GB SSD drives, Power Cables, Xeon E7-8837's) but these will be offset by selling the extra 1200 Watt Power supplies, 192 GB of PC3-10600R Memory, Xeon Processors and Memory Cartridges that came with the DL580's that are not needed. Selling these will actually lower my overall cost.

These three systems are on the Free HiveOS mining Monero at a rate of 9.34 KH/s using 2.49 KW at the wall.

https://www.cryptocompare.com/mining/calculator/xmr?HashingPower=9340&HashingUnit=H%2Fs&PowerConsumption=2490&CostPerkWh=0.071&MiningPoolFee=1

That results in a profit of $4631 yearly so these systems will be paid for in 5.5 months. I have been mining with them for about a month so 4.5 more months to pure profit.

[greyday]

Damn, wish I'd gone that route. Energy wise that is the second most energy efficient monero stat I've seen (affordably, anyway).

[PharmEcis]

The only problem you will have is that a lot of the things you want to sell, won't.  I've had cartridges on Ebay for months now at the CHEAPEST price of all listed, and I've sold ZERO.