First, I want to clarify that I am not an expert on the subject of cryptocurrency mining. I just obsess over stupid things because I find it fun and I have no life. Apparently.

That being said I hope you will find this guide informative and helpful.

All tests were performed with the Oscillator JZMRWN prebuilt as the reference point. The default hashrate for this PC in SuperHash averages out to 160 kH/s. The default part list is as follows:

• Hard Drive: SumSag 2TB SSD IMP90 SU2749
  
• GPU: Azuz RS9 80X
  
• CPU: AMP RiceZen-4350 Black Ed. @3.50GHz
  
• MoBo: EGVA GLOP-774-JK-258-ZX
  
• RAM: Tashibo Sport 4x16GB 4333MHz CL19
  
• Case: BeMuted Silent X-808
  
• PSU: EnerMin PRO 1700
  
• Fans: x3 CrossAir MGN120 Levitation Static
  
• Cooler: FreezerMaster LiSquid 240 PRO
  
• Cooler 2: FreezerMaster LiSquid 240 PRO: Radiator
  
• Optical Drive: Azuz 16DTH

Any time a specific part is not mentioned as part of the test description, assume it is the default part from this list.

Crypto-mining in real life is a delicate balance between the cost of your components, power consumption, and the maximum number of hashes your hardware can complete per second. During its peak, miners could earn back the money they sunk into their mining rigs fairly quickly before turning a small profit. However, in order to maximize profit you needed to make your rig as energy efficient as possible. So it was not unusual for miners to run 6 to 8 high-end GPUs that were undervolted.

That concept doesn't apply in Hardware Engineers. You don't pay for the power you use so there isn't really a downside to overclocking your rig and cranking up the watts consumed.

Additionally, the reason GPU mining rigs are so popular is because a GPU can calculate more hashes per second than a CPU. So GPU miners in real life typically don't bother using their CPU to mine. All that matters is to have a CPU with enough cores to run all of your GPUs simultaneously.

That is... kinda emulated in HE? But not exactly.

So what actually matters when building a mining rig in HE?

Well first, lets get the stuff out of the way that kinda doesn't matter so we can forget about it.

• Power Consumption- Doesn't affect hashrate unless you throttle your GPU to 80% power for whatever reason (like if you enjoy pain). There is no drawback for using as much power as possible. But there are no gains for pushing your voltages either other than to make an overclock stable.
  
  
• Case- Doesn't affect actual hashrate. But it does affect how many fans you can use and thus how far you can overclock before hitting temperature limits.
  
  
• Motherboard- Doesn't affect actual hashrate. The only time this matters is if you have mods installed that give you MoBos with more than 3 GPU slots. But in the base game all MoBos have 3 GPU slots. I tried running the Oscillator build with the Egva GLOP-ZX board (best AMP MoBo in the game) swapped out for the OBits IP24 board (worst AMP MoBo in the game). SuperHash did not care at all.
  
  
• RAM- Doesn't matter. I tested different capacities, CAS latencies, and frequencies. There was no difference.
  
  
• Operating System- Doesn't matter. WinDoor 7, 8, and 10 performed identically.

The following are the four 22nm AMP CPUs I compared and their resulting average hashrates:


As you can see, there is a law of diminishing returns at play. The only reason the RiceZen-3350 comes out on top is due to the number of cores being doubled rather than simply increased by 2.

The following are the four 12nm AMP RiceZen-4350 Black Edition CPUs I compared with various frequency clocks and their resulting average hashrates. I attempted to push each CPU to their limit without using any VCore changes:


There appears to be a fairly linear relationship between overall CPU frequency and hashrate with a rough gain of +1kH/s per +100MHz.

Lastly for CPUs, I looked at the lithography of two different processors with otherwise identical stats.


While minimal, there is a slight gain of +0.3kH/s per nm reduction of the processor's lithography.

Alright, on to the GPU section.

For this test I simply borrowed two Azuz RS9 80X from neighboring PCs and added them one at a time to a single machine. The results:


Again, there is a law of diminishing returns at play. If each additional GPU were used at 100% we would see a linear increase of 200% and 300% respectively for x2 and x3 cards. Instead we lose 25% of the second card's potential and 50% off of the third card.

This is very similar to how a real PC has a limited number of PCI-E lanes for the CPU and the Chipset on the motherboard. As such, you usually can't achieve full performance potential when you add multiple GPUs to a rig.

For this next test I overclocked four Azuz RS9 80x to their limit without increasing voltage to the Core or the Memory. I was able to achieve a maximum stable overclock of +130MHz for each.


Normally a GPU has a separate Clock Speed for the Core and the Memory. However, in HE this appears to have been simplified to where each card only has a single overall Frequency Clock that can be altered by adjusting either the Core or the Memory. So in the end what seems to matter most is just pushing the clock speeds as high as possible. Though it should be noted that the Core clock gives a higher percentage multiplier per MHz of Frequency.

This is where things get weird. For this part I took four nearly identical RS9 GPUs with different amounts of HBM memory to see if there was any noticeable differences. I had to alter each card's Core clock from their stock frequency to achieve a uniform result, unfortunately the overclocking program only adjusts by increments of 10MHz, so while the frequencies are as close as I could get them, they aren't exact.


I'm a little sus of these results since I'm still not 100% on how the game assigns Memory clock. But at least in this instance it seems as if we are seeing exponential growth rather than diminishing returns.

Ok, home stretch! Just got to keep breathing...

For this final test I wanted to see if the type of GPU Memory had any effect on the performance. This is the most sketchy of the tests. I paired off four cards with a partner to test and see which memory type was superior. Once again I had to alter Core clocks to make each pair as close to identical as possible. Here are the results:


So there you go. For reasons I do not understand it appears that DDR5X has an edge over the competition when it comes to running cryptocurrency hashes.

Honestly? It's hard to say what the best setup for mining is if you don't have endless money. There are a lot of factors at play.

If you do have the money, get the cheapest MoBo and RAM in the game and then buy 3 DDR5X GPUs. From there, get the biggest PSU you can find and a CPU with the most cores you can afford while still being able to purchase the best cooling system. From there, overclock the crap out of the CPU and GPUs.

And yeah, I was curious and ran a benchmark test on the machines I used for the Motherboard swap. Turns out RAM and the Motherboard do change your score in 3DBench. They just don't have any relevance to bitcoin mining. Which is about as accurate as you can get for an indy game like this.

At any rate, now that I've been at this for over 12 hours, I think I'm going to go lie down.


Happy mining ya mad lads! And if you find this info useful, give me a shoutout!

Oh, and once I've recovered I may try my hand at creating a few mining-specific mod parts. We'll see. First I need to close my eyes for awhile.