Optimize Your GPU for Ethereum Mining
Ethereum Mining

Optimize Your GPU for Ethereum Mining

Ethereum Mining

Ethereum GPU mining is again and  — at the moment, at the least — extremely worthwhile. However there’s extra to it than simply firing up the software program and letting it run within the background, particularly for those who’ve managed to acquire one of many greatest graphics playing cards. A lot of the graphics playing cards in our GPU benchmarks hierarchy can theoretically earn cash proper now, relying on how a lot you pay for energy. Nevertheless, you will wish to tune your graphics card with the optimum settings, and the model and card mannequin can have a huge impact on general efficiency and effectivity. Our outcomes right here have been used to create our greatest mining GPUs, for those who’re occupied with extra protection.


First, let’s word that we’re not making an attempt to actively encourage anybody to begin a mining farm with GPUs. If you wish to know learn how to mine Ethereum, we cowl that elsewhere, however the “how” is sort of totally different from the “why.” In actual fact, primarily based on previous private expertise that a few of us have operating shopper graphics playing cards 24/7, it’s completely doable to burn out the followers, VRMs, or different parts in your card.

On the identical time, we all know there’s lots of curiosity within the subject, and we wished to shed some gentle on the precise energy consumption — measured utilizing our Powenetics gear — that the assorted GPUs use, in addition to the real-world hashing charges we achieved. In the event you’ve pulled up information utilizing a mining profitability calculator, our figures point out there is a lot of variation between energy and hash charges, relying in your settings. Do not be shocked in case your explicit card does not attain the extent of efficiency others are displaying.

There’s additionally the previous saying: If you cannot beat ’em, be part of ’em. A part of what drives up the worth on cryptocurrency is that if there’s demand for it that may’t be met. Proper now, you can doubtlessly make a number of hundred {dollars} a month, per GPU. Certain, the worth might go up and also you would possibly miss out. It might additionally drop.

And if extra individuals mine, the rewards for mining change into much less, which suggests possibly the big mining farms will cease snapping up all of the graphics playing cards (and gaming laptops!) in pursuit of Ethereum. Do as you’ll, however do not assume that us discussing it will actually make the scenario any worse. The one manner this will get higher is that if profitability of mining plummets, and the extra people who mine of their spare GPU cycles, the much less there’s for the mining farms.

We’ll begin with the newest era of AMD and Nvidia GPUs, together with the latest RTX 3060 12GB and RX 6700 XT, however we even have outcomes for many earlier era GPUs under. There are some things it is best to know earlier than getting began. First, Ethereum GPU mining requires greater than 4GB of VRAM, so for those who’re nonetheless hanging on to an RX 570 4GB, it will not work.

Second, there are lots of totally different software program packages for mining, however we’re taking the straightforward route and utilizing NiceHash Miner It consists of assist for the most well-liked mining options, and it’ll even benchmark your card to find out which one works greatest.

Besides, it is not at all times correct, so we have standardized on NBMiner for the Nvidia GPUs and PhoenixMiner on the AMD GPUs. (Our take a look at outcomes are typically inside a number of p.c of one another on both miner, although T-Rex or Phoenix miner might do a bit higher on Nvidia and AMD, respectively. YMMV.)

Nvidia Ampere and AMD RDNA2 Mining Efficiency


We’ve used our commonplace GPU testbed (listed to the precise) for these assessments, operating a single GPU. This is not a completely optimum miner PC configuration, nevertheless it’s seemingly near what most of our readers are utilizing.

You do not want a high-end CPU, motherboard, or reminiscence for mining functions, and lots of bigger installations will use Pentium CPUs and B360 chipset motherboards with extra PCIe slots. Crucial components for a mining PC are energy and cooling, as they each immediately affect general profitability. In the event you can maintain your GPU and different parts cool, they’re going to last more and never break down as usually. In the meantime, energy might be very costly for bigger mining setups, and poor effectivity PSUs (energy provide items) will generate extra warmth and use extra electrical energy.

We have run these preliminary benchmarks utilizing NiceHash Miner, taking a look at precise realtime hash charges somewhat than the outcomes of its built-in benchmark. We examined every graphics card in inventory mode, after which we additionally tried to tune efficiency to enhance general effectivity — and ideally maintain temperatures and fan speeds at cheap ranges. We let the mining run for at the least quarter-hour earlier than checking efficiency, energy, and so on., as usually issues will decelerate as soon as the graphics card begins to warmth up.


It is also vital to notice that we’re reporting uncooked graphics card energy for your complete card, however we do not account for the ability consumption of the remainder of the PC or energy provide inefficiencies. Utilizing an 80 Plus Platinum PSU, we ought to be operating at round 92% effectivity, and wall outlet energy consumption is usually about 50-80W increased than what we present within the charts. About 40W of energy goes to the CPU, motherboard, and different parts, whereas the rest depends upon how a lot energy the GPU makes use of, together with PSU inefficiencies.

There’s rather a lot to debate with these charts, particularly, what will we imply by “tuned” efficiency? The reply: It varies, usually massively, by GPU. (After the preliminary article, we have up to date and added extra GPUs, however we have skipped the “inventory” testing and solely included our tuned outcomes.)

Let’s speak in regards to the massive image shortly earlier than we get into the small print. The quickest GPUs for Ethereum mining proper now are the RTX 3080 and RTX 3090, by fairly a big margin. Our baseline RTX 3080 FE measurement obtained 85MH/s, and the baseline 3090 FE obtained 105MH/s. Further tuning improved the 3080 efficiency to 93MH/s, whereas the 3090 FE restricted us (on reminiscence temps) to round 106MH/s. It is vital to notice that each the 3080 and 3090 Founders Version playing cards run extremely popular on the GDDR6X, which limits efficiency. Modding these with higher thermal pads, or shopping for a 3rd get together card, can enhance efficiency and decrease reminiscence temperatures.

In the meantime, the RTX 3060 Ti and 3070 playing cards began at near 52MH/s — despite the fact that the 3070 is theoretically quicker. That is as a result of Ethereum hashing relies upon fairly closely on reminiscence bandwidth. Overclocking the VRAM on these GPUs obtained efficiency as much as round 60MH/s. AMD’s RX 6800, 6800 XT, and 6900 XT playing cards began at near 60MH/s, and with tuning, we achieved 65MH/s — there wasn’t a lot distinction between the three AMD GPUs, largely as a result of they’re all utilizing the identical 16GB of 16Gbps GDDR6 reminiscence.

Lastly, the newer RTX 3060 and RX 6700 XT each have 12GB of GDDR6, with a 192-bit reminiscence bus width, successfully chopping bandwidth by 25% relative to the 256-bit playing cards. This in flip limits efficiency to round 47–48 MH/s after tuning. With out the reminiscence overclock, efficiency drops to round 40MH/s.

You’ll be able to try the ability chart, however the general effectivity chart is extra vital than uncooked energy use. Right here, the decrease energy RTX 3060 Ti and 3070 shoot to the highest, after which there is a average step all the way down to the RTX 3060, 3080, RX 6700 XT, RX 6800, and so forth. A lot of the playing cards are fairly shut when it comes to general effectivity for Ethereum mining, although the extra GPU cores on the 6800 XT and 6900 XT ended up dropping effectivity a bit — extra tuning would possibly enhance the outcomes, notably for those who’re prepared to sacrifice a little bit of efficiency to scale back the ability use.

Lastly, now we have temperatures. These are GPU core temperatures, however they’re truly not the vital issue on lots of the playing cards. AMD’s playing cards ran sizzling at inventory settings, however the entire playing cards profit enormously from tuning. Extra importantly, whereas we could not get GDDR6 temperatures on the 3060 Ti and 3070, we did get VRAM temps on the 3080 and 3090 in addition to the AMD playing cards. At inventory, the 3080 and 3090 Founders Version playing cards each hit 108-110C on the GDDR6X, at which level the GPU followers would kick as much as 100% (or almost so).

The playing cards settled in at 106 levels Celsius, with GPU clocks fluctuating fairly a bit. AMD’s RX 6000 playing cards additionally peaked at round 96C on their GDDR6 at inventory, however tuning dropped VRAM temps all the way in which all the way down to round 68-70C. This brings us to the primary space of curiosity.

Tune Your Graphics Card’s Ethereum Mining Efficiency

Let’s begin by noting that each card mannequin is totally different — and even playing cards of the identical mannequin might fluctuate in efficiency traits. For the 3080 and 3090 playing cards with GDDR6X reminiscence, holding that cool is vital. We have seen examples of playing cards (particularly, the EVGA RTX 3090 FTW3) that may run at as much as 125MH/s, whereas the reminiscence sits at round 85C. That is as a result of EVGA seems to have put lots of effort into cooling the reminiscence.

With out altering the playing cards, the Nvidia 3080/3090 Founders Editions let the reminiscence get extremely popular whereas mining, which may dramatically hinder efficiency and/or cut back the cardboard’s lifespan. Let’s take every card in flip.

GeForce RTX 3090 Founders Version: Whereas technically the quickest card for mining that we examined, we actually don’t love the concept of operating this one 24/7 with out {hardware} modifications or critical tuning. At inventory, the followers find yourself operating at 100% to attempt to maintain the GDDR6X under 110C, and that is not good.

For our functions, we tuned the cardboard by dropping the GPU core to the utmost allowed -502MHz, set the reminiscence clock to +250MHz, and put the ability restrict at 77%. That gave us a GDDR6X temperature of 104C, which remains to be increased than we might like, and efficiency remained at round 106MH/s. Energy use additionally dropped to 279W, is sort of good contemplating the hash price.

Alternatively, you may go for broke and max out fan pace, set the ability to 80%, drop the GPU clocks by 250-500MHz, and enhance the VRAM clocks by 750-1000MHz. In the event you do not mod the cardboard to enhance GDDR6X cooling, you will sometimes find yourself at 106-110C (relying in your card, case, cooling, and different components), and the maxed out fan pace is not going to be good for fan longevity. Take into account your self warned.

GeForce RTX 3080 Founders Version: Tuning this card was similar to the 3090 FE. It does not like inventory settings, because the GDDR6X will get very toasty. We once more dropped the GPU core the utmost allowed (-502MHz), set the reminiscence clock to +750MHz, and put the ability restrict at 70%. That resulted in the identical GDDR6X temperature of 104C because the 3090 FE, and efficiency was solely barely slower at 93-94MH/s.

Once more, maxing out fan speeds and reminiscence clocks whereas dropping the GPU core clocks and energy restrict are key to bettering general hash charges. Modding the cardboard and changing the VRAM thermal pads with thicker/higher pads is feasible and can assist cooling and efficiency. We would desire utilizing an RTX 3080 with higher GDDR6X cooling, nevertheless. Which brings us to…

Colourful RTX 3080 Vulcan: We have eliminated this card from the charts to keep away from having an excessive amount of litter, nevertheless it’s an instance of a 3080 mannequin with higher VRAM cooling than Nvidia’s reference card, so the reminiscence did not get fairly as sizzling.

Nevertheless, we nonetheless discovered we achieved one of the best end result by dropping the ability restrict to 80-90% after which setting the GPU core clocks to the minimal doable worth in MSI Afterburner (-502MHz). Then we overclocked the reminiscence by 750MHz base clock, which gave a last pace of 20Gbps (the Ampere playing cards appear to run at 0.5Gbps under their rated reminiscence pace when mining by default).

That yielded common GPU clocks of simply 1425MHz (down from 1920MHz), however general efficiency elevated to 93MH/s, whereas fan pace, GPU temperature, and energy consumption all dropped. Most significantly, on the identical (comparatively) efficiency because the 3080 FE, GDDR6X temperatures stabilized at 100C. It is not very best, however at these temperatures a 4C distinction might be vital.

The RTX 3070 and RTX 3060 Ti have the identical 8GB of 14Gbps GDDR6, and as we’ll see with the AMD GPUs, that seems to be the limiting issue. Our preliminary outcomes had been poor, as these had been the primary playing cards we examined, however we have revisited the settings after trying on the RX 6000 collection.

GeForce RTX 3070 Founders Version: The principle factor for bettering efficiency on the 3070 was to spice up the GDDR6 clock. We had been ready so as to add 1200MHz, giving a 16.4Gbps efficient pace in principle, however the reminiscence truly ran at 16Gbps (vs. 13.6Gbps at inventory settings). Together with the enhance to reminiscence clocks, we dropped the GPU clock all the way down to the utmost -502MHz in Afterburner and set the ability restrict at 55%. That resulted in precise GPU clocks of 960MHz on common.

You’d assume that would not be adequate, however boosting the GPU clocks as much as 1.9GHz resulted in the identical efficiency whereas considerably rising the quantity of energy used. On the 55%, in the meantime, the 3070 is second solely to its little sibling in general effectivity.

GeForce RTX 3060 Ti Founders Version: As with the 3070, we bumped the reminiscence pace up as the primary change to enhance efficiency. We went from inventory to +1200MHz (in comparison with +750, which is about so far as we might go for gaming overclocks), giving a most pace of 16.4Gbps, with a 400MHz detrimental offset for GPU compute workloads. Different settings had been related: -502MHz GPU clock, 55% energy restrict, and 50% fan pace. Efficiency was very near the 3070 whereas utilizing much less energy, making this the general winner in effectivity.

Asus RTX 3060 Ti TUF Gaming OC: Once more, we eliminated this from the up to date charts, however not like the 3080 and 3090, third get together playing cards weren’t markedly totally different in hashing efficiency with the 3060 Ti and 3070 GPUs. Our tuned settings ended up with increased clocks (because of the manufacturing facility overclock) and extra energy use than the 3060 Ti Founders Version, however mainly the identical hashing efficiency. Optimum effectivity can fluctuate a bit, however for the GA104-based playing cards, GDDR6 pace is the limiting issue on efficiency.

EVGA RTX 3060 12GB: There isn’t any official reference card for the brand new GA106-based RTX 3060, so we’re utilizing the EVGA card that we utilized in our RTX 3060 12GB assessment. We set the ability restrict to 75%, elevated the GDDR6 clocks by 1250MHz, and ended up with a comparatively excessive fan pace of round 80% with the default fan curve. (The cooling on this card is not almost as strong as lots of the different GPUs.) We did not file ‘inventory’ efficiency, nevertheless it was round 41MH/s.

(Picture credit score: AMD)

Radeon RX 6900 XT (Reference): Tuning all three of AMD’s reference RX 6000 playing cards ended up very related. The GPU clocks can go very excessive at inventory, however the reminiscence bandwidth seems to be the primary bottleneck. Working with GPU clocks of two.2-2.5GHz simply wastes energy and generates warmth with out bettering efficiency.

We cranked the ability restrict to the utmost 115% simply to make sure the VRAM wasn’t being held again, then set the reminiscence at +150MHz (the utmost allowed in Radeon Settings), enabled quick RAM timings, and dropped the utmost GPU clock all the way down to 70%. That gave us last clocks of 1747MHz in comparison with 2289MHz at inventory and about 8% increased hash charges general. Extra importantly, energy consumption took an enormous dive, and effectivity improved to one of many higher leads to our testing. However this truly is not AMD’s greatest general displaying.

Radeon RX 6800 XT (Reference): Identical method as above, however attributable to variations within the core configuration and … actually, we’re undecided what the deal is, however we ended up with an optimum most GPU frequency setting of solely 50% this time, which gave us clocks of 1206MHz as a substitute of 2434MHz — and efficiency nonetheless went up, matching the RX 6900 XT and RX 6800. On the identical time, energy necessities dropped considerably, from 281W to 186W.

No matter is happening behind the scenes, it seems totally different AMD Navi 21 GPUs run optimally at totally different “most frequency” settings. Like Nvidia, AMD’s GPUs are largely restricted in efficiency by their reminiscence pace, and with out instruments to overclock past 17.2Gbps, there’s not a lot to be carried out.

Radeon RX 6800 (Reference): With solely 60 CUs (in comparison with 72 on the 6800 XT and 80 on the 6900 XT), you would possibly anticipate the 6800 vanilla card to finish up slower. Nevertheless, the reminiscence proves the deciding issue as soon as once more. We set the GPU energy restrict on the identical 115%, which does make a distinction, oddly — common energy dropped about 15W if we set it to 100%, despite the fact that the cardboard was operating properly under the official 250W TGP.

We additionally maxed out the reminiscence slider at +150MHz (17.2Gbps efficient), and this time achieved optimum efficiency with the GPU set to 75% on most clocks. That resulted in a 1747MHz clock in comparison with 2289MHz at inventory, however fan pace was increased this time. That is as a result of we set the fan to run at 40% at 50C, 60% at 60C, 80% at 70C, and 100% at 80C — and it ended up at 50% pace, which is probably greater than is required, however we really feel it is higher secure than sorry for those who’re taking a look at 24/7 mining.

Radeon RX 6700 XT (Reference): AMD cuts down to only 40 CUs on the Navi 22 chip, however clocks are fairly a bit increased. The reminiscence additionally will get reduce to a 192-bit bus, a 25% discount in bandwidth that interprets immediately into hashing efficiency. We set the utmost GPU frequency to 50% and ramped the VRAM clocks up by 150MHz, which resulted in precise clocks of round 1300MHz whereas mining.

Boosting the clocks again to 2.0GHz did not enhance hash charges, so it is best left alone. With these settings, we obtained 47MH/s, up from about 40MH/s at inventory, with energy draw of 120W. Temperatures had been fairly good, with fan pace of round 50%.

Actual-World Profitability and Efficiency

After the testing we have accomplished, one factor we wished to do was take a look at real-world profitability from mining. There is a purpose individuals do that, and outcomes can fluctuate fairly a bit relying in your particular {hardware}. Our outcomes positively do not match up with what we have seen reported on some GPUs at locations like NiceHash or WhatToMine. We have used the optimum tuned settings, in addition to energy draw figures.

Nevertheless, word that the ability draw we’re reporting does not embody PSU inefficiencies or energy for the remainder of the PC. We’re largely taking a look at reference fashions as properly, which frequently aren’t the best choice, however this is how our information compares to what NiceHash stories.

Mining Efficiency Comparability – Tom’s {Hardware} vs. NiceHash
TH Hashrate NH Hashrate % Distinction TH Energy NH Energy % Distinction
RTX 3090 FE 106.5 120.0 -11.3% 279 285 -2.1%
RTX 3080 FE 93.9 96.0 -2.2% 234 220 6.4%
RTX 3070 FE 61.3 60.6 1.2% 123 120 2.5%
RTX 3060 Ti FE 60.6 60.5 0.2% 116 115 0.9%
RTX 3060 12GB 48.6 49.0 -0.8% 119 80 48.8%
RX 6900 XT 64.6 64.0 0.9% 183 220 -16.8%
RX 6800 XT 64.5 64.4 0.2% 186 190 -2.1%
RX 6800 64.6 63.4 1.9% 166 175 -5.1%
RX 6700 XT 47.1 47.0 2.1% 120 170 -29.4%

There are some fascinating outcomes. The RTX 3090 Founders Version is unquestionably not one of the best pattern of mining efficiency, and NiceHash’s quantity (maybe with maxed out followers) is 12% increased than what we obtained, but in addition used 2% extra energy — or for those who desire, our numbers had been 11% slower whereas utilizing 2% much less energy. On the 3080, the FE ended up simply 2% slower whereas utilizing 6% extra energy.

NiceHash specifically recommends operating your followers at 90-100% on the 3080 and 3090, which may positively enhance efficiency. Our stance is that this can be a Very Unhealthy Thought ™. Not solely will the followers make lots of noise, however they’re additionally destined to fail sooner somewhat than later. In the event you’re okay changing the cardboard’s followers sooner or later, or if you wish to mod the cardboard with higher cooling pads within the first place, you may positively obtain the NiceHash efficiency figures. Energy use (as measured utilizing Powenetics) would in fact improve.

We had been comparatively shut on the 3060 Ti efficiency, and our earlier energy information confirmed significantly better outcomes than NiceHash, however now these figures have been up to date and are barely decrease than our measured energy. The RTX 3060 in the meantime ended up with related efficiency, however our energy outcomes had been considerably increased — maybe our EVGA pattern simply wasn’t start line.

Shifting over to AMD, the whole lot was very shut on efficiency — inside 2% throughout all 4 playing cards. This time, nevertheless, our energy testing confirmed wherever from a number of p.c to as a lot as 29% decrease energy necessities than what NiceHash stories.

The present pondering for lots of miners is that Nvidia’s RTX 30-series playing cards are superior to AMD, however that is actually solely true for those who take a look at pure hashrates on the 3080 and 3090. Think about energy effectivity and issues are a lot nearer. Apart from, it is not like you should buy any of those GPUs proper now — until you are prepared to fork out some huge cash or have some good trade contacts for constructing your mining farm.


Mining with Earlier Technology {Hardware}

Nvidia Ampere and AMD RDNA2, aka Large Navi, GPUs are getting a lot of the headlines in the present day, however what about earlier era {hardware}? The rules we have outlined above typically apply to the older GPUs as properly. We have included Nvidia’s RTX 20-series, GTX-16-series, and GTX 10-series playing cards, together with AMD’s RX 5000, Vega, and 500 collection elements.

We’ll skip all of the baseline efficiency metrics this time, and simply soar straight to optimized efficiency. Observe that exterior of the RX 580 and 590, and the three GTX 1660 variants, all of our assessments had been carried out utilizing the reference fashions from AMD and Nvidia, which are typically extra of a baseline or worst-case situation for many GPUs. We’ll listing our optimized settings under, however listed here are the outcomes.

* – Our GTX 1070 and GTX 1080 are unique Founders Version playing cards from 2016 and seem to carry out worse than different 1070/1080 playing cards. Our Vega playing cards are additionally reference fashions and had been much more finnicky than different GPUs. YMMV!

For Nvidia’s Turing GPUs, efficiency once more correlates just about immediately with reminiscence bandwidth, although with a number of fascinating exceptions. Discover that the 2060 Tremendous, 2070 Tremendous, and 2080 Tremendous all find yourself with almost an identical efficiency? That is not an error. The odd bit is that the 2080 Tremendous requires considerably increased reminiscence clocks to get there.

The 2060 Tremendous and 2070 Tremendous each needed to run with a +1100 MHz offset in MSI Afterburner, giving an efficient pace of 15.8Gbps (with a 400Mbps detrimental offset for compute). The 2080 Tremendous in the meantime has a base GDDR6 pace of 15.5Gbps and we had been in a position to max out the reminiscence overclock at +1500 MHz, yielding a last pace of 18Gbps for compute.

Besides we nonetheless ended up on the identical 44MH/s hash price. Almost definitely the reminiscence timings on the GDDR6 within the 2080 Tremendous are extra relaxed (looser), so despite the fact that bandwidth is increased, bandwidth plus latency finally ends up balancing out. 2080 Ti in the meantime hits the identical ~60MH/s because the 3060 Ti and 3070, due to its 352-bit reminiscence interface.

The GTX 16-series GPUs in the meantime supply an honest mix of efficiency and energy. The 1660 Ti and 1660 Tremendous are mainly the identical pace, although we had higher luck with reminiscence overclocking on the Tremendous. The vanilla 1660 with GDDR5 reminiscence does not clock fairly as excessive on the VRAM and thus performs worse on Ethereum mining. No matter which GPU you are taking a look at, although, the entire GTX 1660 fashions profit enormously from dropping the GPU clocks. That reduces energy use and temperatures and boosts general effectivity.

Stepping again one era additional to Pascal (GTX 10-series), the method modifications a bit. Most reminiscence clocks are nonetheless vital, however core clocks begin to matter extra—the structure is not tuned for compute as a lot as Turing and Ampere. We obtained our greatest outcomes by overclocking the GPU core and reminiscence pace, however then setting an influence restrict. Besides, being almost 4 years previous, two of our GPUs (the GTX 1070 and GTX 1080) actually weren’t proud of reminiscence overclocking. Something greater than a 200MHz bump on the 1080 prompted a tough PC crash, and whereas the 1070 managed +500MHz, our hashing outcomes had been nonetheless a bit decrease than anticipated.

Tweaking AMD’s earlier era GPUs is a bit totally different, in that somewhat than limiting the ability, one of the best method is to tune the voltage and clock pace. Not surprisingly, the older the GPUs get, the decrease the hash charges and effectivity change into. Let’s begin with the earlier era and transfer again from there.

The RX 5000 collection GPUs (aka RDNA, aka Navi 1x) all proceed to carry out fairly properly in Ethereum mining. RX 5700 XT and RX 5700 are almost as quick as the newest RDNA2 chips, the primary distinction being in reminiscence bandwidth. Most playing cards will not go a lot past round 1800MHz on GDDR6 clocks, which is equal to 14.4Gbps, although some would possibly have the ability to do 14.8Gbps. The RX 5600 XT is identical GPU however with a 192-bit reminiscence interface, which cuts efficiency down by about 25% once more, proper in line with reminiscence bandwidth. General, the 5700 and 5700 XT find yourself as probably the most environment friendly earlier era GPUs for mining.

Transfer again to the Vega structure and the big reminiscence bandwidth that comes from HBM2 comes into play. The Radeon VII with its monster 4096-bit reminiscence interface and a modest overclock for good measure finally ends up with 1178 GBps of bandwidth—solely the RTX 3090 comes shut. However Vega was additionally a power-hungry structure, and it advantages from turning down the GPU clocks. We ended up at a 1650MHz setting (inventory is 1800MHz) on the Radeon VII, and we set the core voltage to 880mV. That gave mining clocks of 1620MHz.


See Also Best Mining GPUs Benchmarked and Ranked

Vega 64 and Vega 56 used related settings, however half the reminiscence capability and bus width limits efficiency fairly a bit relative to the Radeon VII. Additionally, our outcomes on the reference blower playing cards are most likely far lower than very best—nearly any customized Vega card can be a better option than these blowers. We skilled a lot of crashing on the 2 Vega playing cards whereas making an attempt to tune efficiency.

After which there’s Polaris. Very like the Pascal GPUs, our tuning efforts took extra effort and time. Many locations suggest VBIOS flashing, however that is not one thing we wish to do (as we’re simply testing mining efficiency, not constructing a 24/7 mining farm). As such, our most hash charges ended up about 10% decrease than what some locations present. Apart from setting a low voltage of 800-900mV, relying on the cardboard, we set the reminiscence timings to level2 in Radeon Settings, and that gave one of the best outcomes with cheap energy use. YMMV.

Here is a gallery of all of the ‘tuned’ settings we used for the legacy playing cards. Use at your individual threat, and know that some playing cards desire totally different miner software program (or just fail to work with sure miners).

Is it doable to enhance over our outcomes? Completely. That is only a baseline set of efficiency figures and information, utilizing our particular samples. Once more, non-reference playing cards usually carry out a bit higher, and if you wish to analysis VBIOS flashing and {hardware} modding it is doable to hit increased hash charges. However out of the field, these are numbers that virtually any card utilizing considered one of these GPUs ought to have the ability to match.

Ought to You Begin Mining?

This brings us to the ultimate level we wish to make. Suppose you have already got a graphics card and wish to mine utilizing your GPU’s spare cycles. In that case, it is definitely value contemplating, notably for those who dwell in an space the place energy is not tremendous costly. You can theoretically internet $8–$11 per day from the precise 3080 or 3090 card, and $5–$6 per day with one of many Large Navi or lower-tier Ampere GPUs.

In the event you’re searching for the best choice for mining, from the angle of paying off the GPU as shortly as doable, the RTX 3060 Ti can be the cardboard to get — or at the least, it might be for those who might discover it wherever for lower than $1,000. No, that is not a typo, and no, you should not pay $1,000 for a 3060 Ti!

On the identical time, we strongly suggest towards ‘redlining’ your card for optimum hash price in any respect prices. The followers on shopper playing cards aren’t designed to spin at 100%, 24/7 with out failing. They will burn out for those who run them that manner. We even have critical issues with any part temperature that is constantly at or above 100C (or actually, even 90C).

However for those who do some tuning to get fan speeds all the way down to the 40-50% vary, with temperatures under 70C, you may most likely mine with a card for fairly some time with out having it go stomach up. Will or not it’s lengthy sufficient to recuperate the price of the cardboard? That is the large unknown.

Here is the factor to recollect: Cryptocurrencies are extraordinarily unstable. Ethereum’s value has fluctuated over 30% simply previously month, and sometimes swings by 5% (or extra) in sooner or later. It is also up 151% throughout the previous quarter, and greater than 1300% since one yr in the past. Because of this, as quick as the worth has shot up, it might plummet simply as shortly.

In the event you’re taking a look at $10 per day per GPU in mining earnings and pondering it is solely three months to interrupt even on some GPUs, you are assuming the worth and profitability will stay regular. It will not. There’s an opportunity it might go up, however the reverse is extra seemingly given the current soar in pricing. Simply ask the GameStop ‘investors’ how that labored out for those who assume the sky is the restrict.

Once more, if you have already got a GPU, placing it into service is not a horrible thought — it is your {hardware}, do with it what you please. Paying excessive costs for mid-range {hardware} to try to construct your individual private mining mecca, then again, is a giant threat. You would possibly do nice, you would possibly do nice, otherwise you would possibly find yourself with lots of additional PC {hardware} and debt. Plus, what about all of the avid gamers that will love to purchase a brand new GPU proper now they usually cannot? Any person, please consider the avid gamers!

Anyway, for those who’re searching for extra info, right here is our listing of the greatest mining GPUs, and we have checked profitability and each day returns of every GPU. Even for those who ‘solely’ have a earlier era graphics card, whilst you won’t have the ability to make fairly as a lot from mining, the GTX 1070 and RX 570 8GB playing cards can at the moment do over $3 per day. Simply bear in mind to account for energy prices and money out sufficient cash to cowl that, after which hopefully you will not get caught holding the bag.