How to Optimize Your GPU for Ethereum Mining | Tom’s Hardware

Ethereum gpu mining is still profitable, at least until it switches to proof of stake sometime this year (we hope). But there is more to it than just launching the software and letting it run in the background, especially if you have managed to get one of the best graphics cards or the best gpu for mining. Most of the graphics cards in our hierarchy of GPU benchmarks can make money right now by mining, depending on how much you pay for power. however, you’ll want to tune your graphics card to the optimal settings, and the make and model of the card can have a big impact on overall performance and efficiency. First, let’s keep in mind that we’re not trying to actively encourage anyone to start a gpus mining farm. if you want to know how to mine ethereum, we cover it elsewhere, but the “how” is quite different from the “why”. In fact, based on past personal experience some of us have using consumer graphics cards 24/7, it’s absolutely possible to burn out your card’s fans, vrms, or other elements. Also note that we periodically ‘update’ this article, but the original text was from early 2021. Mining at this stage is much less profitable. At the same time, we know there is a lot of interest in the topic, and we wanted to shed some light on the actual power consumption (measured with our powenetics team) that the various gpu’s use, as well as the real world hashrates that we use. . accomplished. If you got data using a mining profitability calculator, our figures indicate that there is a lot of variation between power and hash rates, depending on your setup and even your particular card. don’t be surprised if you don’t reach the level of performance shown by others. We’ll start with the latest generation of AMD and Nvidia GPUs, but we’ve got results for most older generation GPUs as well. nvidia and its partners now have amp lhr (hash rate lite) cards that run about half as fast as non-lhr cards, though newer versions of nbminer can get into the 70% range (with the appropriate drivers) : older is better, typically). however, nvidia has managed to undo some of those gains with updated drivers, so newer cards like the rtx 3050 might not be as good. All of this is likely to set the stage for nvidia’s next gpus, ada lovelace, which we expect to see in the latter part of 2022.

nvidia ampere and amd rdna2 mining performance

There are a few things you should know before you begin. first of all ethereum gpu mining requires more than 4gb of vram so if you still have a 4gb rx 570 it won’t work and neither will the new radeon rx 6500 xt. second, there are many different software packages for mining, but we are taking the easy way out and using nicehash miner. includes support for the most popular mining solutions and will even compare your card to determine which one performs best. although some algorithms may perform better than others, for now we will focus exclusively on the ethereum hash. We have used our standard gpu benchmark for these tests, running a single gpu. this is not necessarily an optimal miner pc setup, but it will suffice and is a closer representation of what most of our readers are probably using. You don’t need a high-end CPU, motherboard, or memory for mining purposes, and many larger installations will use Pentium CPUs and B360 chipset motherboards with more PCIe slots. The most important factors for a mining computer are power and cooling, as both have a direct impact on overall profitability. If you can keep your GPU and other components cool, they will last longer and won’t break as often. Meanwhile, power can be very expensive for larger mining facilities, and low-efficiency psuses (power supply units) will generate more heat and use more electricity. We’ve run these benchmarks using nicehash miner, analyzing actual hash rates in real time rather than the results of your built-in benchmark. we test each graphics card in stock mode and then also try to tune performance to improve overall efficiency and ideally keep temperatures and fan speeds at reasonable levels. we let the mining run for at least 15 minutes before checking for performance, power, etc, as things often slow down once the graphics card starts to heat up. It’s also important to note that we report raw graphics card power for the entire card, but we don’t take into account the power consumption of the rest of the PC or power supply inefficiencies. with an 80+ platinum power supply, we should be running at about 92% efficiency, and the power draw from the wall outlet is typically 50-80 watts higher than what we show on the charts. About 40w of power goes to the cpu, motherboard, and other components, while the rest depends on how much power the gpu uses, including power supply inefficiencies.

Reading: How to increase hashrate ethereum

There’s a lot to discuss with these charts, specifically what do we mean by “tightened” performance? the answer: it varies, often massively, by gpu. (after the initial article, we updated and added more gpu, but skipped the “stock” tests and only included our adjusted results). Let’s talk about the big picture quickly before we get into the details. the fastest gpus for ethereum mining right now are the rtx 3080 and rtx 3090, by a pretty wide margin. our baseline rtx 3080 fe measurement got 85 mh/s, and the baseline 3090 fe got 105 mh/s. additional tuning improved the 3080’s performance to 93 mh/s, while the 3090 fe limited us (on memory temperatures) to around 106 mh/s. It’s critical to note that the 3080 and 3090 founders edition cards get very hot on the gddr6x, limiting performance. modifying those with better thermal pads or purchasing a third-party card can increase performance and lower memory temperatures. meanwhile, the rtx 3060 ti and 3070 cards started at around 52 mh/s, although the 3070 is theoretically faster. that’s because the ethereum hash is highly dependent on memory bandwidth. overclocking the vram on those gpus got performance around 60mh/s. the amd rx 6800, 6800 xt and 6900 xt cards started at about 60 mh/s, and with tuning we achieved 65 mh/s; there wasn’t much difference between the three amd gpus, mainly because they all use the same 16gb of 16gbps gddr6 memory. finally, the new rtx 3060 and rx 6700 xt have 12gb of gddr6, with a 192-bit memory bus width, effectively reducing bandwidth by 25% relative to 256-bit cards. this in turn limits performance to around 47-48 mh/s after tuning. without the memory overclock, the performance drops to around 40 mh/s. you can refer to the power graph, but the overall efficiency graph is more important than the gross power usage. here the lower powered rtx 3060 ti and 3070 shoot up, and then there is a moderate step up to the rtx 3060, 3080, rx 6700 xt, rx 6800 etc. Most cards are pretty close in terms of overall efficiency for ethereum mining, although the extra gpu cores in the 6800 xt and 6900 xt ended up lowering the efficiency a bit; more tweaking could improve results, especially if you’re willing to sacrifice a bit of performance to reduce power usage. Finally, we have the temperatures. These are the core temperatures of the gpu, but they are not really the critical factor on many of the cards. amd cards worked with standard settings, but all cards benefit greatly from tuning. more importantly, while we couldn’t get gddr6 temperatures on the 3060 ti and 3070, we did get vram temperatures on the 3080 and 3090, as well as amd cards. In stock, the 3080 and 3090 founders edition cards hit 108-110c on gddr6x, at which point gpu fans bumped up to 100% (or close to it). cards were set at 106 degrees celcius, with gpu clocks fluctuating quite a bit. amd’s rx 6000 cards also maxed out at around 96c on their stock gddr6, but the tweak dropped vram temps down to around 68-70c. this brings us to the main area of ​​interest.

how to adjust the ethereum mining performance of your graphics card

Let’s start by noting that every card model is different and even cards of the same model can vary in performance characteristics. for 3080 and 3090 cards with gddr6x memory, maintaining that temperature is critical. we have seen examples of cards (specifically the evga rtx 3090 ftw3) that can run at up to 125mh/s while the memory is around 85c. that’s because evga seems to have put a lot of effort into cooling the memory. without altering the cards, the nvidia 3080/3090 founders editions allow the memory to get very hot during removal, which can dramatically hamper performance and/or reduce the life of the card. Let’s take each card in turn. geforce rtx 3090 founders edition – While technically the fastest mining card we tested, we really don’t like the idea of ​​running this 24/7 with no mods from hardware or serious adjustments. on stock the fans end up running at 100% to try to keep the gddr6x below 110c, and that’s not good. For our purposes, we tuned the card by lowering the gpu core to the maximum allowed -502mhz, set the memory clock to +250mhz, and set the power limit to 77%. that gave us a gddr6x temperature of 104c, which is still higher than we’d like, and performance hovered around 106mh/s. power usage also dropped to 279w, that’s pretty good considering the hash rate. alternatively you can go broke and max fan speed, set power to 80%, lower gpu clocks by 250-500mhz and increase vram clocks by 750-1000mhz. if you don’t modify the card to improve gddr6x cooling, you’ll typically end up at 106-110c (depending on your card, case, cooling, and other factors), and max fan speed won’t be good for fan longevity. consider yourself warned. geforce rtx 3080 founders edition: The tuning of this card was very similar to the 3090 fe. doesn’t like the stock configuration as the gddr6x gets very toasty. We lowered the gpu core back to the maximum allowed (-502mhz), set the memory clock to +750mhz, and set the power limit to 70%. that resulted in the same gddr6x temperature of 104c as the 3090 fe, and performance was only slightly slower at 93-94mh/s. Once again, maximizing fan speeds and memory clocks while lowering gpu core clocks and power capping are keys to improving overall hash rates. It is possible to modify the card and replace the vram thermal pads with thicker/better pads, which will help cooling and performance. however, we would prefer to use an rtx 3080 with better gddr6x cooling. which brings us to a card that we have since removed from the lists. the colorful rtx 3080 vulcan is an example of a 3080 model with better vram cooling than nvidia’s reference card, so the memory didn’t get as hot. however, we found that we achieved the best result by reducing the power limit to 80-90% and then setting the gpu core clocks to the minimum possible value in msi afterburner (-502mhz). We then overclocked the memory to a base clock of 750mhz, which gave a final speed of 20gbps (amp cards run 0.5gbps below their rated memory speed when removed). That produced similar hash rates of 93mh/s, while fan speed, gpu temperature, and power consumption all dropped. most importantly, with (relatively) the same performance as the 3080 fe, gddr6x temperatures stabilized at 100c. It’s not ideal, but at these temperatures a 4°C difference can be significant.

the rtx 3070 and rtx 3060 ti have the same 8gb gddr6 14gbps and as we will see with amd gpus that seems to be the limiting factor. our initial results were poor as these were the first cards we tested, but we revised the settings after seeing the rx 6000 series. geforce rtx 3070 founders edition – main thing to improve performance on the 3070 was boosting the gddr6 clock. we were able to add 1200mhz, giving an effective speed of 16.4gbps in theory, but the memory was actually running at 16gbps (vs. 13.6gbps in standard configuration). Along with increasing memory clocks, we lowered the gpu clock to the maximum -502mhz in afterburn and set the power limit to 55%. that resulted in actual gpu clocks of 960mhz on average. You’d think that wouldn’t be enough, but bumping the GPU clocks up to 1.9GHz resulted in the same performance and substantially increased the amount of power used. meanwhile, at 55%, the 3070 is second only to its little brother in overall efficiency. geforce rtx 3060 ti founders edition: As with the 3070, we increased memory speed as a primary change to improve performance. We went from stock to +1200mhz (compared to +750, which is as far as we could go for gaming overclocks), giving a maximum speed of 16.4gbps, with a negative 400mhz offset for workloads gpu computing Other settings were similar: -502mhz gpu clock, 55% power limit, and 50% fan speed. performance was very close to the 3070 but used less power, making it the overall efficiency winner. asus rtx 3060 ti tuf gaming oc – Again, we removed this from the updated graphics, but unlike the 3080 and 3090, the 3rd party cards weren’t much different in hashing performance to the 3060 you and the 3070 gpus. Our tweaked configurations ended up with higher clocks (due to factory overclocking) and more power usage than the 3060 Ti Founders Edition, but basically the same hashing performance. Optimum efficiency may vary a bit, but for GA104 based cards, GDDR6 speed is the limiting factor in performance. evga rtx 3060 12gb – There is no official reference card for the new ga106 based rtx 3060, so we are using the evga card we used in our rtx 3060 12gb review. we set the power limit to 75%, increased the gddr6 clocks by 1250mhz, and ended up with a relatively high fan speed of around 80% with the default fan curve. (cooling on this card is not as robust as many of the other gpu’s). We didn’t record “stock” throughput, but it was around 41 mh/s.

See also: How to solve pending (stuck) transactions on Ethereum | by Karel Striegel | Venly

radeon rx 6900 xt (reference): The fit of the three amd rx 6000 reference cards was very similar. gpu clocks can go up a lot in stock, but memory bandwidth seems to be the main bottleneck. running with gpu clocks of 2.2-2.5ghz just wastes power and generates heat without improving performance. we increased the power limit to max 115% just to make sure vram wasn’t held up, then set memory to +150mhz (the max allowed in radeon settings), enabled fast ram times and lowered max clock from gpu at 70%. that gave us final clocks of 1747mhz compared to 2289mhz in stock and 8% higher hash rates overall. More importantly, power consumption took a huge dive and efficiency improved to one of the best results in our tests. but this is actually not amd’s best overall presentation. radeon rx 6800 xt (benchmark) – Same approach as above, but due to differences in core configuration and… honestly, we’re not sure what the deal is, but we end up with a maximum optimal this time the gpu frequency setting was only 50% giving us clocks of 1206mhz instead of 2434mhz and the performance kept increasing matching the rx 6900 xt and the rx 6800 at the same time, the power requirements were reduced substantially, from 281 watts to 186 watts. Whatever is going on behind the scenes, it appears that the amd navi 21 gpus are performing optimally at different “max frequency” settings. Like nvidia, amd gpu’s are limited in performance largely by the speed of their memory, and without tools to overclock beyond 17.2 gbps, there’s not much you can do. radeon rx 6800 (benchmark): At just 60 cus (compared to 72 on the 6800 xt and 80 on the 6900 xt), expect the standard 6800 card to be slower. however, memory is once again the deciding factor. we set the gpu power limit to the same 115%, which does make a difference, interestingly: the average power dropped by about 15w if we set it to 100%, even though the card was running well below the official tgp of 250w . We also maxed out the memory slider to +150mhz (17.2gbps effective), this time achieving optimal performance with the gpu set to 75% at max clocks. that resulted in a clock of 1747mhz compared to 2289mhz in stock, but the fan speed was higher this time. That’s because we set the fan to run at 40% at 50°C, 60% at 60°C, 80% at 70°C, and 100% at 80°C, and it ended up at 50% speed, Which may be more than necessary, but we think it’s better to be safe than sorry if you’re looking to mine 24/7. radeon rx 6700 xt (benchmark): amd throttles down to just 40 cus on the navi 22 chip, but clocks are slightly higher. memory is also reduced to a 192-bit bus, a 25% reduction in bandwidth that directly translates to hashing performance. We set the maximum gpu frequency to 50% and increased the vram clocks by 150mhz, resulting in actual clocks of around 1300mhz while mining. boosting the clocks back to 2.0ghz didn’t improve hash rates, so it’s best left alone. With this configuration, we got 47 mh/s, against 40 mh/s in stock, with a power consumption of 120 watts. temperatures were pretty good, with a fan speed of around 50%. radeon rx 6600 xt (asrock): navi 23 reduces the gpu to 32 cus, but the memory interface is now just a 128-bit bus. after tuning the rx 6600 xt basically gets about 2/3 of the hashrate of the rx 6700 xt. We reset the maximum gpu frequency to 50% and increased the vram clocks by 150mhz, resulting in actual clocks around 1300mhz while mining. we got 32 mh/s, against 28 mh/s in stock, with a power consumption of 75 watts. the efficiency looks good, but the raw hashrate is definitely lacking: it’s only slightly faster than an 8gb rx 470 from five years ago.

performance and profitability in the real world

After the tests we completed, one thing we wanted to do was look at the actual profitability of mining. there’s a reason people do this, and results can vary quite a bit depending on your specific hardware. our results definitely do not match what we have seen reported on some gpus in places like nicehash or whattomine. we have used the optimum tuned configuration as well as the power consumption figures. However, please note that the power consumption we report does not include power supply inefficiencies or power to the rest of the PC. We mainly also look at benchmarks, which are often not the best option, but this is how our data compares to nicehash’s reports.

there are some interesting results. the rtx 3090 founders edition is definitely not the best display of mining performance, and the nicehash number (maybe with maxed out fans) is 12% higher than what we got, but it also used 2% more of energy, or if you prefer, our numbers were 11% slower while using 2% less energy. in the 3080, the fe finished only 2% slower and used 6% more energy.

nicehash specifically recommends (opens in a new tab) running your fans at 90-100% on the 3080 and 3090, which can definitely improve performance. our position is that this is a very bad idea ™. fans will not only make a lot of noise, but they are also bound to fail sooner rather than later. if it’s okay to replace the card’s fans in the future, or you want to mod the card with better cooling pads in the first place, you can definitely achieve nicehash’s performance figures. power usage (measured using powenetics) would of course increase.

We were relatively close to the performance of the 3060 ti and our previous power data showed much better results than nicehash, but now those numbers have been updated and are slightly lower than our measured power. Meanwhile, the RTX 3060 finished with similar performance, but our power results were significantly higher; maybe our evga sample was not a good starting point.

switching to amd, everything was very close to performance: within 2% on all four cards. this time, however, our power tests showed anywhere from a few percent to 29% less power requirements than nicehash reports.

See also: Benefits of Ethereum Social Media Platforms | Gemini

The current thinking of many miners is that nvidia’s rtx 30 series cards are superior to amd, but that’s only true if you look at the pure hash rates on the 3080 and 3090. note the power efficiency and things are much closer. Also, it’s not like you can buy any of these gpu’s right now, unless you’re willing to shell out a lot of money or have some good contacts in the industry to build your mining farm.

mining with older generation hardware

nvidia ampere and amd rdna2, aka big navi, gpus are getting most of the headlines nowadays, but what about previous generation hardware? The principles we’ve outlined above generally apply to older GPUs as well. We’ve included nvidia rtx 20, gtx-16, and gtx 10 series cards, along with amd rx 5000, vega, and 500 series parts. this time we’re going to skip all the benchmark performance metrics and jump right into optimized performance. note that outside of the rx 580 and 590, and the three variants of the gtx 1660, all of our testing was done using the amd and nvidia reference models, which tend to be more of a baseline or the worst of the cases for most gpus. we’ll list our optimized settings below, but these are the results.

*: Our gtx 1070 and gtx 1080 are original founders edition cards from 2016 and seem to perform much worse than other 1070/1080 cards. our vega cards are also reference models and were much finer than other gpus. mmmm! for nvidia’s turing gpus, performance again correlates fairly directly with memory bandwidth, albeit with some interesting exceptions. have you noticed that the 2060 super, 2070 super and 2080 super all end up with almost identical performance? that’s not a bug the weird thing is that the 2080 super requires substantially higher memory clocks to get there. the 2060 super and 2070 super had to run with a +1100 mhz offset in msi afterburner, giving an effective speed of 15.8 gbps (with a negative 400 mbps offset for compute). meanwhile, the 2080 super has a base gddr6 speed of 15.5 gbps and we were able to max out the memory overclock to +1500 mhz, giving a final speed of 18 gbps for compute. except we still end up with the same hash rate of 44 mh/s. most likely the memory timings on the gddr6 in the super 2080 are looser (more flexible), so even though the bandwidth is higher, the bandwidth plus latency ends up balancing out. meanwhile, the 2080 ti hits the same ~60 mh/s as the 3060 ti and 3070, thanks to its 352-bit memory interface. Meanwhile, the GTX 16 series GPUs offer a decent mix of performance and power. the 1660 ti and 1660 super are basically the same speed, although we had better luck overclocking the memory on the super. the vanilla 1660 with gddr5 memory doesn’t run as high on vram and thus performs worse when mining ethereum. however, regardless of which gpu you’re looking at, all gtx 1660 models benefit greatly from reduced gpu clocks. which reduces energy use and temperatures and increases overall efficiency. going back one more generation to pascal (gtx 10 series), the approach changes a bit. max memory clocks are still critical, but core clocks are starting to become more important: the architecture isn’t tuned for computation as much as turing and amperage. We got our best results by overclocking core gpu and memory speed, but then we put a power cap on it. except, being almost four years old, two of our gpu’s (the gtx 1070 and the gtx 1080) really weren’t happy with the memory overclocking. anything over 200mhz on the 1080 caused a computer crash, and even though the 1070 handled +500mhz, our hash results were a bit lower than expected.

tuning old gen amd gpu is a bit different in that instead of limiting power the best approach is to tune voltage and clock speed. not surprisingly, the older the gpus, the lower the hash rates and efficiency. let’s start with the previous generation and work backwards from there. the rx 5000 series gpus (aka rdna, aka navi 1x) continue to perform quite well mining ethereum. rx 5700 xt and rx 5700 are almost as fast as the latest rdna2 chips, the main difference is memory bandwidth. Most cards do not exceed 1800 MHz in gddr6 clocks, which equates to 14.4 gbps, although some can reach 14.8 gbps. the rx 5600 xt is the same gpu but with a 192 bit memory interface, reducing performance by about 25% again, at the same rate as memory bandwidth. Overall, the 5700 and 5700 XT end up as the most efficient previous generation GPUs for mining. Go back to the vega architecture and the high memory bandwidth that comes from hbm2 comes into play. the radeon vii with its monstrous 4096-bit memory interface and a modest overclock for good measure ends up with 1178 gbps of bandwidth; only the rtx 3090 comes close. but vega was also a power hungry architecture, and it benefits from lowering gpu clocks. we ended up with a 1650mhz setting (stock is 1800mhz) on the radeon vii, and set the core voltage to 880mv. which gave 1620mhz miner clocks. vega 64 and vega 56 used similar configurations, but half the memory capacity and bus width severely limit performance relative to radeon vii. Furthermore, our results on benchmark blower cards are probably far less than ideal – just about any custom Vega card would be a better choice than these blowers. We experienced a lot of crashes on both vega cards when trying to tune performance. and then there’s polaris. like pascal gpu, our tuning efforts took more time and effort. many places recommend flashing vbios, but that’s not something we want to do (since we’re just testing mining performance, not building a 24/7 mining farm). as such, our maximum hash rates ended up 10% lower than what some places show. in addition to setting a low voltage of 800-900mv, depending on the card, we set the memory timings to level 2 in the radeon settings, and that gave the best results with reasonable power usage. hmm.

here is a gallery of all the “tweaked” settings we use for legacy cards. use at your own risk, and know that some cards prefer different miner software (or just don’t work with certain miners). Is it possible to improve our results? absolutely. this is just a reference set of performance figures and data, using our specific samples. again, non-reference cards often perform a bit better, and if you want to look into vbios flashing and hardware modding, higher hash rates are possible. but out of the box these are numbers that almost any card using one of these gpus should be able to match.

should you start mining?

This brings us to the final point we want to make. Let’s say you already have a graphics card and you want to mine using your gpu’s spare cycles. in that case, it might be worth considering, especially if you live in an area where energy is not very expensive. however, while before you could theoretically get more than $10 a day with a 3080 or 3090 card, today’s profitability has dropped significantly and less than $4 a day is typical for a 3090. At the same time, we strongly recommend that you don’t “draw” your card to get the maximum hash rate at all costs. consumer card fans are not designed to spin at 100%, 24/7 without failing. they will burn if you run them that way. We also have serious concerns with the temperature of any component being consistently at or above 100°C (or for that matter, even 90°C). but if you make some adjustments to reduce the fan speed to the 40-50% range, with temperatures below 70°c, you can probably mine with a card for quite some time without it crashing. Will it be enough to recover the cost of the card? that is the great unknown. here’s the thing to remember: cryptocurrencies are extremely volatile. Ethereum price fluctuated more than 30% in the last month alone and often swings 5% (or more) in a day. this means that as fast as the price shot up, it could plummet just as fast. At one point, it might have been possible to break even on the cost of a new GPU in a few months. these days, it would take more than a year at current rates, assuming nothing changes. could go up, but the opposite is more likely. just ask gamestop ‘investors’ how it worked out if you think the sky is the limit. again, if you already have a gpu, putting it into service isn’t a terrible idea – it’s your hardware, do what you want with it. Paying extreme prices for mid-range hardware to try to build your own personal mining mecca, on the other hand, is a huge risk. You may do well, you may do very well, or you may end up with a lot of additional PC hardware and debt. also what about all the gamers who would love to buy a new gpu right now and can’t? someone please think about the players! Anyway, if you are looking for additional information, here is our list of the best gpu for mining, and we have verified the profitability and daily yields of each gpu. just remember to factor in energy costs and collect enough coins to cover that, and then hopefully you won’t get caught holding the bag.

See also: Ethereum StarkWare Completes Sequoia Led Funding Round With 50M

Related Articles

Leave a Reply

Your email address will not be published.

Back to top button