The Nvidia GeForce GTX 1070 was an absolute gaming beast when it released back in June 2016. Despite its age, its popularity hasn’t dwindled, and it was actually making a comeback before all this crazy GPU pricing started. Before, you’d be able to find one of these graphics cards at less than $200, which is incredible considering I paid $329 back in 2019 for the card I’ve got here. But with the crypto boom, supply issues, and the ridiculous pricing that they’ve brought along with them, you’re now looking at a price tag more around $300 – 500 on a good day… Once the bubble bursts again, prices will come back down, and in preparation for that, let’s take a look at the GTX 1070 in 2021, and see if this card is worth keeping an eye out for once prices finally stabilize!
The Specs of the Card
In typical Proceu Tech fashion, let’s start off with the hardware specifications of the card, and discuss the different requirements you would need in order to run the card in your system. The onboard power connectors are rated for a single 8pin PCIe connector on the Founders Edition, however this can vary depending on where you buy the card from. In terms of the silicon in this card, we’re presented with a cut-down GP-104 die. This means we’re getting an overall backbone similar to that of the higher-end GTX 1080, with a memory feature set more in line with the GTX 1060. You’re getting the aesthetics of the 1080 as well, which with this specific model is definitely a good thing!
This also bring us to the power requirements, which for this card are a rated TDP of 150W and a recommended PSU of 450W. To be safe, I would go for at least a 500W unit, especially if you plan on overclocking the card to anything beyond stock clocks, but to get the card up and running, a 450W unit is all you need. A 600W unit is probably what I’d recommend if you’re interested in serious overclocking, or need the utmost stability. Power connectors for the card are also relatively basic and speak to the card’s relative efficiency when compared to its Turing- and Ampere-based successors. Requiring only a single 8-pin PCIe power connector, this card draws most of its power through the PSU cables, rather than the PCIe slot on the motherboard, which helps with compatibility on older boards. It should be noted, however, that this specification can change from vendor to vendor, and even depending on the model.
Some require an 8 + 6-pin connector, while others may require all the way up to two 8-pin connectors, so it’s important to check the listing for the card you’re looking at to ensure proper compatibility. Overclocking on this particular model is also not recommended, as it’s the reference model, meaning thermal dissipation isn’t this card’s strong suite. Speaking of clock speeds, this card has a stock clock of 1,506 MHz and a rated boost clock of 1,683 MHz. In-game, clock speeds usually hovered around the 1,800 MHz mark, with 1,792 MHz coming up quite frequently.
However, as time went on, the clocks began to droop down to the rated boost clock due to thermal limitations. This is not anything to be worried about, however it does bring into question the value of a hybrid mod, or buying a model with a larger heat sink. This is the reference design after all, and it’s important to keep in mind that the heatsink is only roughly the size of a can of soda. Cosmetically, I also modded my card and painted the metal parts of the shroud an opaque gold color, which looks kind of tacky but doesn’t effect thermal performance. I was bored one weekend and decided to paint it. That’s really the only explanation I can give, but it was a fun little mod.
Moving onto the feature set of this card, the GTX 1070 has most of the modern extensions, however it does lack Ray Tracing and Tensor Cores. For strictly gaming without DXR, this isn’t a big deal, but if you were looking to get into DLSS, then this card is not the best choice. During the current price hike, if you need DXR for a specific workload, then the RTX 2060 is probably the go-to card at the moment, even though it’s a bit pricey. While yes, this card is technically capable of ray tracing, it does so in software, meaning it’s beyond slow, even for a basic recursive implementation. Jumping into Rasterization performance, the GTX 1070 packs in 1,920 CUDA cores, each capable of either a 32-bit floating point or 32-bit integer calculation per clock cycle. This is a quirk in the design of the 10-series, and if you’re familiar with my GTX 1080 review, that card works in the exact same way.
All Pascal cards have this feature, and it ultimately limits the amount of mixed data type computations that can be carried out. In performance terms, this doesn’t lead to any significant issues when gaming, however it is limited in terms of FP16 and FP64 compute, both of which are common in scientific and work station use cases. If you’re going to be running protein folding simulations or doing some hard, high precision math, then the GTX 1070 might be best left to the gamers, because this card just doesn’t have the horsepower. Hell, even the card’s bigger brothers from this generation struggle with those workloads.
If you’re really into heavy compute, then I would suggest a Turing card if integer workloads are a big part of your workflow, and Ampere if you’re looking for tons of FP32 and FP16 power. The cache structure found in this card is also incredibly similar to the pipeline found in Maxwell 2.0, with 48KB of L1 cache per SM, and an overall shared L2 size of 2MB. Compared to more modern offerings, such as the RTX 3070, this amount of cache is tiny, and limits the amount of data that can be processed at once. Less cache means more of the data has to be stored in the card’s VRAM, meaning the 1070 is overall less efficient with its memory allocation. Compounded with the lower bandwidth of GDDR5, this leads to this cards ultimate Achilles heel – the memory configuration.
Rocking 8GB of GDDR5 split into eight 1GB chips, this card makes use of a 256-bit Bus, with clocks on the memory allowing for 8Gbps on the chips, and 256 GB/s on the GPU. Just for comparison’s sake, this card’s memory configuration was actually excellent for the time, and allowed the card to fully saturate a nice, wide pipeline. In the years since, GDDR6 has moved the base bandwidth up to 448 GB/s, putting the 1070’s memory a solid 42% behind the configuration found in the RTX 2070. In-game, you’ll notice more frame dips on the 1070 when particle effects fill up the screen, however the GP104 chip itself is ultimately being held back by the memory configuration.
I wish there were a way to see how this card would perform with GDDR5X, because the specs of the chip itself are still impressive and rather capable. With the 15 SMs on board, we get 120 Texture Mapping Units and 64 Rasterization Operation Pipelines. The TMUs are definitely being held back here by the lack of on-die cache and the slower memory, meaning they can’t do their job as effectively as they possibly could. You could arguably say that about most graphics cards, but the overall age of the 1070’s G5 configuration is holding it back from being a no-brainer recommendation. Sure, it’s got 8GB, which is plenty for 1440p gaming, but in newer games where particle effects are used extensively, the 1070 will start to struggle a bit more, as it lacks the high levels of bandwidth found in its GDDR6-based successors.
This card also lacks the memory compression algorithms found in Ampere, although delta color compression is in use here, which really helps to keep the 1070 and it’s Pascal brethren more in line with modern cards. Overall, the feature set is the Gold Standard Nvidia setup, minus ray tracing. For some of us, this isn’t that big of a deal, as while it’s technically impressive and graphically stunning, it’s so costly that some of us would prefer to just use traditional rasterization to get the job done. The memory configuration is what’s theoretically holding the cores back the most, however techniques introduced by Nvidia can help to mitigate this, along with a slight memory overclock! Just looking at the spec sheet of the 1070 is enough for your mouth to start watering, especially when you realize its implications for gaming!
This brings us to the specifications of our test system, and our methodology for capturing the numbers we will be discussing! For our CPU, we’re rocking the beastly Intel Core i9-9900k, an 8-core 16-thread chip, socketed in my MSI Z390-a Pro motherboard. Alongside our CPU, we paired the card with 32GB of DDR4 clocked at 2,933MHz. This is kind of an unusual clock speed, however I am running them undervolted to cut down on heat. It’s also an overclock anyway, as our 9900k comes stock with a 2,666MHz controller. I switched out my GTX 1080 with the GTX 1070, however I will only be testing it with the stock cooler.
Either way, we’ve isolated GPU performance to help give us an idea as to where the 1070 sits in the power ladder, and to find out whether it’s ultimately worth picking up. In order to test our card, we also ran all of our games at 1440p for 10 minutes, with the graphical values set to High. This helps to stress the GPU more so than the CPU, while 1440p is also a middle ground between 1080p and 2160p. It’s also important to keep in mind that performance will look even better if you’re planning on gaming at 1080p, but for the benchmarks all the games will be presented in 1440p! With the specifications and testing sequence out of the way, let the benchmarks begin!
Starting off with a rather popular battle royale title, Apex Legends performed exceptionally well, and left enough headroom to possibly play this game in 4K. This game does feature a dynamic resolution scaler though, so if you want locked framerates, then that’s definitely then way to go as opposed to running at your native resolution at all times! An average of 89 FPS kept the game feeling smooth, and the minimums show that performance never dropped to an unplayable level, however dropping settings and resolution is a quick way to get these numbers up! VRAM usage can also be adjusted via the in-menu texture quality slider, but with the 8GB on this card, the highest buffer allocation ran well enough to make me comfortable recommending turning that setting up! For coming up on 5 years old, the 1070 holds up incredibly well, especially at 1440p and at these settings. This card paired with a cheaper hexa-core chip would make for an amazing setup for dominating the battlefield in this rather fun game!
Our first Unreal Engine game, ARK: Survival Evolved, performed admirably on the 1070, however the framerate leaves a lot to be desired. With an average of 56 FPS, this game is right on the cusp of being playable at 1440p, but sadly misses the mark ever so slightly. Even the GTX 1080 struggled a bit with this game, so I’m sure the optimization of the game had something to do with the low performance figures we saw. However, you can always lower the resolution or turn on the resolution scaler if you prefer framerate over resolution. Keep in mind that ARK performs somewhat poorly no matter your hardware, so if you’re looking to play this game at 1440p and with console framerates, then the GTX 1070 is happy to oblige, and even offers incredible high-FPS gameplay at 1080p! Even if you wanted to keep it set to 1440p, lowering the settings would do wonders, easily allowing you to claw back frames and stabilize your overall experience. Regardless, it’s probably worth looking into dropping your settings in this game no matter your hardware configuration.
Call of Duty: Black Ops – Cold War performed well in our test session, giving us an average framerate of 69 FPS. The minimum tells another story though, with the game stuttering frequently when explosions fill up the screen. Tying this back into the memory configuration we discussed earlier, this is most likely due to the onboard memory, rather than the fault of the chip itself. While the game looks great at these high settings, the amount of alpha channel particle effects in this game just bog the 1070 down. This can all be mostly solved by lowering the number of objects that are sorted through transparencies, which can easily be done through the game settings. With this filter setting kept at its ‘High’ value, the game performed well, and dropping the resolution is an easy way to get higher framerates with these higher settings! It was a blast to run and gun through Nuketown on the 1070, and it was even more fun to slay zombies in the forests of Poland!
Our Cry Engine title, Crysis 3, returned an average of 74 FPS, with the minimum clocking in at 51 FPS. Crysis, like Doom Eternal, has a weird settings menu, where the ‘High’ graphics setting is equivalent to most other games’ medium settings. Overall, this puts the 1070 a nice chunk behind the 1080. However, like with most of the games we’re testing today, a ton of these performance issues can be solved by dropping the resolution to 1080p. It’s not optimal, however if you want the utmost smoothness, I would stick to maxing the game out at 1080p. It’s glorious to experience this game with every graphics option set to the highest possible values, and it feels much better to play in 1080p thanks to the rendering taking less time. The 1070 also had a weird issue in this game whereby explosions caused the framerates to drop. They weren’t catastrophic by any means, but the dips were still there, and like with Black Ops – Cold War, I suspect this is due to the memory limitations, rather than the chip itself not being able to handle the game!
Another Source Engine game, CS:GO returned an impressively playable 297 fps. Looking at the 1080’s performance though, it looks as if we’re CPU-limited. We didn’t see much of a gain in framerates with the upgrade to the -80-class card, meaning that the 1070 can max out an 8-core 16-thread Skylake chip no problem! In fact, if you’re looking at the 1070 for a CS:GO-centric build, I’d be looking more into your CPU choice, as even a GTX 1050ti can max out CS:GO and stay above 200 FPS. While this isn’t a CPU review, I would recommend a chip like the i5-10400 if you’re looking at getting onto LGA 1200, or a Ryzen 5 3600, however the i5 is cheaper at the time of writing! Back to graphics performance though, and even at 4K, the 1070 is able to sustain similar framerates, and although I didn’t capture data from that test, I can confirm that the FPS hung around the 300 FPS mark, meaning the 1070 didn’t really care all that much about pumping more pixels in this game. That’s probably because this game is leveraging the CPU more than the graphics card, but for a competitive gaming PC, this card seems hard to beat if you’re able to find a decent deal!
Up next is Cyberpunk 2077, and if you saw my 1080 review, you’ll know that this game loves to give Pascal cards a hard time. The trend continues here, with the 1070 attaining an average of 23 FPS. I do want to mention that the minimum and maximum are actually grouped relatively tightly, though, showing some level of consistency with the card. While it didn’t perform well at 1440p, dropping the resolution to 1080p or dropping some settings should help to get the game above 30FPS, and move it into more playable territory. I wouldn’t necessarily write the 1070 off for Cyberpunk though, as all Pascal cards run this game somewhat poorly at higher settings. However, turning settings down to medium helps to provide a bump in performance, while maintaining the graphically pleasing Cyberpunk look! It was fun to explore Night City, and it was fun to hack things and enjoy the weird NPC behavior we’ve come to expect from this game!
Doom Eternal, the first Vulkan game in our test suite, ran flawlessly at 1440p. Like Apex Legends, this game include a dynamic resolution scaling option, meaning performance can be locked at almost all times! However, with an average of 106 FPS and a minimum of 68, it’s hard to justify using the setting when it natively runs so well. This game was a blast to play on the 1070, and this is just a strong demonstration for Vulkan in general. Even the older OpenGL-based Doom 2016 performed well on this card, although it is notable how much more efficient ID Tech 7 is over its predecessors! Either way, Vulkan is being displayed in full force in Doom Eternal, and the 1440p goodness was made all the more enjoyable thanks to the high and relatively stable framerates, which never dipped below the 60FPS mark. This was a very stable and smooth experience, leaving me open to recommending the 1070 for Doom Eternal-focused builds!
Moving into our first Open World RPG, Far Cry 5 performed relatively well, and like Doom Eternal never dipped below the 60FPS mark. It’s kind of impressive that this game holds up this well graphically, but I think what’s more impressive is how well it runs on this hardware! It’s a 5-year-old graphics card running a newer game at 1440p at 60 FPS and above. It really shows how far we’ve come in a few short years, but I’m glad that the 1070 is still holding up well with this game 3 years on from release. This does bring up the question of lowering the resolution though, and like I’ve shown in previous reviews, this card is able to stand up to this game and deliver some monstrous performance in 1080p. In terms of framerates, a drop in resolution would give you roughly 50% more frames, which makes the game feel even smoother. It was fun to experience Hope County in 1440p at 60 FPS, and it’s even better when its in 1080p at 120Hz!
Our next Unreal Engine game is honestly an incredible graphical showcase for the 4th iteration, as Gears 5 performed very well, and at 1440p we averaged 81 FPS. This is Xbox One X killing levels of performance, and its smooth gameplay was honestly a treat to experience, especially since my original experience with the game was on the original Xbox One. Despite the praise, there were still some issues, particularly with clock speeds. I began to notice that, as time went on, the overall clock speed of the card began to drop. This is normal, and is usually done to keep the card running within operational spec, however the cooler being used on this model was insufficient to the point where I was simply sustaining the base clock rather than the boost clock. This makes me curious as to how much a hybrid mod would help out this card, especially in this incredibly demanding game. It might be worth picking up a model other than a Founders Edition if you’re looking to get the absolute most from your card. However, even when limited to an extremely restrictive heat sink, the 1070 flew through Gears 5. It was fun to blast and chainsaw enemies at 60FPS and above, made all the sweeter by our 1440p resolution!
Grand Theft Auto 5, a now older game built on Rockstar’s proprietary R.A.G.E. Engine, ran like an absolute champ, and actually beat our GTX 1080 by a measurable margin. This is probably due to some randomized in-game event, such as an explosion that we didn’t see, but either way the card proves to be very competent when it comes to GTA 5. You’d probably be able to hit a relatively smooth 60 FPS at 4K as long as you’re willing to turn some of the settings down a notch. The game looked great for being 8 years old, and the high settings really help to sell Los Santos as a living, breathing world. Even though some of the effects can be a bit janky at times, such as the ragdolls, the resolution we tested at combined with the high framerates make this card an easy recommendation if you’re looking to get into Grand Theft Auto. The game was a real treat to experience at these settings, and I can’t wait to see what Rockstar brings to the table next!
Moving into another first-person shooter, and Metro Exodus performed well enough to the point where it delivered a last-gen console killing experience. Even though things weren’t optimal from a performance standpoint, with the average coming in at 61 FPS and a minimum of 49 FPS, it paints a playable, albeit moderately unstable picture. The card delivered a tight grouping of the overall values, indicating there was less variance than what we found on our 1080. All this means is that your framerates will generally hang lower, and they also won’t go as high. Some prefer this, as it leads to less noticeable stuttering when you get erratic framerate jumps, while others might prefer to just have the extra frames. I’m personally in the former camp, so the 1070 delivered a nice experience, even though it was marred by frame drops when particle effects took up a nice chunk of screen space, especially those cobwebs. This brings us back to the memory configuration, and I suspect that we loaded too much for the memory buses to handle. It’s a beautiful game, but it’s probably worth just turning down the resolution to 1080p and calling it a day!
Our third Unreal Engine title, PUBG, returned an incredibly playable average framerate of 93 FPS. The minimum may look kind of rough, however it only occurred during the drop-in and was not present in gunfights. It may not look great, but it was ultimately an outlier in the grand scheme of things. Overall, playing in 1440p allowed us to keep our sights on our enemies at longer ranges, which could be a benefit to you if you’re sniping or trying to take pot shots on another player. It performed well, looked great, and was truly a blast to play! Lowering the resolution to 1080p would also deliver an incredible performance jump, at the cost of some image clarity. Turning settings down is another way to save on processing time, which for PUBG might actually be beneficial thanks to the way it renders vegetation. So far, the 1070 has proven to be very competent when it comes to gaming, and PUBG is just another feather in the card’s cap!
Rainbow Six Siege, our second Vulkan-enabled title, gave us an average of 107 FPS with our 1440p test settings. I tested for this game on DirectX, as I haven’t had time to sit down and compare the Vulkan version of the game, but the general discussion online seems to be pretty neutral on which API you use. Even so, with DirectX our 1070 crushed this game, and the overall e-sports-like design of the graphics really worked in our card’s favor. It’s a simplistic yet effective art style, and that combined with the performance-saving techniques – such as the pseudo-checkerboard MSAA trick – helps our 1070 keep its competitive edge. While these aren’t ultra-high framerates, simply lower the resolution to 1080p and you’re in for an easy 144Hz lock. And if you want to keep the resolution high, turning down the settings isn’t a bad idea either! It was beyond playable at 1440p, so if you’re happy with 60FPS, this game is a treat to experience on the GTX 1070!
Our final game, Call of Duty: Warzone, performed well, but not exceptionally well. With an average of 57 FPS, we aren’t hitting what I would consider to be the minimum for an online competitive FPS, but this is an easy problem to fix. Simply turning down the resolution of the game or turning on the dynamic resolution scaler will allow performance to lock at 60 FPS, and turning down settings wouldn’t hurt either! Warzone did have a problem though, with particle effects causing the game to explosions fill up the screen, but with the rate at which things explode in Call of Duty, it becomes kind of distracting. To remedy this, turning down the particle lighting quality as well as the shadow quality helped to smooth performance out a bit more, although the stutters still persisted. That’s what ultimately dragged our average down, but this once again brings up the memory configuration.
This is ultimately probably what’s holding us back here, with clock speeds being the main issue rather than capacity. Even with a memory overclock – which keep in mind isn’t possible without thermal issues on this card – Warzone just doesn’t perform as nicely as it would at lower resolutions. For the utmost stability and framerates, turning down settings is probably the way to go if you’re thinking about sticking with 1440p, but 1080p with maxed-out settings would also be an excellent experience as well!
Overall, the GTX 1070 is still a very competent graphics card that packs lots of computational power into a silicon chip the size of a nickel. While the GPU cores themselves are powerful and can tackle most modern AAA games with ease, the memory configuration is what’s holding us back. Even though this memory configuration was the gold standard when this card came out, keep in mind that it came out 5 years ago. Most older cards will struggle with high numbers of modern sprites, and this is due more to the clock speed of the memory, rather than the bus widths. On a more modern GDDR6 configuration, we’re now seeing memory chips that run at a minimum of 14Gbps. On this GDDR5 config, we’ve got just 8Gbps. You can probably see now why I’ve been repeatedly bringing up the memory configuration, and it’s mostly because of its age.
As we continue to move forward, GDDR6 will become the standard for video cards which, while a good thing overall, also means that hardware gets left behind. For gaming at 1080p, this card has more than enough bandwidth to shred any modern title, but once the resolution starts to go up, that’s where we start to see some struggling. 1440p, while definitely doable on many of our test games, also crippled our card in 2 of our games (Cyberpunk 2077 & Metro Exodus), leaving me more comfortable recommending this card for 1080p gaming and light work station use. Speaking from personal experience, I’ve been using this card in my render machine for just over a year now, and while it’s not the fastest or newest kid on the block anymore, it puts up one hell of a fight and is able to hold its own in games that should, in theory, not run as well as they actually do.
If we want to make a comparison, imagine trying to run a game from 2015 on a card that came out in 2010; it would run, but it would be incredibly slow. The 1070 thankfully doesn’t suffer from this, and with its lower hash rates, it also is not being targeted by crypto miners quite as frequently. This all ultimately begs the question: is the GTX 1070 worth picking up in 2021? Well, right now with the inflated pricing, not particularly. But as prices come down, so will the price of this card, which in time will be back down to pre-2020 levels. Once that happens, then it’s 110% worth pulling the trigger on one of these cards, despite the negative aspects I’ve discussed at length.
With six and a half teraflops of FP32 compute, this card can stand up against the Xbox Series S and Xbox One X, despite its much older architecture and higher power draw. If you’re able to find a 1070 at or around the $200 mark, then it’s a steal and is definitely worth checking out! It’s an amazing card that offers strong FP32 compute, a high memory capacity, and a sweet spot in terms of efficiency. Keep an eye out for the 1070 if you’re in the market for a graphics card, because one might be lurking nearby!