My 2070 vs 1070 comparison (28 games)

[BFG10K]

Introduction

After 3 years of solid service from my 1070 I picked up a 2070 as an upgrade. Here are the two cards:

https://www.gigabyte.com/Graphics-Card/GV-N1070IXOC-8GD#kf
https://www.gigabyte.com/Graphics-Card/GV-N2070IX-8GC#kf

I like mini cards. They offer fantastic performance in a very small footprint. Unfortunately while the 1070 was pretty quiet, the 2070 sounds like a vacuum cleaner in comparison. I don’t think this style of form factor can cope with a hot Turing chip.

Spoiler

Aside from the huge memory bandwidth advantage, the 2070’s other specs are mediocre relative to the 1070. But we’ll see how the cards compare in real benchmarks.


Setup

4790K | 16GB DDR3L-1600 | Asus Z97-K | 960GB Crucial M500 | Seasonic Titanium Fanless 600W | Corsair 400C | 27" Acer XB271HU |  nVidia driver 431.36 | Win7-64 (Oct 2019 rollup).

I’ve been targeting 1440p up until now but I plan on buying a 4K gaming monitor soon. So to test the effect of the 2070 upgrade, I set my games to 4K DSR but left their other settings as-is, with the exception of disabling anti-aliasing across the board.

My target is 75FPS locked to gsync and these are the worst-case scenarios for the games. So if they run fine here, they’re a solved problem for me.


Results

Spoiler

Comments

I was expecting about 25% overall so this is better than I thought. Some of those creaky old games run over 40% faster.

#10 is my slowest game and I badly needed extra performance there. It’s still too slow for 4K but it also now runs much better at 1440p as well.

Historically #13 seems to be memory bandwidth bound, and this continues to show here.

#7 - can it run Crysis at 4K? No, some settings are using high instead of very high.

I’ve returned the card because it’s unbearably loud. Also my motherboard and PSU both died a day after installing it! It’s probably a coincidence as the store stress-tested the card for 2 hours without issue, but I didn’t want to take the risk. I still plan on upgrading my GPU soon so I’ll keep an eye out for more deals. I’ll have more numbers from my new system when I do.


New System Results

I’ve retested the games on the new system with the 1070. The specs that have changed are below:

9600KF | 32GB DDR4-2666 | Asus B3650M-K | EVGA Supernova G3 Gold 550W | Win10-64 (May 2019)

Spoiler

#8 seems to slow down related to Windows 10 running DX8/7 games in borderless window mode. I’ve also noticed some games have higher input lag compared to 7, though this problem can be reduced by disabling full screen optimizations. Aside from that it seems to run all my old games just fine, which is a huge relief. Classics like C&C Renegade, Quake 3 and No One Lives Forever run just like they did on 7.

 

[psolord]

Thanks for sharing ing.

It's very interesting for me, as a 1070 user. However I aim for >50% upgrades for my graphics cards, so the 2070 does not cut it. The 2070s is much better but too expensive.

Oh well, next round I guess.

 

[Hitman928]

Some of those older games seem pretty low fps wise. Are you applying mods or heavy AA as well as the 4K resolution?

 

[Arkaign]

I know you're an old pro and may have already done so, but have you checked the thermal compound application and transfer pads? I've yet to find a card that I can't improve on at least a little bit with a few minutes effort and the right supplies. Sometimes the difference is quite small, but occasionally the difference is quite substantial in heat, therefore noise (or increased clock range).

I think with things as competitive as they are, some corners can be cut for that extra little bit of profit, and apparently Shenzhen doesn't have the glut of cheap employees they once enjoyed. Some model for model replacements the quality has taken a serious nosedive.

Look at the Gigabyte Aorus 1080ti Gaming vs the Gigabyte Aorus 2080ti Gaming, the decline was stunning. I ended up AMP Xtreme as it was far more robust, like the Pascal Aorus had been. But I was still able to drop the temps at stock clocks about 6C just by reapplication of high quality compound and premium thermal pads.

 

[GodisanAtheist]

Some of those older games seem pretty low fps wise. Are you applying mods or heavy AA as well as the 4K resolution?

- I wonder if, in a sort of counter intuitive way, older games runworse on modern hardware because they don't benefit from a lot of the driver tricks and architectural shortcuts new cards use to squeeze more performance.

Rendering 4k resolution on an old brute force engine that doesn't have all sorts of screenspace only gimmicks, modern texture streaming, using an old inefficient dx revision etc you end up running into bottlenecks that modern software was designed to work around.

Or the game is just massive bottlenecked by single core ipc...

 

[BFG10K]

Some of those older games seem pretty low fps wise. Are you applying mods or heavy AA as well as the 4K resolution?

No. A lot of older games (especially DX8/DX7) don't scale well to higher resolutions due to engine and/or API bottlenecks. Trying to go above 2560x1440 in those situations either results in a severe performance loss or outright crashing.

Also Windows 10 runs DX8.x (and earlier) games in borderless windowed mode so there's a further performance hit compared to Windows 7 (not applicable to my Win7 results above).

Plus the tests above are the slowest points in the game. I do this intentionally because if my scores are good above, I don't have to worry about the rest of the game.

 

[Hitman928]

No. A lot of older games (especially DX8/DX7) don't scale well to higher resolutions due to engine and/or API bottlenecks. Trying to go above 2560x1440 in those situations either results in a severe performance loss or outright crashing.

Also Windows 10 runs DX8.x (and earlier) games in borderless windowed mode so there's a further performance hit compared to Windows 7 (not applicable to my Win7 results above).

Plus the tests above are the slowest points in the game. I do this intentionally because if my scores are good above, I don't have to worry about the rest of the game.

Can you tell me where you're testing for Oblivian and Bioshock 2?

 

[mopardude87]

Nice numbers and thanks for the hard work putting that together op Was the 4790k at stock by chance ? I got a spare rig with a 4670 that MAY get a 4790 non k in the future. Still doing research if the 4790 non k can handle BF1/BF5 as i will be pairing it with a 3gb 1060 soon. Might even upgrade the rig to a 1070/1080 gpu as 3gb on the 1060 can't do max anyways. I know for 1080p anyways a gtx1080 would be overkill for the 4670.

If the cpu tanks in either of those titles, the 1060 will prob be the last stop for the 4670 and the only upgrade for that rig will eventually be my 8700 non k when its upgraded years down the road. Most demanding game i play that runs good on 4c/4t is BF4 and the 1060 with a mild oc JUST fits the bill for a 60fps minimum experience on max settings at 1080p.

 

[Arkaign]

Nice numbers and thanks for the hard work putting that together op Was the 4790k at stock by chance ? I got a spare rig with a 4670 that MAY get a 4790 non k in the future. Still doing research if the 4790 non k can handle BF1/BF5 as i will be pairing it with a 3gb 1060 soon. Might even upgrade the rig to a 1070/1080 gpu as 3gb on the 1060 can't do max anyways. I know for 1080p anyways a gtx1080 would be overkill for the 4670.

If the cpu tanks in either of those titles, the 1060 will prob be the last stop for the 4670 and the only upgrade for that rig will eventually be my 8700 non k when its upgraded years down the road. Most demanding game i play that runs good on 4c/4t is BF4 and the 1060 with a mild oc JUST fits the bill for a 60fps minimum experience on max settings at 1080p.

4790 @ stock will run BFV nicely if you have a 60hz or 75hz display. From my experience a 6GB 1060 or an RX580 with a mix of high and ultra settings is perfect for that level of performance.

BFV is really happy with lots of threads. I got a buddy an S30 with a Xeon 1650 v2 (6C/12T), and he now has a 1070ti, which he uses to run 1440p 144hz BFV. With a mild OC using the Intel tool (1650 v2 is unlocked!) he is seeing 100+ steady, which is absolutely unbelievable value for the system. We got the whole box, with CPU, 2TB HDD, Win7 Pro (use for W10 pro legal activation), 16GB Ram, and 600-something power supply for less than $200. Just added an SSD and GPU.

 

[Furious_Styles]

4790 @ stock will run BFV nicely if you have a 60hz or 75hz display. From my experience a 6GB 1060 or an RX580 with a mix of high and ultra settings is perfect for that level of performance.

BFV is really happy with lots of threads. I got a buddy an S30 with a Xeon 1650 v2 (6C/12T), and he now has a 1070ti, which he uses to run 1440p 144hz BFV. With a mild OC using the Intel tool (1650 v2 is unlocked!) he is seeing 100+ steady, which is absolutely unbelievable value for the system. We got the whole box, with CPU, 2TB HDD, Win7 Pro (use for W10 pro legal activation), 16GB Ram, and 600-something power supply for less than $200. Just added an SSD and GPU.

Hell of a deal there!

 

[mopardude87]

We got the whole box, with CPU, 2TB HDD, Win7 Pro (use for W10 pro legal activation), 16GB Ram, and 600-something power supply for less than $200. Just added an SSD and GPU.

Bargain for sure.

 

[BFG10K]

FYI I've added more results to the OP using the new system with the 1070.

I know you're an old pro and may have already done so, but have you checked the thermal compound application and transfer pads?

Yeah...no. Checking thermal paste is fine on a $25 CPU cooler but I won't be doing that on a $400 graphics card. It's not my job to finish their manufacturing process on an already overpriced part.

Was the 4790k at stock by chance ?

It was but it also had turbo core enabled due to XMP, so all four cores always turbo'd to 4.4GHz regardless of load.

Can you tell me where you're testing for Oblivian and Bioshock 2?

For Oblivion I used the Gold Coast area and for Bioshock 2 I used the final prison area.

 

[Arkaign]

Dang, well, if you ever feel like your temps or noise are more than you'd like, it's worth a look. As long as you have the supplies handy, it only takes a few minutes and can have great results.

Quality coming out of GPU I think peaked in the Kepler to Pascal days, things are really getting shaky these days. I think it's because of the shortage of workers in Shenzhen now. They're cutting corners, which results in seemingly obvious errors being made, which pass as long as they can call it 'good enough'. Some models are especially egregious, like the RX 5700 in the video above. RTX cards are also seeing improvements with repasting/padding. I'm also seeing similar decline in overall quality even for VERY expensive cards. Look up the Gigabyte Aorus 1080ti vs Aorus 2080ti. They are really pinching pennies there and downgrading the quality of the HSF and even PCB and backplate.

 

[Hitman928]

For Oblivion I used the Gold Coast area and for Bioshock 2 I used the final prison area.

Can you by chance record a snippet of your benchmark session in the Gold Coast? I loaded it up and riding around on a horse with 3rd person view and even doing a little bit of fighting I never seem to drop as low as your avg fps in that area. Just curious if we're really running the same thing though. Also, what settings are you using?

 

[IronWing]

I moved from a 1070 to a 2070s, both EVGA open air cards in quiet mid-towers. The 2070s fans barely kick on while gaming. The game that made my old system with the 1070 card stutter and scream is Obduction, far more than more intense games like Doom. I’m not sure what the issue was as the graphics don’t appear to be anything special, maybe poor optimizations. On the new system with the 2070s card, the case fans make a quiet hum during play. I guess I should open the case to see if the card fans spin though opening the case changes the airflow.

 

[Stuka87]

Cool comparison for a lot of games. The new and old system one was very interesting with the older games you used.

 

[CP5670]

Thanks for doing this. It looks like the newer CPU helps a bit but only in the newest games that are more multithreaded. I also found that Dishonored 2 struggled for me in one or two maps, even when there wasn't much going on in view, but the rest of the game ran nicely. I plan to get a Comet Lake once they are out next year. For keeping the video card noise down, you're generally better off with the big 2 or 3 slot cards with multiple fans. My FTW3 cooler is inaudible over my other fans, while various Nvidia blower cards I had in the past had a clear ramp-up in games.

I've found that many DX7/8 games actually run better on 10 than they did on 7 or 8.1, both in terms of performance and bugs/stability. Wrappers like dgVoodoo help in some of them. I used to keep an XP install around just in case but no longer need it for anything.

 

[BFG10K]

In the end I got this one instead: https://www.gigabyte.com/Graphics-Card/GV-N206SGAMING-OC-8GC#kf

It's a tad slower overall than the 2070, but it's far quieter. Subjectively I think it's even quieter than the 1070 which was already pretty good.

Also it's got basic cycling RGB without any software, which is nice. My keyboard is also set to cycle RGB colors.

I also had a chance to sit down and play a bit of Crysis Warhead and Dishonored 2 at 1440p, and the performance gain is quite significant over the 1070. Both games play much better on the new card.

Can you by chance record a snippet of your benchmark session in the Gold Coast? I loaded it up and riding around on a horse with 3rd person view and even doing a little bit of fighting I never seem to drop as low as your avg fps in that area. Just curious if we're really running the same thing though. Also, what settings are you using?

Try first person view outside Fort Sutch. I max all settings possible except grass shadows, and character shadows which look weird.

 

[lobz]

In the end I got this one instead: https://www.gigabyte.com/Graphics-Card/GV-N208SGAMING-OC-8GC-rev-20#kf.

It's a tad slower overall than the 2070, but it's far quieter. Subjectively I think it's even quieter than the 1070 which was already pretty good.

Also it's got basic cycling RGB without any software, which is nice. My keyboard is also set to cycle RGB colors.

I also had a chance to sit down and play a bit of Crysis Warhead and Dishonored 2 at 1440p, and the performance gain is quite significant over the 1070. Both games play much better on the new card.


Try first person view outside Fort Sutch. I max all settings possible except grass shadows, and character shadows which look weird.

What did I miss? You linked in a 2080 Super, how can that possibly be any slower than a 2070??

 

[BFG10K]

What did I miss? You linked in a 2080 Super, how can that possibly be any slower than a 2070??

Sorry, link fixed, should've been 2060 Super.

 

[StrangerGuy]

I sold off my 1070 for $160 and replaced with a $350 2060S after seeing how badly Pascal performed in the new CoD and RDR2 (RX580 on parity with 1070).

Granted I don't play any of the both but I reckoned future games are going to be even less forgiving on Pascal GPUs, so might as well offload it while there's still resale value along with a great deal on my 2060S.

 

[tviceman]

The Turing models with RT and tensor cores are just too power hungry. The efficiency improvements weren't there for traditional rasterization. 7nm Ampere should represent a significant leap forward in efficiency and ray tracing. Those cards will possess the power/efficiency and RT capabilities (at least among the higher end ampere) to really drive next gen. The problem though.... price. The price creeps with Maxwell, then Pascal, and more substantially with Turing along with AMD's new higher pricing are indicative of what to expect from Ampere.

 

[lobz]

The Turing models with RT and tensor cores are just too power hungry. The efficiency improvements weren't there for traditional rasterization. 7nm Ampere should represent a significant leap forward in efficiency and ray tracing. Those cards will possess the power/efficiency and RT capabilities (at least among the higher end ampere) to really drive next gen. The problem though.... price. The price creeps with Maxwell, then Pascal, and more substantially with Turing along with AMD's new higher pricing are indicative of what to expect from Ampere.

And you know this from who exactly? I mean you present all these things about Ampere, like they were facts.

 

[tviceman]

And you know this from who exactly? I mean you present all these things about Ampere, like they were facts.

Let's not start a war of words. Have a beer and relax. I was predicting what Nvidia may be able to do with respect to performance and price based on their leaps in efficiency and performance with Kepler, Maxwell, and Pascal (node jumps with Kepler and Pascal).

 

[aleader]

Interesting comparison that answered a question I'm having right now with my system that I bought used almost 3 years ago for $200 CDN, not including the GPU (i5 4670k @4.4GHz - stock fan, EVGA 1070, 16GB HyperX Fury 1866 Ram, ASUS Z97 Pro-Gamer, W10). Almost every benchmark I see testing new CPUs does not include older systems like mine for comparison, which to me becomes a useless benchmark for upgraders.

Ram and CPUs are so cheap right now I was thinking of an upgrade, but I suspect the in-game difference may not even be noticeable, which to me is a total waste. I play at 1440p/75Hz/FreeSync, mostly DCS World (system HOG), IL2 BOS, Squad, SB Pro, Transport Fever, Post Scriptum, all pretty hardware-hungry games. I can run them at 55 - 75fps without stuttering on High-Ultra depending on the game (DCS with SMAA drops my frames by about 15fps). I would be very interested to see a comparison with something like DCS World between your 1070 and 2070s (it's a free download).

I would like to get a 2060 or 2070 Super to maybe push DCS higher and possibly get into VR. What's holding me back is the 5700XT and it's drivers. I decided to wait until next year to see how/if AMD improves them, and as mentioned, what the new consoles do to the market/prices. The 5700XT is almost $200 CDN cheaper (Newegg) than a 2070 Super...huge difference.

Anyways, thanks for the work that you did here.