My GTX1070 vs Titan 6GB comparison (50 games tested)

[BFG10K]

Introduction & Card

After 3.5 years of solid service from my original 6GB Titan, I decided to upgrade to a GTX1070. I got the smallest one possible, the Gigabyte mini-ITX: http://www.gigabyte.com/products/product-page.aspx?pid=5923#kf

The fan stays off based on a combination of temperature and GPU load. At load it’s audible but not irritating, and it’s subjectively quieter than the Titan. The card’s default core clock is slightly above nVidia’s reference clock by 1%-2%.

In case anyone’s interested, the old xS AA modes still work fine on the Pascal board.


Specs & Setup

Here are the two cards’ theoretical specs:

Of particular interest regarding the 1070 is the loss of the 6-pin power connector courtesy of the 40% reduction in TDP, and the massive pixel fillrate advantage. Also note the small deficit in texture fillrate and memory bandwidth.

The games below use the settings I played them at on the Titan. I intentionally try to test worst-case scenarios where possible, so unusually low results don’t reflect the vast majority of actual gameplay.

System specs: 4790K, 16GB DDR3L-1600, Asus Z97-K, 960GB Crucial M500, Seasonic Platinum Fanless 520W, Corsair 400C, 27" Dell P2714H, Win7-64, nVidia driver 375.70.


Results

Conclusion

This little card is a power-house, being 50% faster on average than the Titan. In particular, #27, #28 and #30 are over twice as fast. Also #12 has the fourth highest increase on the list despite being 13 years old.

I’d also expect all of those 1080p games to now be playable at 4K on the 1070 if I turn off AA.

I’m happy with the results and I expect to get 2-3 years of good service from the 1070.

 

[SlickR12345]

Why do so many 2010 games and older play so badly at hardly 60fps?

 

[Head1985]

Was TITAN OC?GTX1070 alreay run at max(only 10% oc headroom).TITAN have 30-40% oc headroom.

 

[Hi-Fi Man]

Why do so many 2010 games and older play so badly at hardly 60fps?

The games are running at 4K.

 

[Piroko]

Why do so many 2010 games and older play so badly at hardly 60fps?

Because they were still single-threaded for the most part and DX9&DX10 aren't helping either. DX11 definitely was a step in the right direction.

 

[BFG10K]

Why do so many 2010 games and older play so badly at hardly 60fps?

Like I said in the OP: I intentionally try to test worst-case scenarios where possible, so unusually low results don’t reflect the vast majority of actual gameplay.

Extremely low scores (e.g. #29) generally only occur one place in the entire game, probably caused by an unoptimized map hole or similar. I certainly didn't play New Vegas at 22FPS the entire game, because that would be silly. My target is 60FPS and my settings are chosen so that for 95+% of any given game, I'll reach it. Then out of the places that drop below it, I benchmark the worst slowdown from those areas.

With newer games it's generally harder to find the most demanding spots and still make the benchmark completely repeatable, so the scores don't tend to drop as low as older games'.

Was TITAN OC?GTX1070 alreay run at max(only 10% oc headroom).TITAN have 30-40% oc headroom.

I don't overclock anything, and Gigabyte's ~1.5% factory overclock will have virtually no impact on the scores.

 

[amenx]

Titan 6gb's value was already quickly diminished with the arrival of GTX 970, which beat or equaled it in most benches.

 

[Zanovar]

Good stuff.*thumb up*

 

[Annisman*]

Awesome testing thanks. I loved my OG Titan, I think I upgraded 3 times from it already though lol

 

[BFG10K]

Titan 6gb's value was already quickly diminished with the arrival of GTX 970, which beat or equaled it in most benches.

Yep, at one third the price no less. Titan -> 970 convinced me it's not worthwhile buying top-end GPUs anymore.

 

[Phynaz]

Nice job man, thanks for this!

 

[poofyhairguy]

Awesome testing thanks. I loved my OG Titan, I think I upgraded 3 times from it already though lol

I was so excited to get the 970 for that reason! "I have what was a $1000 GPU!"

 

[tential]

OP can you talk about 4k gaming in general? I was about to buy a fury x gpu for 4k gaming. Now after seeing your thread I realize there are a number of games i can play that I haven't played yet and enjoy the 4k beauty.

 

[VirtualLarry]

So, bottom-line, the 1070 is a decent 4K gaming card in 2016? Or only for older games? I've just moved up to a 4K@60 display, and am currently using an RX 460 4GB Nitro card, primarily for desktop usage and Skyrim. But looking at possibilities. 1070 looks like a good one. It would be the most expensive video card I think I have ever purchased, if I get one. Hopefully, the price will drop, when the 1080ti is introduced.

 

[[DHT]Osiris]

So, bottom-line, the 1070 is a decent 4K gaming card in 2016? Or only for older games?

Depends on the game. 1070gtx will push 4k @60fps in Overwatch, doubtful it'll do it for a well modded skyrim (I'm getting 45-60 on a 970 @1440p). Older games, it depends as well.

 

[RussianSensation]

So, bottom-line, the 1070 is a decent 4K gaming card in 2016? Or only for older games? I've just moved up to a 4K@60 display, and am currently using an RX 460 4GB Nitro card, primarily for desktop usage and Skyrim. But looking at possibilities. 1070 looks like a good one. It would be the most expensive video card I think I have ever purchased, if I get one. Hopefully, the price will drop, when the 1080ti is introduced.

I would buy GTX1070 and just lower some settings. GTX1080 doesn't have enough horsepower over 1070 for a much better 4K gaming experience in 2016 AAA games, while 1070 SLI demolishes 1080 in 4K where SLI scales. Titan XP is grossly overpriced. You might want to wait a bit based on this news:
http://www.game-debate.com/news/218...ay-release-in-december-competes-with-gtx-1080

BTW, does your i5-6400 scale beyond 4.5Ghz if you bump the voltage to 1.425-1.45V?

 

[VirtualLarry]

BTW, does your i5-6400 scale beyond 4.5Ghz if you bump the voltage to 1.425-1.45V?

I'm afraid not. It does 4.3+Ghz @ 1.300V, needed a touch more than 1.400V. I could get it to boot and run OCCT without errors at 1.400V, but it wouldn't reboot Windows 10 consistently. Bumping up the vcore to 1.410 or 1.420V seemed to cure that, but I'm already a bit concerned about giving it that much voltage. Don't want to go higher. I'm happy with 4.51Ghz. It's a 67% OC, on a lower-end locked quad, which really isn't too shabby.

Edit: BTW, thanks for the analysis and the tip to pick up an i5-6400 and BCLK OC in another thread. So far, it seems to be working out well.

 

[RussianSensation]

Of particular interest regarding the 1070 is the loss of the 6-pin power connector courtesy of the 40% reduction in TDP, and the massive pixel fillrate advantage.

TDP does not equal power usage. A GPU's power usage may be higher, lower or equal to TDP. It depends and more so since AIBs often don't use reference PCB. This is especially true for Maxwell and Pascal era GPUs where NV's AIB power usage exceeds the TDP rating. Gigabyte G1 Gaming 1070 uses 182W of power. It's highly unlikely that your mini-ITX 1070 uses 32W less power than that. The original Titan used about 251W of power from the same site.

Spoiler: TDP of GTX1070 is below actual power usage. You should not assume TDP always = power usage as it leads to erroneous conclusions.

That means many AIB GTX1070 use about 27-30% less power, not 40% less power. It's a small correction, but important to recognize since TDP is often a misleading metric; and by you comparing power usage = TDP, it just continues to perpetuate the TDP = power usage myth.

-------------------

Congrats on the new card! I am surprised you didn't include more modern games from 2013-2016 era. 90%+ of the games you tested have ancient graphics and the OG Titan was already fast enough. You are quickly getting into diminishing returns throwing 2-4x MSAA and sometimes super-sampling at 4K on older games that look outdated and have last gen graphics compared to many modern AAA games where 1070 will shine a lot more.

- Far Cry 4
- Far Cry Primal
- Metal Gear Solid V: The Phantom Pain
- Metro 2033 Redux
- Metro Last Light Redux
- Crysis 2 & 3
- Alien: Isolation
- Doom (2016)
- The Witcher 3
- Gears Of War 4
- Forza Horizon 3
- Rise of the Tomb Raider
- Deus Ex: Mankind Divided
- Fallout 4
- Dying Light
- SW: Battlefront
- Battlefield 1

etc.

This is why your performance increase is "only" 50% on average but it should be much more. You are actually CPU limited in most of those games and the 1070 is under-utilized.

Spoiler: At 1440p, an AIB 1070 is 72% faster than GTX970 (970 is faster than OG Titan), and At 4K, an AIB 1070 is almost 2X faster than GTX970:

Spoiler: Other reviews back this up. Sweclockers has AIB 1070's about 75% faster than GTX780Ti/GTX970 at 1440p, and at 4K, AIB 1070 is 84-85% faster than 780Ti/GTX970, again cards that are faster than the OG Titan.

The point I am making here is an AIB 1070 is much, much faster than 50% over the OG Titan at 1440p and 4K when testing newer games. You have a CPU limitation in your tests for 2 reasons: (1) your processor is a 15-20% slower than an i7 6700K @ 4.7-4.8Ghz, and (2) the vast majority of the games tested are using older generation game engines that do not benefit from compute/modern features of Pascal architectures. The combination of these factors is why your 1070 is "only 50%" faster than your OG Titan, when it should be 85-100% faster. That doesn't mean there is anything wrong with your testing. On the contrary, your testing shows that GTX1070 is a huge overkill for most of those games and its true potential will only be realized in newer 2013-2017 AAA games. If you keep your OG Titan around for the next 2-3 months and have a chance to test some newer AAA titles, you should see 85-100% delta (slightly less due to the i7 4790K).

Spoiler: *** AIB GTX1070 is actually faster than GTX970 SLI (it's roughly equal to R9 295X2) but 970 SLI is 93% faster than the OG Titan.***

Also note the small deficit in texture fillrate and memory bandwidth

This is not correct on both accounts. Careful. You are comparing paper specs which start to lose meaning since Kepler and Pascal are different architectures.

GTX1070 has superior Int8, FP16 andd FP32 texture filtering compared to the OG Titan. In fact, GTX1070 is superior to the Maxwell GTX980Ti in this regard.

Spoiler: OG Titan texture filtering performance

http://techreport.com/review/24381/nvidia-geforce-gtx-titan-reviewed/6

vs. GTX1070 is superior.

Spoiler: GTX1070's texture filtering performance even supersedes GTX980Ti's

http://techreport.com/review/30413/nvidia-geforce-gtx-1070-graphics-card-reviewed/4

The memory bandwidth comparison you made doesn't account for 3rd generation colour compression of Maxwell (33% more effective than Kepler's) and 4th generation colour compression of Pascal (20% more effective than Maxwell's).

Spoiler: Maxwell's Memory Compression = 33% more efficient than Kepler's

Spoiler: Pascal's Memory Compression = 20% more efficient than Maxwell's

To derive "OG Titan" equivalent memory bandwidth, we have to gross up GTX1070's bandwidth by 33% and then by another 20%. To achieve that, we have to divide by 0.75 (for 33% effective) and by 0.833 (for 20% effective):

GTX1070 = 256GB/sec / (0.75 for 3rd gen) / (0.833 for 4th gen) = 410GB/sec

That means GTX1070 has 42% "Kepler equivalent" memory bandwidth than the OG Titan.

In conclusion, your GTX1070 is a lot better than it appears based on your early testing. Enjoy!

I'm afraid not. It does 4.3+Ghz @ 1.300V, needed a touch more than 1.400V. I could get it to boot and run OCCT without errors at 1.400V, but it wouldn't reboot Windows 10 consistently. Bumping up the vcore to 1.410 or 1.420V seemed to cure that, but I'm already a bit concerned about giving it that much voltage. Don't want to go higher. I'm happy with 4.51Ghz. It's a 67% OC, on a lower-end locked quad, which really isn't too shabby.

Edit: BTW, thanks for the analysis and the tip to pick up an i5-6400 and BCLK OC in another thread. So far, it seems to be working out well.

No problem! I am happy it worked out for you. I have to respond in this thread too as it seems some posters didn't understand the point I was making. I was suggesting that if someone cannot stretch the funds to the i7 6700/6700K, and don't have the $ for the i5-6600K, might as well get a Z170 or BLCK capable board and push the i5 to the max. There are a lot of PC gamers who have i5-6400/6500/6600 that just leave them stock which is a wasted opportunity!

 

[tential]

So you're saying for old games for me to enjoy 4k, I need Skylake or kabylake? I plan on getting fury x so that cpu bottleneck will even be worse.... Darn.

 

[RussianSensation]

So you're saying for old games for me to enjoy 4k, I need Skylake or kabylake? I plan on getting fury x so that cpu bottleneck will even be worse.... Darn.

That's not what I am saying. I am saying older games will often not stress the graphics card enough, causing a CPU bottleneck. It doesn't mean you can't enjoy them on a 2600K-4790K OC. If you look at the tests for OP, he used the worst case scenario and often applied Super-Sampling or 2-4xAA at 4K. Using those settings, you can't get 60 FPS on even Titan XP in 2014-2016 games. I would much rather have 2014-2016 graphics with 0xAA, HQ than ancient 2007-2011 games at 4xSSAA. I don't see the point of buying an expensive 4K monitor and $300-400 graphics card to play 8-10 year old games. That's just me.

Hear me out. Let's imagine it's 2008 and you are playing Crysis / Warhead on a GTX275. Those games looked better than ANY game released from 1998-2008.

Forza Horizon 3, The Witcher 3, SW:BF1, BF1, Rise of the Tomb Raider look better than all 95% of the games tested, but you can't play those games with 2-4xSSAA at 4K @ 80-100 FPS. So what's the solution, wait 6-10 years to play them?! An older looking game with 16xSSAA will not look as good as the best looking modern 2015-2017 AAA games with minimal AA and HQ instead of UQ.

 

[BFG10K]

TDP does not equal power usage. A GPU's power usage may be higher, lower or equal to TDP.

It's true that TDP is not exactly the same power usage, but it's close enough to not matter:

https://tpucdn.com/reviews/NVIDIA/GeForce_GTX_1070/images/power_maximum.png
https://tpucdn.com/reviews/NVIDIA/GeForce_GTX_Titan/images/power_maximum.gif

Max 1070 = 154W, Max Titan = 263W. So if anything I was being generous to the Titan.

I am surprised you didn't include more modern games from 2013-2016 era.

I benchmark the games I play. If any are missing from last time it's because I never really liked them from a gameplay perspective, and they've since been deleted. In time newer games will be added at my own pace.

You are actually CPU limited in most of those games and the 1070 is under-utilized.

This is demonstrably false. Aside from a few outliers (e.g. #22, #35), the observed GPU load was 95+%. That and dropping the resolution raised performance significantly in almost every case, which again demonstrates a GPU limitation.

This includes #32 which got the lowest performance gain out of the lot (8%). Here you go, this is the 1070 with a constant 99% GPU load in Fear 1, measured right now:

Fear is actually an interesting case:

1. 580 to 680 = -4%.
2. 680 to Titan = +53%.
3. Titan to 1070 = +8%.

Looking at all four cards' theoretical specs, it looks like the game is bottlenecked primarily by GPU memory bandwidth.

The fact is different games stress GPUs differently (especially older games), so a small performance gain doesn't necessarily mean you're CPU limited.

This is not correct on both accounts. Careful. You are comparing paper specs which start to lose meaning since Kepler and Pascal are different architectures.

It was a simple chart of theoretical specs, not an architectural deep-dive. You may as well complain that clock speed doesn't factor IPC.

That's why I prefaced the chart with Here are the two cards’ theoretical specs.

Listed specs were good enough to diagnose what's happening with Fear, which can make them useful for explaining outliers.

 

[BFG10K]

OP can you talk about 4k gaming in general? I was about to buy a fury x gpu for 4k gaming. Now after seeing your thread I realize there are a number of games i can play that I haven't played yet and enjoy the 4k beauty.

Using my original settings above and assuming a 60FPS target, 95% of the time you'll reach it on a Titan. For the 1080p games above, I've just finished testing the 1070, and it can handle all of them where the Titan couldn't.

For demanding games you'll need to disable AA but that's fine since 4K on 1080p is 4xSSAA anyway, which is high-end IQ. You'll also need to disable performance-sucking "features" such as SSAO, DOF, and the fanciest shadows, but that's also fine since they seldom add much to IQ. Most of the time you can only see a difference in still screenshots.

People were also asking about Skyrim. On the 1070 I have the Legendary edition of the game (no mods) with max details (except FXAA off and 2xAA), and it never drops below 60FPS, including the demanding areas where the Titan dropped lower. The Titan was also using 0xAA so the 1070 is faster despite using 2xAA.

So you're saying for old games for me to enjoy 4k, I need Skylake or kabylake? I plan on getting fury x so that cpu bottleneck will even be worse.... Darn.

No, this is absolute nonsense. Like to said to Russian to earlier, almost every result above is GPU limited on both cards. CPU limitations are always overblown and any half decent one will be fine.

I compared a 4790K to a 2500K (both at stock) under similar tested conditions, and the overall performance gain was just 6%:

• http://www.portvapes.co.uk/?id=Latest-exam-1Z0-876-Dumps&exid=threads/geforce-titan-2500k-vs-4790k-66-games-tested.2389580/

Now replace the Titan with a 1070 and you get 50% gain overall, as shown above. So is there even a question of where to spend your money?

If someone gave me the choice between a 2500k + 1070 and a 4790K + Titan, I'd take the first option with no hesitation because 90% of the time I'll get better performance.

 

[PigSkinWytBoy]

[redacted]

rasism comments are not allowed
Markfw anandtech moderator

 

[Red Hawk]

I don't know what your deal is, but this is not a thread for talking about social justice stuff and your dislike of it.

 

[Head1985]

Depends on the game. 1070gtx will push 4k @60fps in Overwatch, doubtful it'll do it for a well modded skyrim (I'm getting 45-60 on a 970 @1440p). Older games, it depends as well.

I am running GTX1070 2140/9500 and its decent 1440p card(still cant manage 1440p/60fps in demanding games).Nowhere near good for 4k.Moded skyrim SSE with ENB i have around 35fps in 4k.
GTX1080 is around 30% faster at same clock than GTX1070 in 4k so lets say i have 35fps and GTX1080 will have 46Fps in same scene.Its big jump but not really 4k/60fps..GP102 will be 4k/60fps card.