Geforce GTX 1060 Thread: faster than RX 480, 120W, $249

[Sweepr]

Nah, I leave it to the professionals:

Right, I forgot only the Golem results you copy and paste in all NVIDIA threads to prove your point count as 'from professionals'. Meanwhile here's some Guru3D, Hardware Canucks, ComptoirduHardware, Hardware Unboxed, AnandTech for you:

RX 480 3% faster (7 DX12 games + Doom Vulkan)

GTX 1060 only 4% slower

AMD leads big (15-20%) in Forza 6

Wrong, GTX 1060 is actually faster in some websites:

Hitman

Wrong, 12% according to AnandTech.

Quantum Break

Actually 12% @ HardwareCanucks.

and biggest in Doom (~25%).

The game that uses GCN Shader Extensions + Async Compute exclusively on AMD cards. Yet your 25% is still misleading:

There goes the imaginary '15% overall'.

 

[bystander36]

I'm pretty certain the settings used will play a big factor in the results. If no AA is used, the 480 will fair better than if AA is used.

 

[parkerface]

I can't help but giggle as these volleys take place.

I definitely appreciate the wealth of info presented, but maaaan... If only I could look over some shoulders as you go back and forth, scouring the interwebs in frustration just to support a narrative. All of it comes across as borderline zealotry and that isn't healthy behavior.

 

[HiroThreading]

I can't help but giggle as these volleys take place.

I definitely appreciate the wealth of info presented, but maaaan... If only I could look over some shoulders as you go back and forth, scouring the interwebs in frustration just to support a narrative. All of it comes across as borderline zealotry and that isn't healthy behavior.

I disagree.

It's a healthy discussion where they're focusing on the topic and not resorting to personal attacks.

It's also very informative for someone like me when I recommend hardware to friends and whatnot.

 

[HiroThreading]

By the time 2 GB of VRAM isn't enough, 7970 and gtx 680 class cards will be woefully under-powered.

By the time 3 GB of VRAM isn't enough, 290x and 780 ti will be woefully under-powered....

Oh wait!

Except those GPUs brought a big leap in compute performance, and there was an expectation that VRAM would become a bottleneck.

The 480 and 1060 don't even match Hawaii in terms of GPU performance.

 

[parkerface]

point missed entirely

 

[crisium]

Except those GPUs brought a big leap in compute performance, and there was an expectation that VRAM would become a bottleneck.

The 480 and 1060 don't even match Hawaii in terms of GPU performance.

Developers will optimize around what's becoming more common in the market. Last gen isn't as important as you think come this fall and next year.

4GB already has a few issues at 1440p, and you know that's gonna get worse. 1080 8GB, 1070 8GB are the new 1440p cards, and developers will see less reason to get below 4GB there. Even the lower tier cards are the 1060 6GB and 480 8GB. I'm already bracing for "High instead of Ultra" textures on many games this fall/winter.

1080P won't be that far behind in usage - it's never drastically.

We've seen this before. At launch 7850 was slower than the 1.25GB GTX 570. And the 1GB 7850 seemed pretty fine at 1050p/1080p in 2012 IIRC. But certainly in 2013+ the 2GB really benefited. I can attest with using more than 1GB at times on my GTX 460 equivalent which is a much slower card.

Why did this happen? Because all the top 4 GPUs in 2012 for each company were 2-3GB minimum. So developers saw less and less need for keeping VRAM below 1GB, or even 1.5GB (570-580 started to feel the pain too). And the Top 4 Nvidia GPUs are 6-12GB, and the top 2 AMD Gpus (which are equal to the 4th and likely 5th for Nvidia) also have 8GB options available. As the 980 and 970 ownership starts to fade away a bit, there will be less desire for <4GB textures on Ultra 1080p. Not the biggest deal to run a texture setting lower, but something to consider.

1060 6GB should be fine for quite some time though, although if you hold onto it long enough the day will come that 1080p textures will require more than 6GB, but you'd probably have the other settings at Medium or something and would want to upgrade by that point anyway.

That's why as I posted earlier I find the choices for $150-$250 (MSRP anyway) to be rather fascinating with VRAM considerations. If your max was $200, what would you take? Honestly, I'd recommend either going under budget for the 470 4GB and upgrading sooner with the savings, or going 1060 / 480 8GB (depending on preferences) and riding it out longer.

We still need 470 and 1060 3GB (1050?) benchies, but I anticipate that being my recommendation. $150/$250, and round up or down if your budget is in between. Looking forward to 1050 and 470 reviews to see how things are looking.

 

[littleg]

I think with vram you need to take the overall performance of the card into account aswell. I probably wouldn't be happy with a top end card only having 6GB right now as I believe the GPU should be capable of pushing to a level where that quantity of vram will be limiting.

For the 1060 / 480 class of GPUs, though, 6GB is absolutely fine. Nobody is going to be buying these cards to game a 4k resolutions or push ultra detailed textures and gobs of AA. 6GB, maybe even 4GB, is going to be fine for this class of performance for the next few years at least. In my opinion of course, i'm not Nostradamus

 

[Silverforce11]

The game that uses GCN Shader Extensions + Async Compute exclusively on AMD cards. Yet your 25% is still misleading:

There goes the imaginary '15% overall'.

~25%, spot on.

But if it makes you happier, the 15% overall can be 12%.

 

[raghu78]

GTX 1060 only 4% slower

The only DX12 game Nvidia won big was RoTR. Now that lead has been cut or even erased.

http://hexus.net/tech/reviews/graphics/94969-sapphire-radeon-rx-480-nitro-4gb-8gb-oc/?page=10

https://www.youtube.com/watch?v=7ay4rSDKy2M

RoTR was the only game saving them from a larger lead for Rx 480 in DX12 perf avg. Now that too is gone. The next time those avg nos are going to show a lead closer to 7-8%. btw hardwarecanucks comfortably left out Doom Vulkan. That would move the lead well above 10%.

The game that uses GCN Shader Extensions + Async Compute exclusively on AMD cards. Yet your 25% is still misleading:

There goes the imaginary '15% overall'.

dude multiple sites are showing Rx 480 thrashing GTX 1060 in Doom Vulkan by 15-30%.

http://www.hardocp.com/article/2016...x_1060_founders_edition_review/4#.V5vuLoMrLIU

http://hexus.net/tech/reviews/graphics/94969-sapphire-radeon-rx-480-nitro-4gb-8gb-oc/?page=7

https://www.youtube.com/watch?v=wbR9d6aF8iA

http://www.pcgameshardware.de/Doom-2016-Spiel-56369/Specials/Vulkan-Benchmarks-Frametimes-1202711/

(look at Vulkan adjusted fps in pcgh)

 

[zinfamous]

With the latest driver, AMD's weak performance in Rise of the Tomb Raider is gone. They've reached parity.

https://youtu.be/ECTGKqWpszk?t=2m25s

^ Some sites find close performance, this guy is a pretty big youtuber, he found the RX 480 was actually faster than the 1060 in Rise now.

Ashes is a tie. Warhammer is about 5-10% lead. AMD leads big (15-20%) in Forza 6, Hitman, Quantum Break and biggest in Doom (~25%). The AVG lead overall is ~15% already.

We'll see how the next wave of big DX12 games turn out, then we can say with a high level of certainty.

But I find it odd NV isn't sponsoring more DX12 games, make more like Rise of the Tomb Raider where AMD runs it gimped. What's going on?

Crap! I guess we will learn that RotR is now an AMD sponsored game and it is only fair to toss it out of benchmarks. :\

Dang, I thought OP would have at least 6 months before needing to update the thread title, but that might be much sooner than anticipated.

 

[HiroThreading]

Developers will optimize around what's becoming more common in the market. Last gen isn't as important as you think come this fall and next year.

4GB already has a few issues at 1440p, and you know that's gonna get worse. 1080 8GB, 1070 8GB are the new 1440p cards, and developers will see less reason to get below 4GB there. Even the lower tier cards are the 1060 6GB and 480 8GB. I'm already bracing for "High instead of Ultra" textures on many games this fall/winter.

1080P won't be that far behind in usage - it's never drastically.

We've seen this before. At launch 7850 was slower than the 1.25GB GTX 570. And the 1GB 7850 seemed pretty fine at 1050/1080 in 2012 IIRC. But certainly in 2013+ the 2GB really benefited. I can attest with using more than 1GB at times on my GTX 460 equivalent which is a much slower card.

Why did this happen? Because all the top 4 GPUs in 2012 for each company were 2-3GB minimum. So developers saw less and less need for keeping VRAM below 1GB, or even 1.5GB (570-580 started to feel the pain too). And the Top 4 Nvidia GPUs are 6-12GB, and the top 2 AMD Gpus (which are equal to the 4th and likely 5th for Nvidia) also have 8GB options available. As the 980 and 970 ownership starts to fade away a bit, there will be less desire for <4GB textures on Ultra 1080p. Not the biggest deal to run a texture setting lower, but something to consider.

1060 6GB should be fine for quite some time though, although if you hold onto it long enough the day will come that 1080p textures will require more than 6GB, but you'd probably have the other settings at Medium or something and would want to upgrade by that point anyway.

That's why as I posted earlier I find the choices for $150-$250 (MSRP anyway) to be rather fascinating with VRAM considerations. If your max was $200, what would you take? Honestly, I'd recommend either going under budget for the 470 4GB and upgrading sooner with the savings, or going 1060 / 480 8GB (depending on preferences) and riding it out longer.

We still need 470 and 1060 3GB (1050?) benchies, but I anticipate that being my recommendation. $150/$250, and round up or down if your budget is in between. Looking forward to 1050 and 470 reviews to see how things are looking.

You can't simply increase VRAM usage without putting some sort of pressure on the GPU (TMUs or ROPs, for example).

As Arachnotronic eluded to, and the point I was making, these GPUs will hit bottlenecks in terms of compute performance first before they hit a VRAM limit.

In regards to your point about developers targeting the largest market possible, that's GPUs in the AMD 280-390 class and Nvidia 960-970 class. With the exception of the 390, those GPUs don't have large amounts of VRAM. Developers can't increase VRAM requirements/usage without a corresponding leap in GPU performance.

If we take a historical look at GPU performance and VRAM requirements, we see that in most cases the GPU is the one that starts to tap out first. Exceptions to this are Nvidia who typically under-equip their top end GPUs.

Actually, I would say that the GPU architects at AMD do a very good job in forecasting the optimal amount of VRAM to match their GPU performance. They sometimes get it wrong though (e.g. A optional 1GB allocation to R600 -- not to mention the 512 bit ring-bus memory controller).

 

[Sweepr]

~25%, spot on.

Not so spot on depending on where you look (1, 2). And we all know that's because of the GCN Shader Extensions + Async Compute only on AMD, so a very biased outlier.

But if it makes you happier, the 15% overall can be 12%.

Can also be less than that. Single digits, and that includes the AMD Gaming Evolved titles.

 

[zinfamous]

You can't simply increase VRAM usage without putting some sort of pressure on the GPU (TMUs or ROPs, for example).

As Arachnotronic eluded to, and the point I was making, these GPUs will hit bottlenecks in terms of compute performance first before they hit a VRAM limit.

In regards to your point about developers targeting the largest market possible, that's GPUs in the AMD 280-390 class and Nvidia 960-970 class. With the exception of the 390, those GPUs don't have large amounts of VRAM. Developers can't increase VRAM requirements/usage without a corresponding leap in GPU performance.

If we take a historical look at GPU performance and VRAM requirements, we see that in most cases the GPU is the one that starts to tap out first. Exceptions to this are Nvidia who typically under-equip their top end GPUs.

Actually, I would say that the GPU architects at AMD do a very good job in forecasting the optimal amount of VRAM to match their GPU performance. They sometimes get it wrong though (e.g. A optional 1GB allocation to R600 -- not to mention the 512 bit ring-bus memory controller).

I feel like we will soon be hitting 8gb as a lower limit and 12gb will be the standard-midrange.

It also appears that with current gen and earlier, higher clocked memory and/or wider bus can push a lot of overall performance on the card. Wouldn't it make sense to shift focus towards faster/wider memory before going bonkers with gb? Or is that a more complicated and costlier solution?

 

[crisium]

You can't simply increase VRAM usage without putting some sort of pressure on the GPU (TMUs or ROPs, for example).

As Arachnotronic eluded to, and the point I was making, these GPUs will hit bottlenecks in terms of compute performance first before they hit a VRAM limit.

While I'm sure it has more of an affect on weaker cards, changing texture settings exclusively often has a very minor performance penalty. It is also the setting most tied to VRAM usage, naturally.

So until I see otherwise I stick by what I say. People tend to look at the whole presets such as "Ultra everything" vs "High everything" and of course whenever Ultra anything uses more than 4GB there's a strong chance a 480 class card can't run it at playable framerates anyway.

I'm speaking of Ultra textures and High everything else. Or High textures and Medium everything else. In my own experiences this is only a few % penalty and thus close to free if you had a higher VRAM GPU.

I still want to see more testing of modern games and GPUs though, as my experience is mostly Fermi 2GB with ~2013 games. What are some of the weaker 4GB cards, 960/380 class? I'd want to see a user find a game where Ultra preset is at least 2GB of VRAM, but is generally unplayable. Then I want to see what preset takes it below 2GB (say, High) and see those framerates. Then see if High everything except Ultra textures has a meaningful penalty. This is the best way to see if VRAM is worth it in cards that have choices - not just straight presets, imo. Any takers to test? Probably not many 960/380 users here though =/.

 

[raghu78]

Not so spot on depending on where you look. And we all know that's of the GCN Shader Instructions + Async Compute only on AMD, so a very biased outlier.

dude get used to GCN shader intrinsics being used more and more in AAA games as its a part of GPUOpen which is designed to bring console optimizations to PCs. Developers do not need to spend extra effort as AMD is enabling to bring their console optimization effort onto PC. GPUOpen is AMD's answer to Gameworks except that its not built to harm the competition. Its built to extract the maximum performance out of AMD's GPUs.

http://gpuopen.com/welcometogpuopen/
http://gpuopen.com/gcn-shader-extensions-for-direct3d-and-vulkan/
https://en.wikipedia.org/wiki/GPUOpen

Can also be less than that, according to the websites I linked. Single digits, and that includes the AMD Gaming Evolved titles.

Keep on with your 2017 mantra while others enjoy the fact that GTX 1060 is faster overall today, draws less power and has plenty of nice custom models available (starting from $249).

On the contrary with Rx 480 people are enjoying excellent performance in 2016 DX12 / Vulkan games and competitive performance in older DX11 games from 2013-2015. The way I see it AMD has built a card for the present and the future while Nvidia has built it for the past (DX11). We are going to see review sites add upcoming games like Deus Ex Mankind Divided, Battlefield 1, Watch Dogs 2, Civilization 6, Gears of War 4, Forza Horizon 3. By year end the competitive landscape is going to change. 2017 will only accelerate the DX12 transition. Meanwhile AMD card users are enjoying outstanding DX12 / Vulkan performance on R9 390X / R9 390 / R9 290X / R9 290 while the GTX 980 and GTX 970 have fallen off a cliff. The current generation is going to be a repeat of the last gen as AMD cards age very well while Nvidia cards lose appeal quickly. Look at GTX 600 vs HD 7000 , GTX 700 vs R9 200 and GTX 900 vs R9 300 series. 3 generations is a a pattern and not an accident.

 

[Sweepr]

The only DX12 game Nvidia won big was RoTR. Now that lead has been cut or even erased.

http://hexus.net/tech/reviews/graphics/94969-sapphire-radeon-rx-480-nitro-4gb-8gb-oc/?page=10

Funny because the custom GTX 1060 still leads by 9-15% here (1080P). If that's 'cut or erased' then AMD's minor advantage (3-4%) in DX12/Vulkan titles overall looks even more insignificant.

The current generation is going to be a repeat of the last gen

 

[Silverforce11]

The only DX12 game Nvidia won big was RoTR. Now that lead has been cut or even erased.

http://hexus.net/tech/reviews/graphics/94969-sapphire-radeon-rx-480-nitro-4gb-8gb-oc/?page=10

https://www.youtube.com/watch?v=7ay4rSDKy2M

RoTR was the only game saving them from a larger lead for Rx 480 in DX12 perf avg. Now that too is gone. The next time those avg nos are going to show a lead closer to 7-8%. btw hardwarecanucks comfortably left out Doom Vulkan. That would move the lead well above 10%.

Yep, RotTR was the only game where the 480 wasn't matching or beating the 1060.

Driver update, brings extra perf for 480 already, not bad!

NV should get sponsoring more DX12 titles...

 

[HiroThreading]

I feel like we will soon be hitting 8gb as a lower limit and 12gb will be the standard-midrange.



It also appears that with current gen and earlier, higher clocked memory and/or wider bus can push a lot of overall performance on the card. Wouldn't it make sense to shift focus towards faster/wider memory before going bonkers with gb? Or is that a more complicated and costlier solution?

https://forum.beyond3d.com/threads/is-4gb-enough-for-a-high-end-gpu-in-2015.56964/page-1

https://forum.beyond3d.com/posts/1849915/

 

[jpiniero]

While I'm sure it has more of an affect on weaker cards, changing texture settings exclusively often has a very minor performance penalty. It is also the setting most tied to VRAM usage, naturally.

It's always been that way. The problem of course if you have to pull from system memory because the card is full.. that's when problems start.

4 GB shouldn't be a problem for 1080p. It's resolutions above that you would run into issues. Remember that there is a power cost to more memory, so crazy amounts just isn't worth it, especially in a world where the AAA games are console ports and the consoles have 6-7 GB available to the game.

 

[Cookie Monster]

Heh. Seems like a sketchy way to rationalize the idea that nV isn't behind AMD in next gen API support. Your argument is: "even though they've got more fundamental support, they don't have all the support. Maybe one day nV will blow them away lol".

So I assume the answer is to ignore what we're seeing in current gen cards? And ignore how all AMD GCN cards from the past few years are performing better than comparable nV cards, especially in DX12/Vulcan.

...Because considering four and a half years worth of data would be a bit "premature" ;D

Im not saying don't ignore them. Im saying that it is abit premature to say which card will age better with DX12/Vulkan API era. With Kepler and Fermi, you do have an argument there seeing as they don't support many DX12 features (hardware wise) so the performance benefits for these generation of video cards don't benefit from the new low level APIs. However those video cards as of today still perform ok (its not like performance drops to 0%) but not as well as their old competitors.

Both IHVs don't fully support DX12 and infact Intel out of the 2 provides full support interestingly enough (source - https://en.wikipedia.org/wiki/Feature_levels_in_Direct3D).

Its also true that most of the DX12 titles are console titles i.e. optimised for the GCN architecture as both consoles use a GCN based GPU. It is probably the primarily reason why the first wave of DX12 games perform well on GCN based video cards (however maxwell/pascal based video cards also perform well). There are documentations for it, its been done on the consoles already etc. However there are still many other features developers have yet to fully use if you look at what DX12 has to offer.

One good example is the game Just Cause 3. It maybe a DX11 game, but uses many DX12 features already (source - http://www.dsogaming.com/news/just-...ing-dx12-pc-exclusive-dx12-features-revealed/). Now if you go look at the benchmarks of this game, Maxwell/Pascal based cards which support CR and ROV perform better than its competing products.

Then you have that 1060 vs 480 Doom Vulkan benchmark where using a slower CPUs benefit the former rather than the latter. Now Im not sure of their testing methodology (would be great if a reputable website did some more indepth analysis into the effects of CPU performance in the new low level APIs) but to me it just shows me that as of now, developers are still at large with the DX12 learning curve.

By the time the feature set is maximised and games are embracing DX12 fully, we will have new toys to play with and the old toys will be woefully underpowered as what Arachnotronic has said earlier.

 

[raghu78]

Funny because the custom GTX 1060 still leads by 9-15% here (1080P). If that's 'cut or erased' then AMD's minor advantage (3-4%) in DX12/Vulkan titles overall looks even more insignificant.

I don't know how you do your math. hwc and other reviews used older drivers. the 16.7.3 driver released yesterday brought the significant perf gain in RoTR. hexus.net used the latest drivers and saw huge gains.

http://forums.anandtech.com/showthread.php?t=2481587

http://hexus.net/tech/reviews/graphics/94969-sapphire-radeon-rx-480-nitro-4gb-8gb-oc/?page=10

Other reviews will start reflecting the gains in RoTR soon. I think the Rx 470 review is going to reflect this gain. hwc's dx12 perf lead for Rx 480 was reduced significantly due to RoTR which was a strong win for GTX 1060. The new perf gain for Rx 480 in RoTR closes the gap with GTX 1060 thus increasing its avg DX12 perf lead. :biggrin:

 

[Maverick177]

Funny because the custom GTX 1060 still leads by 9-15% here (1080P). If that's 'cut or erased' then AMD's minor advantage (3-4%) in DX12/Vulkan titles overall looks even more insignificant.

Dude, after all the benchmarks threw at you, I still don't get why you can't accept that P10 is faster than GP106 by quite a margin.

 

[Erenhardt]

Wrong, GTX 1060 is actually faster in some websites:

Yeah. And it is only 10-20% slower than 1070. wright

Cherrypick your numbers better.

 

[mohit9206]

4 GB shouldn't be a problem for 1080p.

6GB(2016)4GB(2015) 3GB(2014) 2GB(2013) 1GB(2012) shouldn't be a problem for max settings 1080p.
But i agree with you about 4gb if medium to high settings is acceptable.