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

[sirmo]

This advantage can't be underestimated. Bodes very well for notebooks as well.

It's overestimated.. even with a slight OC 1320Mhz my rx480 uses avg. 75 watts in overwatch. It's undervolted though (GPU only in GPU-Z).

rx480 used in these tests has 2gb more GDDR5 ram which consumes a significant chunk of power as well.

1060 is more efficient, but the difference isn't as large as some would suggest. And looking at those Doom Vulkan benchmarks at least where reference rx480 is 32% faster I would give it an edge on perf/watt in that game.

AIB rx480 4gb versions will be on par in perf/watt with 1060.

 

[Sweepr]

It's overestimated.. even with a slight OC 1320Mhz my rx480 uses avg. 75 watts in overwatch. It's undervolted though (GPU only in GPU-Z).

GPU-Z doesn't read power consumption for the entire VGA.

rx480 used in these tests has 2gb more GDDR5 ram which consumes a significant chunk of power as well.

A few watts, not significant at all.

AIB rx480 4gb versions will be on par in perf/watt with 1060.

It would need to be >50% more efficient, not happening. And same argument can be used for 3GB 1060 or whatever competes with RX 480 4GB on price.

 

[Bacon1]

Nice find. DX12/Vulkan is all about specific vendor optimizations, and using Gaming Evolved titles as definite proof that the competitor is superior in all titles running on these new APIs is laughable at best.

And the card is a step up from Maxwell here as well.

Thats impossible. Maxwell supports Async Compute. Nvidia claimed so many times. Are they purposely gimping their old hardware to sell you new cards? Has driver support already been removed for Maxwell? Quality drivers indeed

 

[sirmo]

GPU-Z doesn't read power consumption for the entire VGA.

You just repeated what I said?

It would need to be >50% more efficient, not happening. And same argument can be used for 3GB 1060 or whatever competes with RX 480 4GB on price.

Most reviews measured a 25-35 watt difference between two cards. 2Gb of VRAM uses 10 watts by my estimation.. which brings the difference to 15-25 watts. Between two cards.

Which is less than 15% difference.. on 4 out of 5 DX12/Vulkan games tested rx480 is faster.. The difference is really not big at all.. and in fact it is much better than the previous generation of GCN cards. As mentioned in Doom Vulkan rx480 is 32% faster which makes its perf/watt better than 1060s in that game (which is a sign of things to come).

Add in HBM2 into the mix and Vega will be quite competitive in perf/watt to 1080 imo.

 

[Sweepr]

Reviews added to the OP!

Geforce GTX 1060 Reviews

4Gamer
ArabHardware
ArsTechnica
ComptoirduHardware
ComputerBase
Clubic
ElChapuzasInformatico
eTeknix
GamersNexus
GameStar.de
GameSpot
Guru3D
Hardware.info
Hardware Canucks
Hardware.fr
Hexus
HKPEC
HotHardware
NordicHardware
PCGamesHardware.de
PCWorld
PC Perspective
Phoronix
Pure PC
PCLab
PC Watch
SweClockers
Tom's Hardware
TechPowerUp
Tweakers
Xfastest

 

[SPBHM]

the 480 only lasted 3 weeks as the obvious choice :'(

1060 is looking really strong, still a bit behind where we expected (most DX12 games), but not by much, and there is all the bonus of being a better finished product (no power concerns, clear OC margin, DVI port)

I think some of the advantages on DX11 games (like Witcher and fallout) cause a bigger impact than the deficit on things like Hitman, in the end the 1060 is clearly the faster card.

the 480 4GB for $200 is still a very good choice, better value but AMD needs to drop the 480 8GB price urgently, only $10 less is not going to work well for them I think.

perhaps the 480 will age better but for now...

something else, the removal of SLI seems stupid, almost no one uses it I'm sure, but why now? they had it there since the 6600GT from 2004/2005 on al "x60" cards, and this is actually a very good "x60" card, unlike the 960 which was OK, but not on the same level as this,

the funny thing is that I think people who would buy a 970 last year will probably buy a 1060 now (which is a lot cheaper) because it really looks good compared to the 960 and the 1070 is a lot more expensive than the 970 was, kind of feels like Nvidia was forced to the $250 price point because of the 480, competition is good.

 

[Sweepr]

You just repeated what I said?

Most reviews measured a 25-35 watt difference between two cards. 2Gb of VRAM uses 10 watts by my estimation.. which brings the difference to 15-25 watts. Between two cards.

Which is less than 15% difference.. on 4 out of 5 DX12/Vulkan games tested rx480 is faster.. The difference is really not big at all.. and in fact it is much better than the previous generation of GCN cards.

Add in HBM2 into the mix and Vega will be quite competitive in perf/watt to 1080 imo.

50% better perf/watt according to TPU, even more depending on the game, as Hardware.fr tested:

So I'll have to disagree. It's still very significant and by the time Vega reaches the market in 2017 there's more than enough time for new revisions / improvements from NVIDIA's camp.

Edit: 4GB version doesn't change the picture:

 

[sirmo]

50% better perf/watt according to TPU, even more depending on the game, as Hardware.fr tested:

So I'll have to disagree. It's still very significant and by the time Vega reaches the market in 2017 there's more than enough time for new revisions / improvements from NVIDIA's camp.

As I have explained twice, these numbers don't take VRAM difference into account and they do not test for new APIs.

Yes if you're comparing the two reference cards that's the difference, but comparing the architectures and their efficiency for what they were built Polaris is not inefficient by any means.

Take VRAM into the account and take a game like Doom Vulkan and it has better efficiency than 1060 for instance.

GCN efficiency lags in DX11, this is nothing new, but it's on par or better in DX12/Vulkan ..

 

[guachi]

the 480 only lasted 3 weeks as the obvious choice :'(

It's still the obvious choice. You can say with fairly high confidence that any dx12 or vulkan game will run faster or much faster on a 480.

That the 480 will be cheaper.

That a Freesync monitor is 25% less than the otherwise same Gsync monitor.

That even in a slower and older dx11 game you'll still have very high frame rates (e.g. 99 vs. 115 in GTA V 1080)

That you can get a Freesync monitor and have smooth frame rates even in the dx11 games that are slower.

That you'll have $150 left over for hookers and blow with a 480 + Freesync vs. 1060 + Gsync.

GTX1060 - Better in Yesterday's Games! is not a winning tag line.

 

[Dygaza]

Can someone explain why 480 uses same amount of power under DX12 or Vulkan than it does under DX11? I know you AMD cards are undertutilized under DX11, but why don't they consume less power then? Makes no sense.

 

[StinkyPinky]

Quite nice card, but imo the 480 is the better deal. More vram and better DX12 performance.

 

[Sweepr]

Take VRAM into the account and take a game like Doom Vulkan and it has better efficiency than 1060 for instance.

Still >150W idle/load gap using the 4GB model, so probably a very minor difference which wouldn't change any of the conclusions. GP106 is significantly more efficient. And again, there will be models with less RAM here as well (3-4 GB GTX 1060/1050).

Doom Vulkan is useless for comparisons, currently uses optimized render path + Async only on AMD hardware. Same as picking Talos Vulkam to claim huge advantages for Pascal.

ID said, that they optimzed the render path ONLY for AMD yet and that the Nvidia implemention will follow later. How we can test a half-done game without the optimized driver?

Currently asynchronous compute is only supported on AMD GPUs and requires DOOM Vulkan supported drivers to run. We are working with NVIDIA to enable asynchronous compute in Vulkan on NVIDIA GPUs. We hope to have an update soon.

http://www.gamersnexus.net/hwreviews/2518-nvidia-gtx-1060-review-and-benchmark-vs-rx-480/page-4

 

[sirmo]

Can someone explain why 480 uses same amount of power under DX12 or Vulkan than it does under DX11? I know you AMD cards are undertutilized under DX11, but why don't they consume less power then? Makes no sense.

My guess is implementing power gating on each stream processor within a compute unit would take too many transistors, so they just remain idle (noop) when CUs aren't fully utilized.

There is also a command processor and ACEs which aren't fully utilized under DX11 workloads, as well as full precision compute (which Polaris has significantly more of than Pascal) which is mostly unused under graphics workloads.

 

[sirmo]

Still >150W idle/load gap using the 4GB model, so probably a very minor difference which wouldn't change any of the conclusions. GP106 is significantly more efficient.

Doom Vulkan is useless for comparisons, currently uses optimized render path + Async only on AMD hardware. Same as picking Talos Vulkam to claim huge advantages for Pascal.



http://www.gamersnexus.net/hwreviews/2518-nvidia-gtx-1060-review-and-benchmark-vs-rx-480/page-4

That's not a 4gb model. That's just a bios setting which limits the VRAM to 4gb.. those cards still have 8gb worth of modules using power. Wait for true 4gb models and see for yourself. Should be about 20 watts less (probably a bit more since the 4gb version's VRAM is clocked lower as well).

 

[Thala]

My guess is implementing power gating on each stream processor within a compute unit would take too many transistors, so they just remain idle (noop) when CUs aren't fully utilized.

Yup, power islands with according isolation logic are expensive. However they should be clock-gated in idle state and not executing NOPs. Given how many tape-out runs AMD had with Polaris, i would expect they have clock gating under control by now.
But indeed, the question is interesting. I would also expect a measurable power decrease in partial idle situations.

 

[cusideabelincoln]

Nice find. DX12/Vulkan is all about specific vendor optimizations, and using Gaming Evolved titles as definite proof that the competitor is superior in all titles running on these new APIs is laughable at best.

So using Gameworks titles to prove any kind of point must be laughable at best too.

 

[Sweepr]

Roundup of cards released today:

A wide variety of cards are due to hit today, from most of the major players. The list of cards launching today include...

EVGA GTX 1060 edition: $249
EVGA GTX 1060 Superclocked (SC) Edition: $259
ASUS STRIX-GTX1060-6G-GAMING: $329
ASUS Turbo Edition: $249
PNY GeForce GTX 1060: $259
MSI GeForce GTX 1060 GAMING X 6G: $289
MSI GeForce GTX 1060 Gaming 6G: $279
MSI GeForce GTX 1060 Armor 6G OC: $259
MSI GeForce GTX 1060 6GT: $249
Gigabyte GV-N1060G1 GAMING-6GD: $289
Zotac GeForce GTX 1060 AMP 6GB: $279
ZT-P10600A-10L ZOTAC GTX 1060 Mini 6G: $249

6 VGAs listed at $249 @ Newegg (currently OOS), some of them even include a slight factory OC.

www.newegg.com/VGA/EventSaleStore/I...3512X1535351Xcca30333b0424abe4c9d3e6e9dbcb815

Should be about 20 watts less (probably a bit more since the 4gb version's VRAM is clocked lower as well).

Couldn't find any test showing this kind of gap, but feel free to post if you did.

 

[AnandThenMan]

And BTW, what I said about gamers could have bought a $250-275 R9 290 and skipped x60 NV next gen but instead they will pay $200 for a 960 and $200-250 for a 1060 to get marginally faster performance is also coming true. I called it in January 2015 when the $200 GTX960 vs. 250-270 R9 290 showdown started.

I remember this clearly.

www.newegg.com/VGA/EventSaleStore/I...3512X1535351Xcca30333b0424abe4c9d3e6e9dbcb815

Only one shows in stock the Zotac that has a piddly little cooler.

 

[HurleyBird]

Even though the Perf/Watt looks spectacular*, I can't really see myself recommending the card. There's just too much risk that the architecture is going to age as poorly** as Nvidia's previous offerings, and it seems like it might only be a matter of time before the cheaper RX 480 ends up significantly ahead.

*Even with the potential for relative performance regression vs. GCN, the sheer Perf/Watt of GP106 is going to make it a killer laptop chip. That in itself is a huge setback for AMD as Polaris was designed around laptops, and this chip is undoubtedly superior to their efforts in the area.

**Yes, it's possible that Nvidia will improve their DX12/Vulan performance, but after basically lying about the Maxwell async compute driver (and Maxwell's capabilities in the area in general), I'm not holding my breath. If Nvidia does manage to sort things out (Doom seems like the best barometer for next generation game engines, given its IQ relative to performance) a la next generation APIs, then at that time I'd be hard pressed to recommend any of AMD's offerings for any market segment.

 

[Eymar]

It's still the obvious choice. You can say with fairly high confidence that any dx12 or vulkan game will run faster or much faster on a 480.

That the 480 will be cheaper.

That a Freesync monitor is 25% less than the otherwise same Gsync monitor.

That even in a slower and older dx11 game you'll still have very high frame rates (e.g. 99 vs. 115 in GTA V 1080)

That you can get a Freesync monitor and have smooth frame rates even in the dx11 games that are slower.

That you'll have $150 left over for hookers and blow with a 480 + Freesync vs. 1060 + Gsync.

GTX1060 - Better in Yesterday's Games! is not a winning tag line.

Good post, but I think most people will not be buying a new monitor (I could be wrong). However, if they are I definitely would recommend buying Freesync monitor like you suggest as you really don't notice up to 15% fps differences either way so likely save money in long run as can get by with cheaper GPU. I'm really liking how $200 480 4GB paired with $400 32" 1440p FS monitor is performing (Overwatch, Crysis 3, project cars, Dirt Rally all can do 70FPS with mix of high and ultra settings). Note that Freesync helps with GPU stutter, doesn't help with CPU stutter (I tested with G3258 @4.5 vs 6700k and FS doesn't help at all when G3258 at 100%). I wonder if Gsync is better at CPU loaded scenario? If using non Freesync or non Gsync monitor I would go with 1060 as extra horsepower does help remove Vsync related stutter.

 

[sirmo]

Couldn't find any test showing this kind of gap, but feel free to post if you did.

You won't find any because AMD's review samples were all 8gb cards with a dual bios option.

One bios just underclocked and limited the VRAM to 4Gb. But the VRAM was still powered. (VRAM is powered by a single VRM which gets quite warm on rx480) As far as we know even the retail 4Gb models were all 8Gb.. evidenced by users flashing the free 4Gb.

I believe it was Raja Koduri or someone from AMD who in a recent interview with PCper mentioned that 8Gb of VRAM uses 40 watts on rx480. This certainly matches my experience with my rx480 and the GPU-Z which only monitors the GPU. It shows 75 watts avg. in Overwatch for instance.

We also know that in the past (even when you look at those graphs above you posted) that HBM played a big part in why Fury X had a better perf/watt compared to 390X for instance (and VRAM amount difference).. so VRAM amounts are definitely not something to ignore when comparing perf/watt of a given architecture.

 

[SPBHM]

It's still the obvious choice. You can say with fairly high confidence that any dx12 or vulkan game will run faster or much faster on a 480.

That the 480 will be cheaper.

That a Freesync monitor is 25% less than the otherwise same Gsync monitor.

That even in a slower and older dx11 game you'll still have very high frame rates (e.g. 99 vs. 115 in GTA V 1080)

That you can get a Freesync monitor and have smooth frame rates even in the dx11 games that are slower.

That you'll have $150 left over for hookers and blow with a 480 + Freesync vs. 1060 + Gsync.

GTX1060 - Better in Yesterday's Games! is not a winning tag line.

I wonder how large is the market right now for Gsync and Freesync specially for $250 VGAs?
but sure, Freesync is a lot nicer to have than Gsync,
as for the rest,
Vulkan? it's only a 1 game thing (Doom) for now, I know Valve is also playing around with it, but those Source engine games (same for Talos, these are not very advanced and demanding games, they are doing more because of Linux I think)...

as for DX12, the 480 is faster on most titles, but how is it going to look on the long run? I don't know, 3dmark DX12 is not a win for AMD, rise of the tomb raider is a clear win for Nvidia (and yes, it's a game optimized for them, just like Hitman and ashes are for AMD, if you want to ignore tomb raider, you better ignore those to), the consoles using GCN still gives a boost for the 480 chances in the future, but still, if you look at what Nvidia achieved with Gameworks and such, I can see it going more their way.

 

[Bacon1]

Still >150W idle/load gap using the 4GB model, so probably a very minor difference which wouldn't change any of the conclusions. GP106 is significantly more efficient. And again, there will be models with less RAM here as well (3-4 GB GTX 1060/1050).

Doom Vulkan is useless for comparisons, currently uses optimized render path + Async only on AMD hardware. Same as picking Talos Vulkam to claim huge advantages for Pascal.]

That is completely incorrect, please stop saying that. id has not said that at all, they said that they are waiting on drivers from Nvidia, but never that they optimized solely for AMD. They specifically stated they've been working directly with both.

Where are the promised Maxwell async compute drivers we've been waiting for Nvidia to produce since Ashes of the Singularity first released a year ago?

Since late March 2016 we started working daily with both AMD and NVIDIA. Both have been great partner companies, helping bring full DOOM and Vulkan driver support live to the community. There was a lot of work on all fronts but we are pleased with the results.

https://bethesda.net/#en/events/game/doom-vulkan-support-now-live/2016/07/11/156

The FIRST showing of Vulkan was with Nvidia on stage during the 1080 announcement.

 

[garagisti]

You just repeated what I said?

Most reviews measured a 25-35 watt difference between two cards. 2Gb of VRAM uses 10 watts by my estimation.. which brings the difference to 15-25 watts. Between two cards.

Which is less than 15% difference.. on 4 out of 5 DX12/Vulkan games tested rx480 is faster.. The difference is really not big at all.. and in fact it is much better than the previous generation of GCN cards. As mentioned in Doom Vulkan rx480 is 32% faster which makes its perf/watt better than 1060s in that game (which is a sign of things to come).

Add in HBM2 into the mix and Vega will be quite competitive in perf/watt to 1080 imo.

It's actually 4GB extra on the 4GB cards, as there're no 480 cards out with physical 4gb only. So yeah, you're looking at possibly 15W+ less (35W+ for 8gb GDDR5 iirc) on ones which will only have 4GB, and not only 4GB enabled.

So yeah, while the usual Nvidia pushers will like to suggest gap to be as big as it was, it isn't.

The 3dmark only supports feature level 11_0, so i wouldn't be in too much of a rush to say that all DX12 games should work well on 1060, but that's me being a rationalist. Unlike the 960, which was the only card at the time to have h265 acceleration, it is quite different this time around, and much tighter. In DX12, i would wager that 480 would trump the 1060.

 

[MiddleOfTheRoad]

Yeah, looking at the benches -- I'm not really impressed with the 1060. I think Nvidia really overhyped this one. Add the barely adequate amount of VRAM -- I gotta say the RX 480 is the wiser purchase.