Graphics Pascal and Polaris Discussion Thread

They look totally insane so small.

 

What is happening here with the gtx 1060, it's getting tortured by the rx 480 in doom vulcan:

http://www.hardocp.com/article/2016...x_1060_founders_edition_review/4#.V45B1DV1R_k

I'd be concerned about it's dx12 capabilities in the future, from all the reviews I've read tomb raider is the only one it was faster in.

So basically the mid range price cards are officially dead. No more £160ish price bracket. The closest is the 4gb rx 480 around £175 which isn't perfect itself with all the negativity around power and the terrible cooler.

All in all its a bit....meh... the new gen is supposed to give you more performance for your money, not the same performance as before for the same cash.

 

172 mm on them both, I'll wait for a bit me thinks.

Turns out my 750 ti is only 161. So that should fit, maybe a budget build for now till things settle.

Bigger question becomes what to do with the 980 in the main rig.

 

I looked at that and the average is screwed by the max fps been 260+. The Nvidia card actually has the better minimum which will matter more in the real world.

 

Gonna say, the 1060 minimum is 38.5 whereas the 480 is 28.0. A 10fps difference in minimums is a big thing, especially when the 480 is going below the magic 30.

Also, to my knowledge, NVIDIA have done basically nothing with Vulkan whereas AMD seem to be betting the farm. I'm interested to see NVIDIA's response with driver updates.


EDIT: And I just bought a 1080. Gosh darn it.

 

Did that myself last week, Looking over the review more, Would anyone truly accept such a spiky frame rate, That has got to feel pretty horrible in a real play through across all titles.

Lows of 12 and 15 at 1440p would not be accepted by most players.

Looking at the minimums alone which everyone here would agree is a much better way of judging fps.

Cards are 1060 vs 480 vs 980

Doom 1440p 38 vs 28 vs 41
Doom 1080p N 32 vs 39 Nightmare settings
Hitman 1440p 17 vs 15 vs 17
Hitman 1080p U 12 vs 24 vs 10 Ultra settings
T Raider 1440p 31 vs 28 vs 12
T Raider 1080p 19 vs 28 vs 10 All Settings maxed dx 12
T Raider 1440p 33 vs 33 vs 34
T Raider 1080p 46 vs 40 vs 43 All Settings maxed dx11
Witcher 3 1440p 41 vs 37 vs 39
Wticher 3 1080p 36 vs 33 vs 34 Hairworks enabled ( Can AMD even run this)

1060 benches below 30fps minimum - 3 alot below 30fps minimum and 4 very close to it
480 benches below 30fps minimum - 5 Alot closer to the 30fps target but all still below 2 very close to it as well
980 Benches below 30 fps minimum - 4 Well below the minimum.

So across there tested titles in dx 12 you would have to edit the settings to get a stable 30fps minimum in probably every benchmark tested on every single card at the settings they are testing with.

HardOp has some Crazy maximum fps figures the 980 is above 500 fps at one point, What sort of screen tearing is taking place at this point. Same with the 480 running at 250 + fps on one of the tests.

Hopefully a decent website will run the dx 12 tests and give us some workable results.

 

Did you actually read that review? or were you too busy playing Bottom Trumps?

At 1440p using the same playable settings found on the GeForce GTX 1060 we find the AMD Radeon RX 480 simply blows the GeForce GTX 1060 out of the water in terms of performance. While both are allowing a very high post 60FPS performance, AMD Radeon RX 480 takes it to a new level and even beats out the GeForce GTX 980! The Radeon RX 480 is performing 25% faster than the GeForce GTX 1060 under the Vulkan API, just wow.


And.

The performance gap widens at 1080p. Here the AMD Radeon RX 480 is 32% faster than the GeForce GTX 1060 running "Ultra" settings and "Nightmare" settings at 1080p under the Vulkan API. Performance has increased with the more CPU bound resolution, perhaps showing one advantage of Vulkan API (and coincidently DX12 as well) the fact that CPU overhead can be reduced at lower resolutions improving performance.

Yet you immediately steamed in with "Oh well it's the minimum FPS that counts". Obviously the minimum FPS is less than 27 odd things get jerky but seriously to ignore the fact that the 480 is that much faster is just plain deluded. Sorry dude, usually I put up with the fact that you seem to like Nvidia but on this one you're really reaching.

Min FPS are inevitable. They're usually down to poor coding or a bad driver. Average FPS are far more important here.

 

I agree with Vault-tec, if you look at the frame dips it's only 3 places the rx480 falls below the gtx 1060, all of which are towards the end of the run when the card will have sufficiently heated enough for the gpu to encounter throttling. Don't forget this is a reference cooled 480 vs the 1060 FE. I'd like to see the same run with a custom cooled rx 480 vs gtx 1060. What's the hold up AMD?

 

If the card is throttling in such a short run doesn't that indicate a bigger problem ? So they have compared 2 reference cards what's the problem ?

There's a reason I got rid of SLI and its frame drops below 30fps and micro stutter that is the biggest bug bear of it. Would love to see the frame variance on some of these graphs.

In a racing game a sudden frame drop is you in the wall. A fps shooter you die.

If you take out the 2 huge Anomalies the cards compete pretty well.

As I said early in this thread by the time Vulkan or dx12 is in enough games both the 480 and 1060 will be out of date tech. We both agreed vault it's at least 2-3 years off before either will be widespread enough. 1 game does not make a graphics card.

Hardocp also could of shown the CPU performance benefits by not using a top end Intel CPU to do the testing. How realistic is it that either card will end up with that CPU. For the Vulcan bench it's meant to reduce CPU overhead so if you run both cards on say a i5 2500k what's the performance look like and even that CPU will out pace AMDs range.

Did not say the 480 is not impressive, as I said earlier in this thread it's a nice piece of kit. But I'd never buy one. Same with the 1060 not for me.

Nvidias custom cards are already out so let's run that one. The one everyone wants to use as a test bed will end up more expensive than most of those custom cards. ( sapphire nitro)

If your looking for good performance at £250 either card will do the job, the harsh facts with the £ the way it is the UK will have issues with pricing on certain products. Gpus are one of them sucks but that's life. Buy what you can afford.

Which for most of the threads main contributors is 980+ gpus up to a few 1080s and a Titan X. We have all saw the benches but not one of us has made the purchas of these 2 gpus. What's that say about it's likely target market?

Firestrikes new dx12 bench is fun to run, not actually saw its comparison between the 480 and 1060. Wonder if they will make a Vulkan bench.

 

Yes, a well publicized problem, Amd's amazing cooling solution consisting of a single piece of aluminium with a pressed copper core akin to something you would see 10 years ago from one of it's cpu range. Now compare that to Nvidia's reference solution and you see the issue:

"The cooler has a fairly traditional design, with a copper baseplate connecting with the GPU and feeding a pair of copper heat pipes and an aluminium heatsink above. A metal contact plate also ensures that the memory chips and power delivery components are directly cooled as well. The fins are arranged so as to direct all air out of the rear I/O panel, and everything is cooled by a single radial fan which, sadly, is not semi-passive."

Piss poor from AMD but it is something to consider.

Here is another example from gamersnexus, look at the gpu frequency spikes towards the end of the run:



Full article here if you haven't already read it:
http://www.gamersnexus.net/guides/2499-diy-amd-rx-480-hyrbid-results-with-bigger-overclocking

As you say, both are good cards but equally both have their drawbacks. And yes I'd like to see what the nitro can do with a proper cooler and 8 pin pci-e, that way you rule out cooling and power delivery issues and see a true picture of the doom test above.

 

Err, not in my opinion. As long as the average is acceptable, the next place I look is minimums. A slightly lower average with good minimums is far better than a higher average with awful minimums.

 

High minimum as poss with a low standard deviation from average: consistent performance is key, unless the game or test itself is throwing up wildly challenging loads.

I'm not sure why people really care about ref cards though. After market are so much better designed.

 

Depends on what kind of cooler you want. If you want a rear-exhaust card, then OEM coolers are utterly worthless garbage.

 

Whenever I used to run Heaven in SLI I always got the exact same stutter at the exact same point in the benchmark. Always. Never anywhere else, just in the same one place.

It didn't stop me using SLI, just because of that one stutter that was probably down to other factors.

OK let me put it this way. If you ran a benchmark on the 1080 and it scored over 300 FPS but in one frame dropped to 25 FPS would you then base your opinions on that one dip? don't be silly, of course you wouldn't.

The simple fact is that in Vulkan and more specifically Doom the 480 absolutely thrashes the 1060. There's simply no getting around it at all. And this shouldn't be of any surprise either providing you had done your homework on GCN and how it works with Async etc. IE - AMD have been making Fermi GPUs forever and ever and finally there is some software that uses them properly.

Whilst Nvidia on the other hand have been making "here and now" GPUs with no plans for the future.

And the 1060 is another shining example of a "here and now GPU" and if you don't benchmark Doom or any other DX12 titles the 1060 looks better. But it won't, not for long. Mind you they don't really care about next week even providing they pull enough wool to make some sales.

 

The reason Vulkan is good for AMD is more to do with optimisation than it is to do with Async. AMD have awful OpenGL drivers always have done, they are nearly 30fps behind on one of benches I saw when comparing Vulkan and OpenGL.

With Async force disabled the performance difference is about 5% loss on most of the gpus.

 

There are AIB cards with this now. Prefer 2/3 fan with fan-stop tech tbh.

 

so when the RX 480 thermal throttles everyone goes apeshit - but when the 1070 and 1080 do it at stock in a sealed case then no bats an eyelid

 

Was not aware it is thermal throttling thought it was just reducing the boost as the temps increased ? Thermal Throttling is a different issue in my mind.

 

That is a textbook definition of thermal throttling. People just need to get away from the idea its a bad thing. Its a 'feature' now - thermal throttling back in the day would be slowing down from the core speed, but now its just slowing down from its max boost speed.

Also, lots of decent 1060s at about £240 it seems, only looks like bad value when you compare it to the £174 4GB 480, compared to the £220 8gb version, it seems reasonably priced.

 

Nvidia boost is 3 different levels and based on load and temps, is AMDs system similar noticed how cheap the 1060 is