AMD RX 400 series (Polaris) MegaThread! FAQ & Resources

[TechyInAZ]

LOL, somebody messed up the pricing here: http://www.newegg.com/Product/Product.aspx?Item=9SIA85V4ME3998

 

[-Fran-]

That is one bad value card, lol.

What the hell? xD

Cheers!

 

[AndrewJacksonZA]

Hehehe, or they're just trying their luck! 🙂

 

[BurgerandChips66]

'twas ever thus....

I'm seeing RX 480's and GTX 1060's listed as being 'in stock' here (France) but a Frenchman's definition of 'in stock' does not match up with my own.

I once ordered a card that was supposed to be available, only to be later sent an email informing me there was a 6 week re-stock wait. After 8 weeks and no card I cancelled my order. The website belonged to one of the larger e-tailers in France.

 

[Reynod]

I do not think it is a paper launch. There is stock out there.

 

[Ags1]

GTX1060 has launched with loads of custom cards, while I still only see the ASUS RX480 listed in Holland - and that is due out in August. If it really does give the promised 15-20% performance bump, that would put it ahead of stock GTX1060 in almost all games.

https://rog.asus.com/23922016/gaming-graphics-cards-2/asus-republic-of-gamers-announces-strix-rx-480/

Question is... at what price?

 

[AndrewJacksonZA]

Question is... at what price?

Now that's the question!!

From the link you provided to Asus' website:

and 19% in the latest games such as DOOM

Wait, wait, wait. The RX480 is already about 32% faster than the 1060!!!

 

[17seconds]

Wait, wait, wait. The RX480 is already about 32% faster than the 1060!!!

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 hald-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.

Any question? All current tests are useless 😉

 

[cdrkf]

Wait, wait, wait. The RX480 is already about 32% faster than the 1060!!!

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 hald-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.

Any question? All current tests are useless 😉

Ah yes, nVidia's fabled 'we can do async in software... we'll release a driver, promise!' statement... As obviously software emulation always works out faster than dedicated hardware right? How did that work out for ATOS again?

 

[AndrewJacksonZA]

Wait, wait, wait. The RX480 is already about 32% faster than the 1060!!!

Any question? All current tests are useless 😉

Hi 17! Here's my reply to that over at the 1060 review thread:

Yeah Igor: given the inclusion of the controversial Nvidia-preferring Project Cars, I think that it would not only be fair but honest to include Doom. Update the review when the patch drops. That way we can also see what kind of a difference the patch makes.

If you don't want to do that, then run the AMD cards in Vulkan and the Nvidia cards in OpenGL and show those numbers for now, and then update later.

 

[king3pj]

Wait, wait, wait. The RX480 is already about 32% faster than the 1060!!!

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 hald-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.

Any question? All current tests are useless 😉

Ah yes, nVidia's fabled 'we can do async in software... we'll release a driver, promise!' statement... As obviously software emulation always works out faster than dedicated hardware right? How did that work out for ATOS again?

I'm not sure what you're talking about. Pascal does show improvements with Async Compute. Pascal does have support built in. The 3DMark Timespy results prove that there are improvements to be had with Async support on Pascal cards. Just because a couple early games didn't support it from day one doesn't mean that this will be the case for most games going forward.

The DOOM Vulkan patch notes say that Async Compute support is currently being worked on for Pascal cards. It is also supported on Pascal cards for Rise of the Tomb Raider. The point is looking at current Doom benchmarks with Async enabled is pointless for Nvidia cards because the developers have said it isn't active yet. We won't be able to make those comparisons until it is.

 

[Math Geek]

like dx12 the problem with async is it is not a simple flip a swicth and it's on type thing. there is a command set within async as well. nvidia can work better with certain commands and amd can work better with other ones. so "currently being worked on for Pascal cards" literally means they are working on adding those specific commands that nvidia likes better. people with more time on their hands than i have looked at the time spy benchmark and noted it uses more of the commands that favor nvidia rather than the ones amd handles better.

so your argument on why games are not good sources of info is the same one that should be made for the new 3d mark time spy. it seems to be favoring nvidia making it look better than it does in games.

there is no easy answer to this but do note that your whole argument in favor of time spy is the same one you're using against the games we have seen.

 

[king3pj]

like dx12 the problem with async is it is not a simple flip a swicth and it's on type thing. there is a command set within async as well. nvidia can work better with certain commands and amd can work better with other ones. so "currently being worked on for Pascal cards" literally means they are working on adding those specific commands that nvidia likes better. people with more time on their hands than i have looked at the time spy benchmark and noted it uses more of the commands that favor nvidia rather than the ones amd handles better.

so your argument on why games are not good sources of info is the same one that should be made for the new 3d mark time spy. it seems to be favoring nvidia making it look better than it does in games.

there is no easy answer to this but do note that your whole argument in favor of time spy is the same one you're using against the games we have seen.

My point was that it's possible for Async to improve performance on Pascal cards and Timespy proves that. Do you really think that developers aren't going to use that information to make their games run better on Pascal cards in the future?

I was not trying to say that we can't call Ashes and Hitman wins for AMD. We can. I say it's too early to make that conclusion for Doom since the developers say Async isn't active for Pascal yet but it will be.

My point was that a couple games working better with Async on AMD cards doesn't mean that Async doesn't work on Pascal cards. People keep trying to make that claim and Timespy proves that it can work. Given Nvidia's market share I think it's pretty clear that developers who are working on DX12 and Vulkan games now will take the time to optimize their games to take advantage of Async Compute on Pascal cards. They would be stupid not to.

I want to be clear that I'm not saying they will necessarily do this at AMD's expense. Good developers optimize their games for both. I don't see why Async Compute would be any different.

 

[Math Geek]

works for me. would be nice to see a game optimized for all hardware. that would be my goal if i was developing a game.

it's clear there is more to squeeze from all the hardware out there for sure. dx 12 is still so new it will take time for them to figure out how to make it work better. add in brand new architecture from the gpu's of both sides and developers got their hands full for sure!!

 

[Math Geek]

gigabyte also announced yesterday the G1 480 was released. how long before it hits stores is anyone's guess but they are supposed to be in the wild now.

anyone seen one of these? the G1 is usually a great performer at a decent price.

 

[simon12]

The more I read about it the more I think by the time we know how AMD and Nvidia (recently and soon to be released cards) will compare in future DX12 games the 5xx & 11xx cards will be releasing. My money is AMD will improve more than Nvidia but not by much or every game will just vary wildly.

 

[Math Geek]

that's my thoughts as well. this is one of the pros for consoles. they stay around long enough with the same hardware in every box. developers continuously improve how well they use that hardware over time. so even though an xbox 360 may be 6-7 years old, games look/perform better now than when the system first came out. same hardware, but better performance all those years later. a pc will never touch that level of optimization with similarly aged hardware.

as you noted, by that time, the pc world has moved on to new api's that are not supported or new bigger/better/faster settings that the old parts just can't handle anymore.

 

[BurgerandChips66]

http://sapphirenation.net/NitroYear/

Looks like the Sapphire Nitro RX 480 is about to launch..... I'm rather keen on this one.

 

[cdrkf]

Wait, wait, wait. The RX480 is already about 32% faster than the 1060!!!

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 hald-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.

Any question? All current tests are useless 😉

Ah yes, nVidia's fabled 'we can do async in software... we'll release a driver, promise!' statement... As obviously software emulation always works out faster than dedicated hardware right? How did that work out for ATOS again?

I'm not sure what you're talking about. Pascal does show improvements with Async Compute. Pascal does have support built in. The 3DMark Timespy results prove that there are improvements to be had with Async support on Pascal cards. Just because a couple early games didn't support it from day one doesn't mean that this will be the case for most games going forward.

The DOOM Vulkan patch notes say that Async Compute support is currently being worked on for Pascal cards. It is also supported on Pascal cards for Rise of the Tomb Raider. The point is looking at current Doom benchmarks with Async enabled is pointless for Nvidia cards because the developers have said it isn't active yet. We won't be able to make those comparisons until it is.

Have a look at those numbers though, even though that test is geared as an ideal situation for Pascal, the gain in Pascal is proportionally much less than you see on either AMD card. I mean one view is nVidia cards, Maxwell included, don't perform badly in DX12 or Vulkan- they don't suffer in performance, so using a DX12 or Vulkan render path isn't going to be an issue. That said though they generally doing gain much, whereas AMD cards do (which is probably in part due to the integrated hardware and partly due to AMD's DX11 driver overhead holding their cards back from where they should be based on specifications).

nVidia are desperately trying to downplay hardware Async as it's something they don't have- the fact is however they are adding it into their cards with Volta. That there tells you everything you need to know on the subject- if nVidia's position of 'it doesn't matter, we can do it in software' was valid, they wouldn't be wasting transistors adding it into their hardware. It doesn't mean nVidia cards are terrible or should be avoided, however I do think that the comparatively cheap AMD cards represent a potential bargain- their (comparatively) poor DX11 performance has forced the prices down very low given the capabilities of many of these cards, DX12 and Vulkan are starting to show what the cards can really do.

I mean looking at DOOM for a second- forget the issue of where nVidia might end up with the update for a sec, and just consider an API change is adding 30+% performance to AMD's higher end cards. That's a whole performance tier right there- from a software tweak. I don't doubt that Pascal will gain a bit in DOOM- it will probably put the 1070 back above th RX480 (which it should be it's a much more powerful card) although I'd expect the gain will be less than 30%, as I think the 1070 is already performing close to specification (you can't gain performance if it's already fully utilized after all).

Edit: Some numbers from the 3D mark test- the GTX 1070 gains 7%. The Fury and RX 480 both gain 16%.

 

[BurgerandChips66]

Some unverified benchmarks of a Sapphire Nitro RX 480 OC @ 1420MHz....

https://imgur.com/a/sw5u3#PSkmOZQ

Seems to more than match a GTX 1060 OC @ 2100MHz.

I think the 'official' launch of this card is in about 24 hours time.

 

[TehPenguin]

Some unverified benchmarks of a Sapphire Nitro RX 480 OC @ 1420MHz....

https://imgur.com/a/sw5u3#PSkmOZQ

Seems to more than match a GTX 1060 OC @ 2100MHz.

I think the 'official' launch of this card is in about 24 hours time.

It all comes down to the price. You can buy the cheapest 1060 and OC it to 2GHz no problem, not so much with the 480. IMO, the Nitro has to stay as close to $250 as possible for the 8GB to be competitive.

 

[-Fran-]

Those graphs look promising, but I have a doubt in how the 1060 scales with speed. I got the impression the nVidia cards had better scaling per Mhz than AMD cards (including this round), so a 1060 at 2.1Ghz should keep the same lead or be even faster than the RX480 at 1.4Ghz.

Oh well, official reviews can't come soon enough for the aftermarket RX480s.

Cheers!

 

[Math Geek]

it's gonna be interesting for sure. at the least they will trade blows pretty equally. best case for amd is they beat the 1060 for less money of course. i'm hoping for actually beating the 1060 for cheaper as that will be good for pricing.

 

[king3pj]

Have a look at those numbers though, even though that test is geared as an ideal situation for Pascal, the gain in Pascal is proportionally much less than you see on either AMD card. I mean one view is nVidia cards, Maxwell included, don't perform badly in DX12 or Vulkan- they don't suffer in performance, so using a DX12 or Vulkan render path isn't going to be an issue. That said though they generally doing gain much, whereas AMD cards do (which is probably in part due to the integrated hardware and partly due to AMD's DX11 driver overhead holding their cards back from where they should be based on specifications).

nVidia are desperately trying to downplay hardware Async as it's something they don't have- the fact is however they are adding it into their cards with Volta. That there tells you everything you need to know on the subject- if nVidia's position of 'it doesn't matter, we can do it in software' was valid, they wouldn't be wasting transistors adding it into their hardware. It doesn't mean nVidia cards are terrible or should be avoided, however I do think that the comparatively cheap AMD cards represent a potential bargain- their (comparatively) poor DX11 performance has forced the prices down very low given the capabilities of many of these cards, DX12 and Vulkan are starting to show what the cards can really do.

I mean looking at DOOM for a second- forget the issue of where nVidia might end up with the update for a sec, and just consider an API change is adding 30+% performance to AMD's higher end cards. That's a whole performance tier right there- from a software tweak. I don't doubt that Pascal will gain a bit in DOOM- it will probably put the 1070 back above th RX480 (which it should be it's a much more powerful card) although I'd expect the gain will be less than 30%, as I think the 1070 is already performing close to specification (you can't gain performance if it's already fully utilized after all).

Edit: Some numbers from the 3D mark test- the GTX 1070 gains 7%. The Fury and RX 480 both gain 16%.

Don't get me wrong, I hope AMD does great this generation. I would love to see the market share start to even out and Nvidia prices to be forced down. I would have loved to save $200 and go with a RX 480 instead of my 1070.

Here is the thing though, my main priority is being able to play all my games at a 60 FPS on my 1440p monitor. Currently the best option for that (when price is also a factor) was the 1070 so that's what I went with. The RX 480 might have been able to get the job done in some DX12 and Vulkan games but the 1070 will much more consistently do it. Maybe if DX 12 was the majority and not a very small portion of games I play things would be different. If AMD had a 490 available in the 1070's performance bracket I definitely would have considered that option.

The RX 480 looks like a great 1080p card but since AMD doesn't really have a current 1440p card they weren't an option for me. In PC Gamer's benchmark suite the RX 480 averages 50.3 FPS at 1440p while the GTX 1070 averages 77.4. I know that's not fair because they are in different price and performance brackets but that is what someone looking for a 1440p card currently gets to chose between.

http://www.pcgamer.com/radeon-rx-480-review/

 

[cdrkf]

Have a look at those numbers though, even though that test is geared as an ideal situation for Pascal, the gain in Pascal is proportionally much less than you see on either AMD card. I mean one view is nVidia cards, Maxwell included, don't perform badly in DX12 or Vulkan- they don't suffer in performance, so using a DX12 or Vulkan render path isn't going to be an issue. That said though they generally doing gain much, whereas AMD cards do (which is probably in part due to the integrated hardware and partly due to AMD's DX11 driver overhead holding their cards back from where they should be based on specifications).

nVidia are desperately trying to downplay hardware Async as it's something they don't have- the fact is however they are adding it into their cards with Volta. That there tells you everything you need to know on the subject- if nVidia's position of 'it doesn't matter, we can do it in software' was valid, they wouldn't be wasting transistors adding it into their hardware. It doesn't mean nVidia cards are terrible or should be avoided, however I do think that the comparatively cheap AMD cards represent a potential bargain- their (comparatively) poor DX11 performance has forced the prices down very low given the capabilities of many of these cards, DX12 and Vulkan are starting to show what the cards can really do.

I mean looking at DOOM for a second- forget the issue of where nVidia might end up with the update for a sec, and just consider an API change is adding 30+% performance to AMD's higher end cards. That's a whole performance tier right there- from a software tweak. I don't doubt that Pascal will gain a bit in DOOM- it will probably put the 1070 back above th RX480 (which it should be it's a much more powerful card) although I'd expect the gain will be less than 30%, as I think the 1070 is already performing close to specification (you can't gain performance if it's already fully utilized after all).

Edit: Some numbers from the 3D mark test- the GTX 1070 gains 7%. The Fury and RX 480 both gain 16%.

Don't get me wrong, I hope AMD does great this generation. I would love to see the market share start to even out and Nvidia prices to be forced down. I would have loved to save $200 and go with a RX 480 instead of my 1070.

Here is the thing though, my main priority is being able to play all my games at a 60 FPS on my 1440p monitor. Currently the best option for that (when price is also a factor) was the 1070 so that's what I went with. The RX 480 might have been able to get the job done in some DX12 and Vulkan games but the 1070 will much more consistently do it. Maybe if DX 12 was the majority and not a very small portion of games I play things would be different. If AMD had a 490 available in the 1070's performance bracket I definitely would have considered that option.

The RX 480 looks like a great 1080p card but since they don't really have a current 1440p card they weren't an option for me. In PC Gamer's benchmark suite the RX 480 averages 50.3 FPS at 1440p while the GTX 1070 averages 77.4. I know that's not fair because they are in different price and performance brackets but that is what someone looking for a 1440p card currently gets to chose between.

http://www.pcgamer.com/radeon-rx-480-review/

Oh yeah the RX480 is more of a 1080p card overall. That said the AMD Fury series look comfortable at 1440p, 60fps based on that, the issue is for the same money a 1070 is faster and more efficient (though Fury wasn't too bad power wise thanks to HBM).

I wouldn't rule out the Fury cards out of hand though- there's no major tech they are lacking imo- it's just a question of price which I think is already dropping and should settle out below the 1070.

There's talk of AMD releasing something else this year, although that could be a dual gpu card. They have said the higher tier Vega gpu isn't coming until next year.