AMD Polaris Thread: Radeon RX 480, RX 470 & RX 460 launching June 29th

[sirmo]

Well, here are the news for you. DX12 Explicit multiadapter Linked mode, the basis of the next gen consoles and AMDs future nextgen multigpu dies on interposer GPUs.

http://www.anandtech.com/show/9740/directx-12-geforce-plus-radeon-mgpu-preview/2

Multi Adapter will be a software solution. Consoles will go multi GPU at some point. Unless nanotubes make it into mass produced chips in the next two years which is unlikely, multiGPU is the only way forward.

Tflops need to grow by the orders of magnitude to support 8k per eye VR as well as future 4k and 8k monitors. And it's not going to be accomplished by die shrinks.

 

[tential]

I'd consider it if the chip overclocked well and the Funders Edition had the Fury X CLC.

If the chips OC potential is as currently hyped to be (We got 2 more weeks of hype yay!) and it had the Fury X CLC style cooler.....

for $300?

You'd be weird if you didn't get one....

 

[renderstate]

Well, here are the news for you. DX12 Explicit multiadapter Linked mode, the basis of the next gen consoles and AMDs future nextgen multigpu dies on interposer GPUs.



http://www.anandtech.com/show/9740/directx-12-geforce-plus-radeon-mgpu-preview/2

That's exactly the opposite of what we need to make multigpu dies work efficiently. Ideally a developer wouldn't even know his/her code is running on a multigpu configuration.

For instance a modern GPU is already made of an ensemble of smaller GPUs with their own rasterizer, ROPs, texture units, etc., all working together on the same die. Developers don't write multigpu code for a TitanX and they shouldn't in the future do it if the TitanX-like GPU is physically implemented with multiple chips in the same package.

 

[Glo.]

That's exactly the opposite of what we need to make multigpu dies work efficiently. Ideally a developer wouldn't even know his/her code is running on a multigpu configuration.

For instance a modern GPU is already made of an ensemble of smaller GPUs with their own rasterizer, ROPs, texture units, etc., all working together on the same die. Developers don't write multigpu code for a TitanX and they shouldn't in the future do it if the TitanX-like GPU is physically implemented with multiple chips in the same package.

If you look from Nvidia perspective - yes that is the case because you do not have Hardware Scheduler on the GPU.

AMD hardware does have Hardware Scheduler, therefore it can manage the application, and application can manage itself on the all available hardware.

This exact reason why you were arguing with me about compute on VR without seeing the whole picture.

And you know what? HSA 2.0 is built around Hardware Scheduling. VR built with HSA 2.0 in mind is built with Hardware Scheduling in mind. And VR built with HSA 2.0 in mind will have explicit multi-gpu capability. The same as every application. What will VR games use? DX12.

 

[VulgarDisplay]

And a GTX 1070 murders a single RX 480 by the tune of 45-50%. CFX and SLI aren't practical solutions. Scaling is game and developer dependent, as well as driver dependent. P10 is just looking more and more meh. The only way P10 is a great deal is if it comes in at $229 or less and the 1070 stays inflated. If the 1070 comes down to MSRP or the 8gb 480's are $249 out of the gate, then it's only a very small increase in perf/$ over $400 1070's which is right in line with how lower performing parts should be priced.

This is a pretty definitive post. Should be interesting to revisit come the 29th.

Every leak we have seen for polaris 10 so far has been on drivers that are at least 4 months old.

Sent from my XT1575 using Tapatalk

 

[tviceman]

Tell me how you're able to state all these claims without seeing a single review? Leaked synthetic benchmarks don't really mean much. A lot of indicators out there that this may be a great OC'ing chip. I'll be damned if I had to read through thread after thread of people saying its not fair to compare a FuryX with a 980ti when the 980ti overclocks so well and then hear everyone go back to "stock vs stock" testing.

Tell me how you're able to state all these claims of it being a great OCing chip and amazing perf/$ without seeing a single review or knowing for sure what prices will be on launch? Leaked synthetic benchmarks don't really mean much...except when comparing scores from the same vendor. I'll be damned if I had to read through Fiji threads with people saying how amazing FuryX was going to be with HBM and 600mm2, the CEO saying it was an overclockers dream, and in the end it ended up only tying the 980 TI at 4k before factoring in that 980 TI's overclocked 20-25% while Fury X OC'd 5-10%. Fool me once, twice, thrice AMD hype.

P.S. 1500mhz being 16% over the 1288 boost speed, btw. Good, but not amazing. GP104 overclocks 16% too. http://www.techpowerup.com/reviews/ASUS/GTX_1080_STRIX/27.html
And 22% as well: http://www.pcgameshardware.de/Nvidi...sts/Asus-Geforce-GTX-1070-Strix-Test-1198327/

 

[renderstate]

If you look from Nvidia perspective - yes that is the case because you do not have Hardware Scheduler on the GPU.

Do you also believe in the existence of fairies and elves?


Threadcrapping and trolling are not allowed
Markfw900

 

[tviceman]

This is a pretty definitive post. Should be interesting to revisit come the 29th.

Every leak we have seen for polaris 10 so far has been on drivers that are at least 4 months old.

Sent from my XT1575 using Tapatalk

It's not defensive, it's reality. I've said it myself that if Polaris comes within 20-25% of a GTX 1070 I'll buy a 4gb card at $199 unless GP106 redefines the competition at this price bracket. But everything I'm seeing is that the 1070 will be 40-45% faster. Now people are frothing over 1500mhz, when in actuality it's only 16% above the RX 480 boost speed. What's special about that considering GP104 also OC's 15% above boost speeds? I've learn to read in between all the AMD hype aka Cayman being 40% faster than GTX 480, Tonga utterly flopping on release and never once getting a supposed full die release, Fiji HBM + 600mm2 will destroy GM200.... Round and round we go.

 

[swilli89]

It's not defensive, it's reality. I've said it myself that if Polaris comes within 20-25% of a GTX 1070 I'll buy a 4gb card at $199 unless GP106 redefines the competition at this price bracket. But everything I'm seeing is that the 1070 will be 40-45% faster. Now people are frothing over 1500mhz, when in actuality it's only 16% above the RX 480 boost speed. What's special about that considering GP104 also OC's 15% above boost speeds? I've learn to read in between all the AMD hype aka Cayman being 40% faster than GTX 480, Tonga utterly flopping on release and never once getting a supposed full die release, Fiji HBM + 600mm2 will destroy GM200.... Round and round we go.

How can you make some claim about AMD hype when nVidia is just as guilty if not worst themselves? What about showing a Flounders Edition 1080 at 2.1ghz and 60C? that's not false hype?

 

[Glo.]

Do you also believe in the existence of fairies and elves?

First things, first: Even if I do, it is not your business.

What you did there is attacked the messenger, not the message.

Secondly, all of what I have written was talked over number of forums for past ... 9 months or so.

 

[tviceman]

How can you make some claim about AMD hype when nVidia is just as guilty if not worst themselves? What about showing a Flounders Edition 1080 at 2.1ghz and 60C? that's not false hype?

I can make claims about AMD because there is a major historical pattern of false hype leading up to big AMD product releases. What Nvidia did with their OC demonstration was also false hype and I've been blasting Nvidia in the Nvidia threads for the various blunders of their 1080/1070 launch. AMD (or at least their rabid fans), though, have done it more often and the discussion we're having right now is about AMD cards anyways, not Nvidia. The only reason I even bring up Nvidia cards is to make comparisons - to show that a 16% overclock is already happening and that perf/$ at a potential $250 price tag is nothing to pee your pants over vs. inflated $400-410 GTX 1070's.

 

[linkgoron]

It's not defensive, it's reality. I've said it myself that if Polaris comes within 20-25% of a GTX 1070 I'll buy a 4gb card at $199 unless GP106 redefines the competition at this price bracket. But everything I'm seeing is that the 1070 will be 40-45% faster. Now people are frothing over 1500mhz, when in actuality it's only 16% above the RX 480 boost speed. What's special about that considering GP104 also OC's 15% above boost speeds? I've learn to read in between all the AMD hype aka Cayman being 40% faster than GTX 480, Tonga utterly flopping on release and never once getting a supposed full die release, Fiji HBM + 600mm2 will destroy GM200.... Round and round we go.

So now it's 40-45 and not 45-50? Also, 40-45 at which resolution?

If we assume that the 480 is faster than the 390x at 1440p then the 1070 is already at most 37% faster, according to TPU. If the 480 rumored benchmarks are correct, then the 1070 should be ~35% faster at 1440p.

1080p paints a different picture, where the 1070 is faster by 43%, but we'll see if AMD have managed to improve their somewhat poor 1080p performance. Sure, we might be disappointed and the 480 could be just 390 level, and not 390x, but I think it's too early to definitely say how much faster the 1070 will really be.

I agree, though, that I don't see the 480 coming 20%-25% close to the 1070 at stock. Overclocked? Maybe, but I don't really get why people are so certain that the 480 is going to be some monster overclocker.

 

[tviceman]

So now it's 40-45 and not 45-50? Also, 40-45 at which resolution?

If we assume that the 480 is faster than the 390x at 1440p then the 1070 is already at most 37% faster, according to TPU. If the benchmarks are correct, then the 1070 should be ~35% faster at 1440p.

1080p paints a different picture, where the 1070 is faster by 43%, but we'll see if AMD have managed to improve their somewhat poor 1080p performance. Sure, we might be disappointed and the 480 could be just 390 level, and not 390x, but I think it's too early to definitely say how much faster the 1070 will really be.

I agree, though, that I don't see the 480 coming 20%-25% close to the 1070 at stock. Overclocked? Maybe, but I don't really get why people are so certain that the 480 is going to be some monster overclocker.

I'm looking at 1440p and 1080p, but mostly 1440p and I don't assume the RX 480 is faster than the 390x. On the contrary, according to the leaks, it's a tad bit slower and I'd put a steam game on RX 480 losing ground to the 390x as the resolution scales up simply due to 390x's massive bandwidth advantage. The GTX 1070 is already 37%-43% faster than the 390x at 1440p and 1080p respectively.

And again, we've got people in this thread ultra excited over 1500mhz on the OC, when that's only 16% faster than the RX 480 boost speed. It's a good OC, but not amazing like Maxwell and pretty much exactly right in line with 1070/1080 overclock on a percentage base.

 

[AtenRa]

It's not defensive, it's reality. I've said it myself that if Polaris comes within 20-25% of a GTX 1070 I'll buy a 4gb card at $199 unless GP106 redefines the competition at this price bracket. But everything I'm seeing is that the 1070 will be 40-45% faster. Now people are frothing over 1500mhz, when in actuality it's only 16% above the RX 480 boost speed. What's special about that considering GP104 also OC's 15% above boost speeds? I've learn to read in between all the AMD hype aka Cayman being 40% faster than GTX 480, Tonga utterly flopping on release and never once getting a supposed full die release, Fiji HBM + 600mm2 will destroy GM200.... Round and round we go.

You have only seen 3D Mark FireStrike, I will suggest we wait for actual gaming numbers, especially latest 2015-2016 games.

 

[tviceman]

You have only seen 3D Mark FireStrike, I will suggest we wait for actual gaming numbers, especially latest 2015-2016 games.

Copy that. I am waiting. I just don't expect an anomaly graphics chip that does poor in synthetic vs. it's counter parts and amazing in real world, and I suggest you and everyone else temper those expectations as well.

 

[AtenRa]

Copy that. I am waiting. I just don't expect an anomaly graphics chip that does poor in synthetic vs. it's counter parts and amazing in real world, and I suggest you and everyone else temper those expectations as well.

Im waiting the reviews but i dont care about 3D Mark 11 and 3D Mark FireStrike performance, im wating for DX-12 gaming performance. We already have early numbers for the RX 470 in Hitman and AoTs, why nobody take those and only evaluate the Polaris performance in DX-11 synthetic Benchmarks ??

What if RX 480 DX-11 performance is close to R9 390/X but DX-12 performance is closer to Fury X ???

 

[Glo.]

What if RX 480 DX-11 performance is close to R9 390/X but DX-12 performance is closer to Fury X ???

It is the other way around.

 

[swilli89]

I can make claims about AMD because there is a major historical pattern of false hype leading up to big AMD product releases. What Nvidia did with their OC demonstration was also false hype and I've been blasting Nvidia in the Nvidia threads for the various blunders of their 1080/1070 launch. AMD (or at least their rabid fans), though, have done it more often and the discussion we're having right now is about AMD cards anyways, not Nvidia. The only reason I even bring up Nvidia cards is to make comparisons - to show that a 16% overclock is already happening and that perf/$ at a potential $250 price tag is nothing to pee your pants over vs. inflated $400-410 GTX 1070's.

Again you don't know what the perf/$ is yet so how about we just wait a week to see what holds true? Its looking fairly likely that a well clocking 480 is going to be competitive with a 1070 for a much lower price. It might be 15-20% less performance but is that difference meaningful for most shoppers in this price range when they are looking at pretty much doubling their cost?

 

[tviceman]

Again you don't know what the perf/$ is yet so how about we just wait a week to see what holds true? Its looking fairly likely that a well clocking 480 is going to be competitive with a 1070 for a much lower price. It might be 15-20% less performance but is that difference meaningful for most shoppers in this price range when they are looking at pretty much doubling their cost?

You meant to say 30% less (aka 1070 being 40+% faster) right? Because there will be nothing close to 15-20% outside of 1 or 2 games.

 

[Headfoot]

I can make claims about AMD because there is a major historical pattern of false hype leading up to big AMD product releases. What Nvidia did with their OC demonstration was also false hype and I've been blasting Nvidia in the Nvidia threads for the various blunders of their 1080/1070 launch. AMD (or at least their rabid fans), though, have done it more often and the discussion we're having right now is about AMD cards anyways, not Nvidia. The only reason I even bring up Nvidia cards is to make comparisons - to show that a 16% overclock is already happening and that perf/$ at a potential $250 price tag is nothing to pee your pants over vs. inflated $400-410 GTX 1070's.

Misattribution... It's not AMD making wild claims about performance (though they certainly have stretched the truth at times) it's almost always rumor mongering by forumites who hype and hype until they are so far removed from reality literally nothing but a miracle would satisfy them. Don't impute the actions of fanboys to the company they fanboy for. Besides that, I would not be surprised if there are people intentionally hyping in order to fuel an eventual let down purposefully. It is the internet after all.

If fanboying was actually the company's position, Honda would release a heck of a lot more Civics with enormous spoilers and ground effects stock

 

[Azix]

considering how close the card is to launch, the leaks are going to be accurate to some degree. These new clock leaks are likely accurate but need to know in what form they will come.

At 1500Mhz I'd expect to beat a 1070 stock card at least in dx12. So things get interesting. But will the 1500Mhz be only lucky chips, expensive AIB cards etc?

More support is added by the fact that AMD is putting together better OC software. No point doing that without adequate headroom. Imagine launching that software with something like 390x or fury x. 50-100 Mhz OC is not so impressive.

 

[monstercameron]

Misattribution... It's not AMD making wild claims about performance (though they certainly have stretched the truth at times) it's almost always rumor mongering by forumites who hype and hype until they are so far removed from reality literally nothing but a miracle would satisfy them. Don't impute the actions of fanboys to the company they fanboy for. Besides that, I would not be surprised if there are people intentionally hyping in order to fuel an eventual let down purposefully. It is the internet after all.

If fanboying was actually the company's position, Honda would release a heck of a lot more Civics with enormous spoilers and ground effects stock

You mean the si or the type r?

 

[AtenRa]

It is the other way around.

How come ??

 

[EightySix Four]

How come ??

AMD's previous architectures were heavily underutilized in DX11/OpenGL which is part of the reason they started the push with Mantle. Their new architecture looks to fix those deficiencies, potentially improving their performance in pre-DX12 games drastically.

 

[VulgarDisplay]

Can anyone point me to a 3d mark with a driver that actually has rx480 support please?

Everything we have seen so far clearly states they are using a driver from either February or January.

Sent from my XT1575 using Tapatalk