AMD Radeon RX 480 Designed To Bring VR To The Budget-Strapped Masses

[jbc029]

The real coup here is that the 480 finishes removing *all* maxwell cards from recommendations.

The 1070 gutted the product line for anything 980 and above (because it's cheaper and performs just as well or better than all of it). The 480 took care of the rest (970 performance for 960 price). In less than two months, Maxwell, Fury and Hawaii have become completely irrelevant.

Not completely. Yes the RX 480 will be replacing the 970, my 960 , 380 and 390 but let us not forget there are still a huge chunk of people there who wants to go lower. It would be really interesting to see a new contender for the GTX 750ti and GTX 950 for these are the really budget champs here for people who does not give a flying poop about VR but is still able to play at 1080p better than today's consoles could offer.

Isn't that what Polaris 11 will be for?

Isn't that what Polaris 11 will be for?

The official pitch for Polaris 11 is mobile-centric but there is supposed to be a 470 in there too.

Yes, P11 is the low end solution. It's roughly equivalent to a 950, maybe a hair faster but doesn't require a power connector at all. It's around a 40 watt card and runs off just the power from he PCIe slot. Pricing is the big question mark here. I imagine the P11 card is going to the be $100 card in the "$100-$300" Polaris launch.

 

[iam2thecrowe]

The one problem i have with AMD at the moment, is performance consistency. Comparable AMD and Nvidia gpu's perform roughly on par with the correct workload, but there are many games that nvidia steams off with the lead with products at the same price point. I know people will say to this, "oh but nvidia pays off/works with devs to make games run better on their gear", but still, I dont want to buy a gpu and have it perform poorly on one or two games i really want to play.... AMD really need more consistent performance.

A year long track record of turning around the drivers, from being buggy to reliable with performance gains, might be seen by some as more consistent performance towards end users.

I disagree, "Turning around the drivers" has absolutely nothing to do with consistent performance, when does "bug fix" = consistent performance? I guess they have to consistently bug fix.... I have swapped between AMD and nvidia quite a few times, I've never actually had problems with buggy drivers except when using crossfire, we all know about the frame time variance/stutter issues, that they fixed, somewhat. When I say consistent i mean for example an r9 390x or fury gets beaten by a gtx960 in some games, and its theoretically much more powerful. nvidia seems to be more consistent across more titles, never really having a game they do really badly at.

 

[jbc029]

The one problem i have with AMD at the moment, is performance consistency. Comparable AMD and Nvidia gpu's perform roughly on par with the correct workload, but there are many games that nvidia steams off with the lead with products at the same price point. I know people will say to this, "oh but nvidia pays off/works with devs to make games run better on their gear", but still, I dont want to buy a gpu and have it perform poorly on one or two games i really want to play.... AMD really need more consistent performance.

A year long track record of turning around the drivers, from being buggy to reliable with performance gains, might be seen by some as more consistent performance towards end users.

I disagree, "Turning around the drivers" has absolutely nothing to do with consistent performance, when does "bug fix" = consistent performance? I guess they have to consistently bug fix.... I have swapped between AMD and nvidia quite a few times, I've never actually had problems with buggy drivers except when using crossfire, we all know about the frame time variance/stutter issues, that they fixed, somewhat. When I say consistent i mean for example an r9 390x or fury gets beaten by a gtx960 in some games, and its theoretically much more powerful. nvidia seems to be more consistent across more titles, never really having a game they do really badly at.

I'd be curious to know what that game is, where a 960 can beat a Fury. Please, share with the class.

 

[mpampis84]

Does anybody know if you can crossfire the RX 480 with an old AMD GPU? My current setup has an HD7850, so I was wondering if I could keep it as a second card in Crossfire.

 

[jbc029]

Does anybody know if you can crossfire the RX 480 with an old AMD GPU? My current setup has an HD7850, so I was wondering if I could keep it as a second card in Crossfire.

Unfortunately, no. There may be something that would let a Polaris 10 card CF with a Polaris 11 card, but even that's a stretch.

 

[InvalidError]

Does anybody know if you can crossfire the RX 480 with an old AMD GPU? My current setup has an HD7850, so I was wondering if I could keep it as a second card in Crossfire.

Even if you could crossfire GPUs across completely different architecture generations, the slowest of the two becomes a bottleneck for the other and you would most likely end up with crossfire performance worse than the 480 alone.

 

[Outlander_04]

A GTX 970 had comparable performance to a 780ti . Does that make the 970 a failure?
Can Nvidia start competing in asynchronous compute? Probably not . Expect the AMD architecture to shine in DX 12 and Vulcan

 

[Outlander_04]

The wait between now and June 29th is not going to help AMD. That's a blunder as it gives NV plenty of time to get all the data they can through their sources and prepare their response. Whether that's lowering prices on Maxwell cards or getting a new card out.

Yeah probably not . Polaris follows the release of the high end GTX cards by 28 days . It happens exactly on the schedule AMD announced months ago .
No amount of soft launches by nvidia where the cards they "release" are not even available will change that. And when they do become available they are price gouging "founders editions" probably because they cannot be produced in large numbers ,
Unless of course an appropriate response from nvidia is to announce their new card now and wait for 6months before it is available ?

I'd say AMD's strategy was brilliant . They will sell way more RX480's than nvidia will sell 1070's and 1080's which are currently a complete waste of money for anyone using a 1080p monitor [ which is nearly everyone]

 

[Aspiring techie]

I'd like to see benchmarks done with the 480 and the 390. The 390 would be locked at 1GHz and the 480 at 1.11GHz (to even out the compute unit difference). If the clock rates are locked, then the raw compute power will be determined by the IPC of each compute unit.

Yes, I like the extra clock rates and the transistor increase that comes with 16nm FinFet, but that shouldn't be all that beefs up Polaris and Pascal, or else the whole competition between red and green is pointless. The real innovation should be measured by how much IPC gained per compute unit or CUDA core. Compare the gains, and then you see how good each team's engineers are.

 

[InvalidError]

Yes, I like the extra clock rates and the transistor increase that comes with 16nm FinFet, but that shouldn't be all that beefs up Polaris and Pascal

One of the key differences between GCN 1.x and 4.x is that instead of being parallel SIMD, AMD has gone serial to allow shaders to make more efficient use of ALUs: most modern shaders use odd mixes of multiplications that operate only on a subset of texture and geometry data, causing tons of resources in the VLIW4 ALU to go severely under-used most of the time or produce multiple discarded results, so they have switched to doing VLIW serially and letting the ALU skip work (the components of a matrix, vector or scalar multiplication) bound to get discarded.

Less ALU time wasted on unnecessary results might be the second biggest reason behind Pascal's massive performance per watt improvement, the first being 14nm.

 

[Valantar]

The real coup here is that the 480 finishes removing *all* maxwell cards from recommendations.

The 1070 gutted the product line for anything 980 and above (because it's cheaper and performs just as well or better than all of it). The 480 took care of the rest (970 performance for 960 price). In less than two months, Maxwell, Fury and Hawaii have become completely irrelevant.

Not completely. Yes the RX 480 will be replacing the 970, my 960 , 380 and 390 but let us not forget there are still a huge chunk of people there who wants to go lower. It would be really interesting to see a new contender for the GTX 750ti and GTX 950 for these are the really budget champs here for people who does not give a flying poop about VR but is still able to play at 1080p better than today's consoles could offer.

Isn't that what Polaris 11 will be for?

Isn't that what Polaris 11 will be for?

The official pitch for Polaris 11 is mobile-centric but there is supposed to be a 470 in there too.

Yes, P11 is the low end solution. It's roughly equivalent to a 950, maybe a hair faster but doesn't require a power connector at all. It's around a 40 watt card and runs off just the power from he PCIe slot. Pricing is the big question mark here. I imagine the P11 card is going to the be $100 card in the "$100-$300" Polaris launch.

Exactly. Expecting Polaris 11 to be "mobile only" is a ridiculous idea - as long as the chips exist, they'll go into a desktop SKU. The GTX 750 Ti/950 have been huge successes for Nvidia, I have no doubt AMD wants in on that market. I would expect the RX(?) 470 to be a cut-down 480 (Polaris 11 with half the hardware slotting in one rung down would be odd to say the least), but the 460 or 450 might be P11. Have to say I really like the idea of an end-to-end new product stack. HDMI 2.0, DP 1.4, TrueAudio and H.265 decode on the low end? Yes please.

 

[Mousemonkey]

The real coup here is that the 480 finishes removing *all* maxwell cards from recommendations.

The 1070 gutted the product line for anything 980 and above (because it's cheaper and performs just as well or better than all of it). The 480 took care of the rest (970 performance for 960 price). In less than two months, Maxwell, Fury and Hawaii have become completely irrelevant.

Not completely. Yes the RX 480 will be replacing the 970, my 960 , 380 and 390 but let us not forget there are still a huge chunk of people there who wants to go lower. It would be really interesting to see a new contender for the GTX 750ti and GTX 950 for these are the really budget champs here for people who does not give a flying poop about VR but is still able to play at 1080p better than today's consoles could offer.

Isn't that what Polaris 11 will be for?

Isn't that what Polaris 11 will be for?

The official pitch for Polaris 11 is mobile-centric but there is supposed to be a 470 in there too.

Yes, P11 is the low end solution. It's roughly equivalent to a 950, maybe a hair faster but doesn't require a power connector at all. It's around a 40 watt card and runs off just the power from he PCIe slot. Pricing is the big question mark here. I imagine the P11 card is going to the be $100 card in the "$100-$300" Polaris launch.

Exactly. Expecting Polaris 11 to be "mobile only" is a ridiculous idea - as long as the chips exist, they'll go into a desktop SKU. The GTX 750 Ti/950 have been huge successes for Nvidia, I have no doubt AMD wants in on that market. I would expect the RX(?) 470 to be a cut-down 480 (Polaris 11 with half the hardware slotting in one rung down would be odd to say the least), but the 460 or 450 might be P11. Have to say I really like the idea of an end-to-end new product stack. HDMI 2.0, DP 1.4, TrueAudio and H.265 decode on the low end? Yes please.

http://www.guru3d.com/news-story/rumor-nvidia-to-use-gtx-1070-and-1080-in-laptops.html

If true what is to stop AMD doing the exact same thing?

 

[InvalidError]

I would expect the RX(?) 470 to be a cut-down 480 (Polaris 11 with half the hardware slotting in one rung down would be odd to say the least), but the 460 or 450 might be P11.

Rumored specs for the 470 say 4GB of 128bits GDDR4, 1024-1280 shaders and 50W TDP. If true, between the architecture's efficiency gains and clock boost, that might put it in the neighborhood of the R9-380.

For such an R(whatever)-470 to make sense next to the $200 RX-480, it would need to be priced under $140.

 

[Aspiring techie]

One thing that would be cool would be a $100 card that requires no power connector. The GTX 950 and 750ti have been rocking the cheap, small form factor, low power consumption, yet decent gaming segment for too long now. Besides, I'm looking to do a build soon with a 750ti, and I want a cheap Pascal or Polaris offering to be released so I don't have to go Kepler.

 

[Valantar]

http://www.guru3d.com/news-story/rumor-nvidia-to-use-gtx-1070-and-1080-in-laptops.html

If true what is to stop AMD doing the exact same thing?

Yes, and..? That hardly even relates to what I was saying. I'd even say such theories corroborate what I'm saying - that the same chips are used for both mobile and desktop. And as full Tonga has been used in mobile (R9 M295X), I'd fully expect its replacement, Polaris 10, to do the same. Especially seeing as it uses less power, but performs even better. That's a no brainer. The differences I expect are in clock speeds and RAM speeds, as these are easy ways to reduce power consumption. I'd bet they can get the RX 480 down to ~100W without sacrificing even close to 33% performance - which would be great numbers for a second-tier mobile chip. And then Vega would step in with HBM and ~150W for high-end mobile.

I would expect the RX(?) 470 to be a cut-down 480 (Polaris 11 with half the hardware slotting in one rung down would be odd to say the least), but the 460 or 450 might be P11.

Rumored specs for the 470 say 4GB of 128bits GDDR4, 1024-1280 shaders and 50W TDP. If true, between the architecture's efficiency gains and clock boost, that might put it in the neighborhood of the R9-380.

For such an R(whatever)-470 to make sense next to the $200 RX-480, it would need to be priced under $140.

Ugh, I hope that's not true. I know the tendency in the GPU business is for a huge drop between third- and fourth tier cards (GTX 960 to 950 and so on), but fully 50%? That would be awful, even if the power consumption dropped to 50-75W. There really ought to be at least one step between Polaris 11 and full Polaris 10, if the rumors about ~1280 shaders are true. Both 1536 shaders (24 CUs) with high clocks and 1920 shaders (30 CUs) clocked lower should provide a good in-between step and a good use for dies with defective CUs. ~$120-140 for full P11, ~$160-170 for cut-down P10 (restricted to only 4GB to not cannibalize the RX 480), and $200 for the 480. Sure, that's a lot of SKUs, but $30 makes a lot of difference at these price points.

One thing that would be cool would be a $100 card that requires no power connector. The GTX 950 and 750ti have been rocking the cheap, small form factor, low power consumption, yet decent gaming segment for too long now. Besides, I'm looking to do a build soon with a 750ti, and I want a cheap Pascal or Polaris offering to be released so I don't have to go Kepler.

Unfortunately, at $100 you won't get much of a GPU no matter what - at those price points you're cutting it mighty close to where PCB costs, connectors, coolers and whatnot have a significant impact on margins. Then again, you don't need much of a cooler for 50W. Still, with $5-10 for the cooler, probably the same for the PCB, $2-3 for packaging, ~$5 for onboard components, ~$5 for manufacturing and assembly, $5-10 for DRAM, another few bucks for shipping & handling, some for R&D, margins for AMD and board partners, margins for distributors and retailers, you're not left with much money at all for the GPU. I agree with you - a decent (for 1080p) GPU around 50W for ~$100 would be AWESOME. But I very much doubt it will happen for a few generations yet.


Edit: typo. Polaris 10, not 12. I keep getting confused by the "lower numbers are better" thing.

 

[Valantar]

Yes, I like the extra clock rates and the transistor increase that comes with 16nm FinFet, but that shouldn't be all that beefs up Polaris and Pascal

One of the key differences between GCN 1.x and 4.x is that instead of being parallel SIMD, AMD has gone serial to allow shaders to make more efficient use of ALUs: most modern shaders use odd mixes of multiplications that operate only on a subset of texture and geometry data, causing tons of resources in the VLIW4 ALU to go severely under-used most of the time or produce multiple discarded results, so they have switched to doing VLIW serially and letting the ALU skip work (the components of a matrix, vector or scalar multiplication) bound to get discarded.

Less ALU time wasted on unnecessary results might be the second biggest reason behind Pascal's massive performance per watt improvement, the first being 14nm.

I thought that was what they did with GCN 1.0, when they moved away from VLIW?

To quoute Anandtech (http://www.anandtech.com/show/10375/arm-unveils-bifrost-and-mali-g71/2) (my emphasis):

In designing Bifrost, ARM’s GPU group likes to joke that they just turned their old shader core 90° to the left, which all things considered is not a bad analogy. Rather than trying to extract 4 instruction’s worth of ILP from a single thread, ARM groups 4 threads together and executes a single instruction from each one.

If all of this sounds familiar to our long-time readers, then it should. This is a very similar transition to what AMD made with Graphics Core Next in 2011, a move that significantly improved AMD’s GPU throughput and suitability for modern rendering paradigms. And for ARM we are going to see a lot of the same."

 

[jbc029]

It's a good place for AMD. As of the now (as long there's stock on shelves for the 29th) nvidia has no cards below $380 to be recommended. The 970 would have to be dropped to around 60% of it's current price to be competitve and that just isnt going to happen.

 

[Martell1977]

It's interesting to see people talking of nVidia releasing a 1060ti if the 1060 can't keep up with the 480. But history tells us that, that is not a given thing. The 960 is handily out-performed by the R9 380, yet a long awaited 960ti never materialized.

I don't expect the 480 to compete with the 1070, just like the 380 doesn't compete with the 970. This "paper launch" is just the tip of the iceberg. There are a lot of products still to come. I have to admit, this trickle out approach could drive ya crazy, lol.

I'm still waiting to hear what is done in the mobile area. I'm looking to get a new laptop, but don't want to buy something that will be crushed by cheaper machines in a month or 2. If I were to buy one now, I'd be looking at the 950m, so that's about my price range ~$700.

 

[rikkof name taken]

GAMES WILL STILL BE RUN BETTER ON NVIDIA HARDWARE, ALWAYS
I will not play with teh performance of my pc and games, so i will gladly pay 150 more bucks to get the 1070.
the rx 480 is a desperate attempt of amd n00bs who cnat even make a driver to work properly to come back in the game
Guess what, they wont
next year this date, there will be 10000000000 threads of "i have a rx 480, why my fps is sooo low"...u know why ? Coz u were cheap and didn't want to shelve 150 more bucks for a card (1070) that would last u at least 3.5 years form now on

 

[rikkof name taken]

GAMES WILL STILL BE RUN BETTER ON NVIDIA HARDWARE, ALWAYS
I will not play with teh performance of my pc and games, so i will gladly pay 150 more bucks to get the 1070.
the rx 480 is a desperate attempt of amd n00bs who cnat even make a driver to work properly to come back in the game
Guess what, they wont
next year this date, there will be 10000000000 threads of "i have a rx 480, why my fps is sooo low"...u know why ? Coz u were cheap and didn't want to shelve 150 more bucks for a card (1070) that would last u at least 3.5 years form now on

yeah i am agree with u
all these ppl happy for 200 bucks vidoe card , and when will p[lay dark souls 3 their fps with rx 480 will be like 20 lmfao
NVIDIA 1070 IS BEST CARD FOR THE PRICE

 

[rikkof name taken]

GAMES WILL STILL BE RUN BETTER ON NVIDIA HARDWARE, ALWAYS
I will not play with teh performance of my pc and games, so i will gladly pay 150 more bucks to get the 1070.
the rx 480 is a desperate attempt of amd n00bs who cnat even make a driver to work properly to come back in the game
Guess what, they wont
next year this date, there will be 10000000000 threads of "i have a rx 480, why my fps is sooo low"...u know why ? Coz u were cheap and didn't want to shelve 150 more bucks for a card (1070) that would last u at least 3.5 years form now on

yeah i am agree with u
all these ppl happy for 200 bucks vidoe card , and when will p[lay dark souls 3 their fps with rx 480 will be like 20 lmfao
NVIDIA 1070 IS BEST CARD FOR THE PRICE

ALSO, FACT:
No 1 cares of stupid VR anyway, what we care is good fps in normal pc games...thats where NVIDIA always win over trAsMD

 

[Martell1977]

GAMES WILL STILL BE RUN BETTER ON NVIDIA HARDWARE, ALWAYS
I will not play with teh performance of my pc and games, so i will gladly pay 150 more bucks to get the 1070.
the rx 480 is a desperate attempt of amd n00bs who cnat even make a driver to work properly to come back in the game
Guess what, they wont
next year this date, there will be 10000000000 threads of "i have a rx 480, why my fps is sooo low"...u know why ? Coz u were cheap and didn't want to shelve 150 more bucks for a card (1070) that would last u at least 3.5 years form now on

You mean a card that isn't meant to or priced to compete with the 1070 can't keep up with it? Who would have ever thought that?!?

Take a couple hours, think about what you did, then come back and apologize for your ignorance.

Love how you posted 3 times to agree with yourself in a sorry attempt to give yourself credibility.

 

[InvalidError]

Ugh, I hope that's not true. I know the tendency in the GPU business is for a huge drop between third- and fourth tier cards (GTX 960 to 950 and so on), but fully 50%? That would be awful, even if the power consumption dropped to 50-75W. There really ought to be at least one step between Polaris 11 and full Polaris 10, if the rumors about ~1280 shaders are true.

If the price difference between 470/1280 and 480/2304 ends up being only ~$60, there really won't be any room to price another GPU in-between. Producing a whole bunch of market fragmentation, especially ones which may require different silicon from planned products, for the sake of providing options in $20 price point increments would be a horrible waste of resources.

 

[iam2thecrowe]

GAMES WILL STILL BE RUN BETTER ON NVIDIA HARDWARE, ALWAYS
I will not play with teh performance of my pc and games, so i will gladly pay 150 more bucks to get the 1070.
the rx 480 is a desperate attempt of amd n00bs who cnat even make a driver to work properly to come back in the game
Guess what, they wont
next year this date, there will be 10000000000 threads of "i have a rx 480, why my fps is sooo low"...u know why ? Coz u were cheap and didn't want to shelve 150 more bucks for a card (1070) that would last u at least 3.5 years form now on

I may be a candidate for this rx480, i need to update my 7970. It's not that i don't want to put more money into it, its that i just cant afford it, along with many others. But if its going to be like current amd cards, that do really poorly in some titles, then i'm going to wait for nvidia's response to the mid-range cards, or snap up a cheap clearance or second hand 970.

 

[Valantar]

GAMES WILL STILL BE RUN BETTER ON NVIDIA HARDWARE, ALWAYS
I will not play with teh performance of my pc and games, so i will gladly pay 150 more bucks to get the 1070.
the rx 480 is a desperate attempt of amd n00bs who cnat even make a driver to work properly to come back in the game
Guess what, they wont
next year this date, there will be 10000000000 threads of "i have a rx 480, why my fps is sooo low"...u know why ? Coz u were cheap and didn't want to shelve 150 more bucks for a card (1070) that would last u at least 3.5 years form now on

yeah i am agree with u
all these ppl happy for 200 bucks vidoe card , and when will p[lay dark souls 3 their fps with rx 480 will be like 20 lmfao
NVIDIA 1070 IS BEST CARD FOR THE PRICE

ALSO, FACT:
No 1 cares of stupid VR anyway, what we care is good fps in normal pc games...thats where NVIDIA always win over trAsMD

Have to say I love the tactic of replying to yourself (in the very next post!) with "yeah i am agree with you." Both for the grammar and the wildly obvious fanboyish shilling. Sure, Nvidia makes great GPUs. So does AMD. And AMD drivers are really not bad. Sure, Nvidia beats them in terms of ekeing out every last drop of performance from their architectures. That's not news. But every single AMD card I've owned has been stable, has performed evenly across titles, and drivers very, very rarely contain noticeable bugs.

And the fact that a ~$400 card outperforms a $200 card? That doesn't surprise anyone. The 1070 and the 480 aren't competing in the same league. Anyone saying that (including the "the 480 should be ~15% below the 1070 because of FLOPS" crowd) are deluded. However, if the 480 performs on par with or better than the 980 and 390X, it's amazing value for money - and will have a lot of happy customers with enough performance for a few years.

Whether or not anyone cares about VR in a year or two is an interesting question, but it seems like it's here to stay - at least as a niche interest. Immersion and good experiences are attractive. It won't be nearly as big as esports or regular gaming, but a noticeable market nonetheless - and an important one for GPU makers. Or did you not pay attention to Nvidias intense VR focus with the Pascal launch?

Ugh, I hope that's not true. I know the tendency in the GPU business is for a huge drop between third- and fourth tier cards (GTX 960 to 950 and so on), but fully 50%? That would be awful, even if the power consumption dropped to 50-75W. There really ought to be at least one step between Polaris 11 and full Polaris 10, if the rumors about ~1280 shaders are true.

If the price difference between 470/1280 and 480/2304 ends up being only ~$60, there really won't be any room to price another GPU in-between. Producing a whole bunch of market fragmentation, especially ones which may require different silicon from planned products, for the sake of providing options in $20 price point increments would be a horrible waste of resources.

I'm not saying they should have custom chips, just that a cut-down Polaris 10 really ought to exist. Or do they not have production defects at all? Also, I disagree that there isn't room for another SKU within $60 - after all, third party coolers and other stuff already muddles the pricing quite heavily. The current R7 360 sits around $100, while the R7 370 starts from around $140, and the R9 380 around $180-200. In other words, their current lineup has room for three GPUs within less than $100. The same should be possible with Polaris. $130-140, $170 and $200 for full P11, cut-down P10 and full P10 respectively? With cut-down P10 limited to 2-4GB, possibly slightly slower GDDR5? That sounds reasonable to me, at least.