Official AMD Polaris Review Thread: Radeon RX 480, RX 470, and RX 460

[Azix]

I really think AMD had a turd in there hands but needed to push something out into the market. Nvidia is right behind them with the GP106. I don't know if the RX480 can hold them off if Nvidia is willing to engage in a price war. Nvidia can easily price the smaller die sized GP106 at $199 and would probably have better performance/watt and comparable performance.

Having said that, AMD did a great job of pricing their card. At $199, you won't find a better video card. Based on this merit, it is a win. But, AMD really needed a slam dunk. It needed something so disruptive, Nvidia will have nothing it can do but engage in a price war.

In my opinion, AMD is in big trouble moving forward. The performance per watt is utterly abysmal. It really is pretty bad. I don't think AMD has had this bad of a node shrink. It really is bad.

The engineering failure is already/going to have huge repercussions on their entire GPU stack this generation. Vega isn't looking too hot (or is it? jokes for days ). There goes the chances of dominating mobile. You can forget about that. Not with that kind of power draw.

The only saving grace here is getting GloFo all to themselves and thus, eliminating supply shortages. The same can't be said about Nvidia with TSMC. I don't think Nvidia would be able to meet all of the demands out there. That's a big plus for AMD right now.

perf/watt is a nice metric, but what matters most after a point is simply not using too much power. Its not like fps etc which have constant relevance. Anyway, perf/watt is at the same level as other cards, so not sure why its abysmal. When did better than 970 perf/watt become abysmal? Its not enough to be more efficient than maxwell without castrating the architecture? AMD gets no credit for what they are pulling off while nvidia took the easy road (just cut out the hardware)

there is no engineering failure. AMD and Nvidia have different architectures with different strengths. If nvidia were to add the hardware AMD has, who knows where they would be on "efficiency"

1060 is the card in a tough spot. It's already moved up and reduced in price from rumours. And there is next to no chance of it being better in dx12.

 

[krumme]

Anyone else finds it strange that the only other power draw demo of Polaris is the one we saw in January with P11 [roughly 1/2 P10] vs a 950 model and it appeared to have a much lower than 1/2 of the present power usage. This led a lot here to expect around 100-120W for RX480.

What might have happened or is happening now that has changed? Any ideas?

That pr spin was not good for anything. Probably just shareholder porn. I never deducted anything from it.
I dont know if it was that that led to the 110 w rumor. I dont think so? Reading the thread it just seemed to me it just popped out of the air and then it was repeated

I actually find it incredible we have this hard launch of finfet from gf. Its a huge progress from prior amd gen and looking at gf prior execution amd could have been 1 year after nv/tsmc. No probs. I wouldnt have bet my money on it beeing this early. Its damn fine we get compettition here and more capacity.

After i have been slamming gf for 8 years i actually think its appropriate to say congrats. Even if its not the finest 14nm process on this planet its a huge step forward and gibbo have his card so to speak.

 

[NomanA]

While the power usage of this card is more than what was generally expected, it'll be good to remind ourselves, that the power consumption is still roughly the same as what stock 970 and 980 consume (little more than 970, less than 980), whereas its performance is in the same ballpark as well, and better when considering DX12 games. Guru3D has a good collection of DX12 and DX11 games tested http://www.guru3d.com/articles-pages/amd-radeon-r9-rx-480-8gb-review,1.html

This is not a hot card. You won't be hit by heat coming out of the PC, as you would on an overclocked 7950 or 290x. Anyone who found 970 and 980 tolerable, should find this card equally so. Comments berating this card on "disastrous" power usage are quite ridiculous.

It's a great card for anyone looking to upgrade Radeon 380x and below or 960 and below, and at a very good price. Anyone paying $200-240 to get this card, in order to get 390/980 level performance, more so at 1080p, but also to some level at 1440p, should likely be satisfied. DX12 performance will also be better especially against Maxwell/Kepler cards.

Yes, there's a discussion worth having of how a more fleshed out chip in this same series can compete with 1080 (or 1080Ti), when Polaris is hitting 150-160 W already, but it's a parallel topic when related to the value of this card. Those who are buying a $200-240 card don't care what their next purchase 18-24 months down the road will be. Whether Vega can barely compete with 1070 or trounces 1080, it doesn't matter at all when analyzing what Polaris can currently achieve at $200-240 mark. Only the upcoming 1060 can affect Polaris value, and we don't know what its performance and especially price will be.

 

[Elixer]

Here is the Sapphire 480 card.

 

[MrTeal]

If I recall, it was severely controlled settings.

Throw in some 60 FPS cap, and lower settings, and you can make a GPU sip power.

Also, I'm of opinion AMD had to juice this card. Mix that with GloFlo and you got cards ranging to maintain the settings desired.

All in all, the QA for review copies was terrible. The one versions that should be golden samples.

From the demo:

From EightySix Four's post

There's going to be a pretty massive difference in efficiency between 1266MHz@1.13V and 850MHz@0.8375V.

 

[GodisanAtheist]

Guess its time to get worked up about the AIB releases next...

Looks like AMD made the decision to "hide" the poor performance of the GF 14nm node in the part where it would have the least impact. They get out from under the wafer agreement with a part that can compete on price rather than raw performance characteristics.

Favorable pricing (or at least not paying a penalty) for each chip plus exclusive access to a fab for the highest selling bracket (NV and AMD are likely small fries in TSMC's world) will allow AMD to compete on the one thing that really matters in this segment: Price.

When it comes to their high performance parts that will sell in substantially lower volume in segments where you're looking for any sort of edge to give you an advantage over the competition, they'll go back to the tried and true TSMC.

It was really the best move a cash strapped underdog like AMD could make. Lets hope the payoff works out as well for them.

As for myself: once the AIB cards hit, drivers mature a bit and prices settle I'll seriously consider the 480 as a replacement for my aging 7950.

 

[Eymar]

Yes, there's a discussion worth having of how a more fleshed out chip in this same series can compete with 1080 (or 1080Ti), when Polaris is hitting 150-160 W already, but it's a parallel topic when related to the value of this card. Those who are buying a $200-240 card don't care what their next purchase 18-24 months down the road will be. Whether Vega can barely compete with 1070 or trounces 1080, it doesn't matter at all when analyzing what Polaris can currently achieve at $200-240 mark. Only the upcoming 1060 can affect Polaris value, and we don't know what its performance and especially price will be.

Yeah it's too bad can't tag post so clear what area a post is targeting (technical, marketing, consumer, market share, etc). From a consumer stand point the 480 is a strong product against current offerings purely on price, performance, features. From a technical standpoint it is disappointing to me that more clock speed didn't come with process shrink, but hopefully that gets refined with next iteration of Polaris and Vega.

 

[Cookie Monster]

So after reading the reviews.. Kyle was actually on point?

Joking aside, the RX480 is abit underwhelming given the hype (mostly user created). I guess this is showcasing the R&D budget differences between nVIDIA and AMD. If Polaris went up against Maxwell, it would have been great. But Pascal is in its own level right now.

I think they went far too high on the frequency/voltage curve as some of the review sites managed to see better perf/w figures with it being downclocked.

 

[krumme]

I'm pretty sure I already said I understand them doing that. I'm not judging them for doing it. I'm just contesting someone's blanket statement because they seem to be unhappy with AMD's decisions. Perhaps respond to them.



A lot of us buy into the hype. More so when you want your team to win. It seems every year I buy into the AMD hype. Even this year when I said I'm strictly neutral I still debated picking up a 480 based on the rumors of high OC's.

With AIB's to see what they can do, it isn't completely off the table. But something tells me they won't be getting that much further from reference model. If this card is already chewing through 150W, I can only imagine what AIBs have to put up with.



My HD 7970 was tolerable. But I replaced the fan. My 980 Ti was not. However, I replaced the fan. In both situations I got lucky, cost me less than an AIB to replace the fans. End results I had quiet cards that OC'd like the dickens, both reference models.

Back to the blanket statement I responded to - not all reference models are garbage.

I hoped the aib card could oc to 1500 and most come 1450 out of the gate. That was wrong. No way looking at theese reviews. Its probably 8-10% below that. And thats the dissapointment besides the 160 w whatever tdp that is 30w above ny expectations. Low oc is a blow imo for me as its fun. Sure the aib will take care of the blower but power is worse than expected. Not that it mattered for desktoo if it wasnt for oc headroom as you say.

I dont know where some of the disapointment or meh for perf comes from.
Tpu that uses a huge selection of bm and that we have used as basis for discussions for years in this forum and especially the 480 thread says this card is 100% 390x at 1080. In a selection of games that ranges from mostly dx11 but also to the few dx12. Why on earth is this card then labelled as 970 perf or under 390 perf? What crap is that? When there was talk about 390x perf most was rightly sceptical. Then it just ends at that level. Damn fine. And as tpu shows it result in a slam dunk perf value rating.

 

[maddie]

Not the same with games. Eth's algorithm was specifically designed to scale bandwidth relatively linearly with compute, this card just falls on the wrong end of that scale. It is bandwidth starved for that algo, not necessarily everything it does.



Based on how aggressive the voltage ramping is, they probably are pushing the silicon a bit harder than they should. Power usage drops off a cliff when you underclock/lower the voltage. That also explains the 6 pin connector on the reference card and the cooler. Made the same exact mistake as the 290x. Pretty amazing really.

Hopefully, like the 290x, this card will have some legs over the long term.

If I recall, it was severely controlled settings.

Throw in some 60 FPS cap, and lower settings, and you can make a GPU sip power.

Also, I'm of opinion AMD had to juice this card. Mix that with GloFlo and you got cards ranging to maintain the settings desired.

All in all, the QA for review copies was terrible. The one versions that should be golden samples.

OK, but I'm seeing some reviewers getting about 25-45 Watts less system power usage vs a 970 powered one. It's quite a bit harder to make a mistake on total system power readings vs component power consumption, so this appears quite confusing. CPU usage should be similar in both cases, so that doesn't appear to be the problem.

Could there be such a vast variation in individual cards?

hardwarecanucks get 24W less
techreport gets 2W more
anandtech gets 12 less
Hardocp gets 53W less
computerbase.de gets 26W less

all of the above is total system power

techpowerup gets 7W more
Toms gets 7W less
guru3d gets 12W more

these do card power

 

[iiiankiii]

Guess its time to get worked up about the AIB releases next...

Looks like AMD made the decision to "hide" the poor performance of the GF 14nm node in the part where it would have the least impact. They get out from under the wafer agreement with a part that can compete on price rather than raw performance characteristics.

Favorable pricing (or at least not paying a penalty) for each chip plus exclusive access to a fab for the highest selling bracket (NV and AMD are likely small fries in TSMC's world) will allow AMD to compete on the one thing that really matters in this segment: Price.

When it comes to their high performance parts that will sell in substantially lower volume in segments where you're looking for any sort of edge to give you an advantage over the competition, they'll go back to the tried and true TSMC.

It was really the best move a cash strapped underdog like AMD could make. Lets hope the payoff works out as well for them.

As for myself: once the AIB cards hit, drivers mature a bit and prices settle I'll seriously consider the 480 as a replacement for my aging 7950.

That makes sense. I agree. It looks like GloFo is continuing to have trouble. The RX480 doesn't need to be super low powered, and/or super high performance. Although, I expected better from a node shrink.

For its targeted audience, it just need to give the them what matters most to them; Perf/$. In the end, that is the most important metric at the $200-$250 range.

 

[Elixer]

I've created 4 "new" jet.com accounts and quadruple dipped their 15% off for first time users on various products. Would recommend a double dip if you intend to CF.

Heh, I suppose that is one way to get cards cheaper.
I officially hate Jet now. First order went fine. The second one they kept cancelling because of "fraud" (and no, I didn't to the double dip of the discount). Idiots don't know what they are doing, each customer service person is clueless, and they keep telling you wait a few days and call again. They never called back, the e-mail they kept saying if this isn't resolved, please give it X more days, and round and round they went.

I'll see if I can find a review with h.265 encoding testing. If there isn't one, are there any particular benchmarks you'd like me to run? I have some 4GB cards on the way for testing and they could be arriving as early as Friday.

Do you have access to 1080 & 4k content that you can encode?
You can download this http://bluesky23.yukishigure.com/en/AsVideoConv.html (it is free) and try encoding some stuff. This is the only free one that uses AMD's hardware encoder, but, I am not sure it will work with the h.265 encoder. I know it works with the h.264 though.
TIA!

 

[Cookie Monster]

Also do we have a clear reason as to why the 14nm GF process is at fault here? Seems like the general consensus is that the process used is the primary cause for the high power consumption..

Compared to Hawaii/Fiji/Tonga, it isn't too bad. Against Maxwell/Pascal however is a different story.

 

[boozzer]

anyone knows when AIB 480s will be out? those will be interesting.

 

[dark zero]

Also do we have a clear reason as to why the 14nm GF process is at fault here? Seems like the general consensus is that the process used is the primary cause for the high power consumption..

Compared to Hawaii/Fiji/Tonga, it isn't too bad. Against Maxwell/Pascal however is a different story.

Mmm... maybe the FinFet hype is not as real as expecting? Maybe time to wait to EUV after all...

 

[Bacon1]

Total system power is actually better, because it takes into account memory, cpu and everything else that is running while the cards are. Afterall, no one games without any of those

They also need to do their testing while gaming, not furmark or other garbage tests.

Computerbase seems to have done the best benchmarking I've seen. They've done an amazing amount of testing

Anyone who cares about this card should read their review, massive amounts of information.

https://www.computerbase.de/2016-06/radeon-rx-480-test/

 

[PeckingOrder]

it seems the PCI-E slot power issues are more than real

source: reddit.com/r/amd

Hardware.fr confirms this too. They also have a retail version (Sapphire 480) which exhibits the same problem. They also confirms the power usage going over 150W with both a review and retail version of the card.
They added also something interesting, they removed power and temperature limits and tested the card with no OC. The card pulled almost 200W in Witcher 3.

Source: http://www.hardware.fr/articles/951-9/consommation-efficacite-energetique.html

Relevant parts:

Rappelons, comme expliqué dans le descriptif de la carte de référence, que la RX 480 n'est équipée que d'un seul connecteur d'alimentation 6 broches, ce qui fait qu'il tire une grosse partie de sa consommation, à peu près la moitié, parfois un peu plus, du bus PCI Express. Elle va à ce niveau bien au-delà de la spécification qui est de 5.5A. Dans Battlefield 4, nous mesurons 6.92A par défaut et 7.10A en 'Uber'. Une valeur qui monte à 7.79A dans The Witcher 3 et qui impose un stress pour lequel toutes les cartes-mères ne sont pas prévues.

"RX 480 comes with a 6 pins PCIE and most of the power drain is shared between the 6 pins and the PCI Express port (sometime more on the latter). The cards does go way over the specs which is 5.5A. In Battlefield 4, we are getting 6.92A and 7.10A with OC (called "Uber" mode in the review). Witcher 3 goes even higher with 7.79A. This stress is not expected on all motherboards."

 

[EightySix Four]

it seems the PCI-E slot power issues are more than real

source: reddit.com/r/amd

Hardware.fr confirms this too. They also have a retail version (Sapphire 480) which exhibits the same problem. They also confirms the power usage going over 150W with both a review and retail version of the card.
They added also something interesting, they removed power and temperature limits and tested the card with no OC. The card pulled almost 200W in Witcher 3.

Source: http://www.hardware.fr/articles/951-9/consommation-efficacite-energetique.html

Relevant parts:

Rappelons, comme expliqué dans le descriptif de la carte de référence, que la RX 480 n'est équipée que d'un seul connecteur d'alimentation 6 broches, ce qui fait qu'il tire une grosse partie de sa consommation, à peu près la moitié, parfois un peu plus, du bus PCI Express. Elle va à ce niveau bien au-delà de la spécification qui est de 5.5A. Dans Battlefield 4, nous mesurons 6.92A par défaut et 7.10A en 'Uber'. Une valeur qui monte à 7.79A dans The Witcher 3 et qui impose un stress pour lequel toutes les cartes-mères ne sont pas prévues.

"RX 480 comes with a 6 pins PCIE and most of the power drain is shared between the 6 pins and the PCI Express port (sometime more on the latter). The cards does go way over the specs which is 5.5A. In Battlefield 4, we are getting 6.92A and 7.10A with OC (called "Uber" mode in the review). Witcher 3 goes even higher with 7.79A. This stress is not expected on all motherboards."

I'm pretty shocked my mobo is holding up under mining loads if this is the case.

 

[Silverforce11]

Let's sum up this thread:

People with team alliances, spinning to make any molehill they can find into a mountain.

And in the real world, people buy $200 GPUs and this is much faster than what $200 would buy you yesterday.

That's not true. I'm usually very favorable towards AMD but this is a failure. 14nm FF, such a small chip, should be not chewing 150W during gaming.

Compare to it's predecessor of this class, anyone remember Pitcairn, the 7850 and 7870?

Polaris 10 is worse when factoring in the node jump.

That tells me GloFo failed them. Call it what it is, but FinFet should NOT be struggling to get 1.26ghz with massive power consumption. The entire point of the FinFet transistor is to minimize current leakage, allowing it to operate at a higher clockspeed & higher voltage tolerances.

Now, I would have thought the RX 480 a much better product if it's gaming load was ~100-110W. That would imply AMD low ball clocks to get perf/w, leaving more performance on the table for overclockers or custom cards. But it's right at the edge.

As a gamer, I still think it's a great GPU at the price.

* I bought 2x RX 480 8GB, $379 AUD each.

GTX 970 3.5GB & AMD 390 8GB are ~$449 AUD. GTX 980 4GB is $629 AUD (got a price cut last week from its usual $749!!).

390X 8GB is $529 AUD. 1070s are ~$779 here and 1080s are $1199, ridiculous prices.

Logically, you can't say RX 480 is a bad GPU for the price. It is good for gamers to have that performance class down at mainstream prices.

But as a tech enthusiast, I am very disappointed at seeing such a small FinFet chip suck down that much power. To me, that's a failure, most likely GloFo but in the final analysis, AMD takes the blame because they should have known better and be more honest about expectations.

You don't get to stand there and claim 2.8x perf/w and talk about all this efficiency and coolness you get from 14nm FF, when the card runs 82C and at the limits of its power PCB.

I can tell you right now with facts, that 1.26ghz is operating beyond it's optimal clocks for the process. Why? Look here:

1.4ghz OC with a aftermarket cooler:

http://oc.jagatreview.com/2016/06/t...deon-rx480-ke-1-4ghz-dengan-cooler-3rd-party/

Power usage jumps to 183W, which is insane for such a small clock speed bump.

All this screams that AMD was forced to clock it outside it's optimal zone, because the node is giving them such a bad result.

I raised these points in the other thread and some of you accuse me of being negative on AMD (falsely even). But, AMD don't get to go to a new node AND HYPE UP efficiency gains and talk about 2.8x perf/w and be so far being Pascal on perf/w.

This is what my logic tells me, I don't need to sugar coat the analysis because I am not a blind fanboy.

 

[PeckingOrder]

Total system power is actually better, because it takes into account memory, cpu and everything else that is running while the cards are. Afterall, no one games without any of those

They also need to do their testing while gaming, not furmark or other garbage tests.

Computerbase seems to have done the best benchmarking I've seen. They've done an amazing amount of testing

Anyone who cares about this card should read their review, massive amounts of information.

https://www.computerbase.de/2016-06/radeon-rx-480-test/

agreed, I enjoy their work. the only problem is that it's in German, which makes it pretty hard to read for the majority of this forum

Google translate results are horrible most of the time when translating German to another language

 

[boozzer]

is there a reason in removing the limits? did fps increase?

 

[IEC]

agreed, I enjoy their work. the only problem is that it's in German, which makes it pretty hard to read for the majority of this forum

Google translate results are horrible most of the time when translating German to another language

If there is a specific part you would like a better translation on I can translate it.

 

[Thala]

Computerbase seems to have done the best benchmarking I've seen.

At least they test with Crimson 16.20.1035.1001-RC1 and have increased the power target, such that 1266MHz is guaranteed for all games.
They also figured, that for each 2% bandwidth increase you gain 1% performance, which is huge -> GPU is bandwidth limited for large parts of the execution time.

 

[Yakk]

anyone knows when AIB 480s will be out? those will be interesting.

Rumor is mid July, per Powercolor's Devil cards giveaway date, and Asus Stix & MSI gaming cards were seen and they appear quite good indeed.

 

[Cookie Monster]

That's not true. I'm usually very favorable towards AMD but this is a failure. 14nm FF, such a small chip, should be not chewing 150W during gaming.

Compare to it's predecessor of this class, anyone remember Pitcairn, the 7850 and 7870?

Polaris 10 is worse when factoring in the node jump.

That tells me GloFo failed them. Call it what it is, but FinFet should NOT be struggling to get 1.26ghz with massive power consumption. The entire point of the FinFet transistor is to minimize current leakage, allowing it to operate at a higher clockspeed & higher voltage tolerances.

Now, I would have thought the RX 480 a much better product if it's gaming load was ~100-110W. That would imply AMD low ball clocks to get perf/w, leaving more performance on the table for overclockers or custom cards. But it's right at the edge.

As a gamer, I still think it's a great GPU at the price.

* I bought 2x RX 480 8GB, $379 AUD each.

GTX 970 3.5GB & AMD 390 8GB are ~$449 AUD. GTX 980 4GB is $629 AUD (got a price cut last week from its usual $749!!).

390X 8GB is $529 AUD. 1070s are ~$779 here and 1080s are $1199, ridiculous prices.

Logically, you can't say RX 480 is a bad GPU for the price. It is good for gamers to have that performance class down at mainstream prices.

But as a tech enthusiast, I am very disappointed at seeing such a small FinFet chip suck down that much power. To me, that's a failure, most likely GloFo but in the final analysis, AMD takes the blame because they should have known better and be more honest about expectations.

You don't get to stand there and claim 2.8x perf/w and talk about all this efficiency and coolness you get from 14nm FF, when the card runs 82C and at the limits of its power PCB.

I can tell you right now with facts, that 1.26ghz is operating beyond it's optimal clocks for the process. Why? Look here:

1.4ghz OC with a aftermarket cooler:

http://oc.jagatreview.com/2016/06/t...deon-rx480-ke-1-4ghz-dengan-cooler-3rd-party/

Power usage jumps to 183W, which is insane for such a small clock speed bump.

All this screams that AMD was forced to clock it outside it's optimal zone, because the node is giving them such a bad result.

I raised these points in the other thread and some of you accuse me of being negative on AMD (falsely even). But, AMD don't get to go to a new node AND HYPE UP efficiency gains and talk about 2.8x perf/w and be so far being Pascal on perf/w.

This is what my logic tells me, I don't need to sugar coat the analysis because I am not a blind fanboy.

Or one could also blame the GCN architecture which has NEVER been power consumption friendly especially for the performance parts (regardless of its superior DX12 perf). And its been awhile since AMD made big changes to its uarch like nVIDIA has done the past couple of generations starting with Fermi.