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

[.vodka]

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.

100% agree. Although the 480 brings great value to the $200-230 price point replacing overpriced turds like the 960 and to a lesser extent the 380/380x and becomes my immediate recommendation for anyone who is on a budget, as a product itself is quite the failure for the enthusiast type of user on a budget. No overclocking potential and power consumption goes through the roof if you try to do so. No aftermarket cooling is going to fix that, unlike Hawaii. The GPU itself is to blame here.

GF has failed hard, sincerely I don't think anyone expected such a level of failure from them, even considering they had help to get their 14nm node running.

Polaris as an architecture certainly is good and a nice upgrade over previous GCN versions, a lot less resources (ROPs and memory bandwidth in particular) with a little clock speed bump equaling a 390/390x in most cases while halving power. This is what was expected in the peformance front yet the final product then is lacklustre because power characteristics aren't what you would expect from a 14nm class product (vs Intel's and TSMC's versions of the node and products)

It's even worse when you think that this very foundry is going to make Vega/Navi and Zen. Logic says one could expect P10 to be some kind of pipe cleaner to refine the node for the next products because as it is, it's not viable. Too much variance, it's unacceptable. Even the HD4770 as a pipe cleaner for TSMC's 40nm node wasn't this bad for being one of the first products out of the gate at that time.


Vega family chips had better come out of the gate with these underlying problems fixed... I wonder how P10 could have been if made on TSMC's 16nm process, I'm pretty sure power wouldn't jump to 183w for a paltry 150MHz overclock.

 

[deanx0r]

I am kind of disappointed that AMD barely caught up with Maxwell in term of efficiency. It uses as much power as a 1070 while being much slower. It might be the best option for the 200-300 price bracket now, but with 390 and 970 being EOL, what is going to happen when NVIDIA drops the 1060? AMD wouldnt have to compete on pricing if they were competitive elsewhere.

 

[Elixer]

67DF:C7 is the version of the chip not the stepping. It's the same as saying RX 480. 67DF:C4 is likely the RX 470 for instance.

I'm not totally convinced this is completely or at all a GloFo issue. Whatever the bug is that is causing the card to over draw more juice from the PCI-E slot could very well be pulling gobs more power than the chip/board need. AMD is already looking into it. Possibly a bug in that new BTC function? Probably a BIOS update coming shortly.

I thought GPU-Z showed stepping...

I wasn't talking about the PCIe power draw issue, that might be a symptom of it, but, there is a power gremlin at work here above that. This node process shouldn't be using that much power.

 

[Stormflux]

I am kind of disappointed that AMD barely caught up with Maxwell in term of efficiency. It uses as much power as a 1070 while being much slower. It might be the best option for the 200-300 price bracket now, but with 390 and 970 being EOL, what is going to happen when NVIDIA drops the 1060? AMD wouldnt have to compete on pricing if they were competitive elsewhere.

As much as people are lamenting the fact that AMD JUST caught up to nVidia's Maxwell in terms of efficiency... AMD also did it refining their current architecture. Which seems key to their current strategies and beneficial for all GCN owners. And when comparing though the GCN iterations, the gain IS pretty substantial.

When nVidia made the big jump in efficiency, they dumped Kepler and made Maxwell. Where are the Kepler cards now? Is the same thing to happen to Maxwell now that Pascal is out?

Double edged sword sticking with GCN? You bet but, given what happened to Kepler...

The wild variances in power consumption from reviews are highly questionable and needs further investigation, definitely.

 

[maddie]

67DF:C7 is the version of the chip not the stepping. It's the same as saying RX 480. 67DF:C4 is likely the RX 470 for instance.

I'm not totally convinced this is completely or at all a GloFo issue. Whatever the bug is that is causing the card to over draw more juice from the PCI-E slot could very well be pulling gobs more power than the chip/board need. AMD is already looking into it. Possibly a bug in that new BTC function? Probably a BIOS update coming shortly.

These are my thoughts also.

The new power saving tech used might be causing many problems, not the 14nm node from GloFlo. We might be having conflicts between BTC, AVFS and adaptive clocking. Something very strange is happening when you have a very wide variation in measured power draw between review sites.

From 53W [Hardocp] less than a 970 system to 12W more [guru3d]. A 65W measurement spread for a 150W card is HUGE and can't be explained away with traditional reasons.

 

[Gorbugal]

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.

Agreed 100%, it represents good value for money, for now, but the performance per watt and the overclocking headroom are seriously disappointing.

I still think Nvidia would have an efficiency advantage even if this was produced on TSMC 16nm but I think the majority of the clocking/efficiency failures are down to GloFo.

That said everyone I know who games doesn't care about any of these metrics except for fps/dollar so maybe they will sell OK?

 

[Janooo]

...
When you look at the power efficiency of Fermi (28nm) -> Kepler (28nm) -> Maxwell (28nm) -> Pascal (16nm) they've all shown significant perf/w gains each generation (with a uarch changes to do so). But in terms of Tahiti (28nm) -> Hawaii (28nm) -> Fiji/Tonga (28nm) -> Polaris (14nm) its quite disappointing because the gains have been relatively been flat.
...

Fermi was 40nm part.

 

[Silverforce11]

That said everyone I know who games doesn't care about any of these metrics except for fps/dollar so maybe they will sell OK?

It will sell great, until the 1060 comes.

960 vs 380 all over again. Similar perf, 960 wins on perf/w. Importantly, 960 has low power enough to enter crap OEM systems as easy upgrades while 380 could not.

That's the RX 480, 150W is too much for mainstream.

 

[Cookie Monster]

Fermi was 40nm part.

True that. Thanks for the correction

 

[SimianR]

Has there been any confirmation that AMD is using Goflo for Vega? Will be interesting to see if they end up going with TSMC what the yields/power usage is like in comparison.

 

[Stormflux]

Where are the details on the RX 460 and RX 470. Seriously disappointing to not see these in reviews, stores or even mentioned. When do the NDAs drop for these and AIB cards?

 

[IEC]

Where are the details on the RX 460 and RX 470. Seriously disappointing to not see these in reviews, stores or even mentioned. When do the NDAs drop for these and AIB cards?

We will likely know more details in the coming weeks. Several RX 480 custom card teasers from ASUS/MSI/Sapphire/PowerColor.

 

[deanx0r]

As much as people are lamenting the fact that AMD JUST caught up to nVidia's Maxwell in terms of efficiency... AMD also did it refining their current architecture. Which seems key to their current strategies and beneficial for all GCN owners. And when comparing though the GCN iterations, the gain IS pretty substantial.

When nVidia made the big jump in efficiency, they dumped Kepler and made Maxwell. Where are the Kepler cards now? Is the same thing to happen to Maxwell now that Pascal is out?

Double edged sword sticking with GCN? You bet but, giving what happened to Kepler...

The wild variances in power consumption from reviews are highly questionable and needs further investigation, definitely.

Was there truly a performance regression with Kepler products? I think Linus made a video with Fermi and Kepler cards and concluded that while drivers releases improved performance over time, they weren't very significant (5-10% at best).

Immature drivers with GCN may be untapped potential, but that's a pretty big gamble and a hard sale for someone looking to get most of his cards now.

 

[atakall]

I was looking to sidegrade from my GTX970 (needed HDMI 2.0) that powers my 4k TV just to lower its power consumption. I wanted 110watts. I am definitely not doing that now. Lets not forget, the idle power consumption on the RX480 is hideous, too!! Man, GloFo sucks big time.

It's been posted (earlier in this thread, iirc) that the poor idle power consumption is due to a driver issue.

 

[Shivansps]

Yes, and when GP106 arrives, imagine half a GP104, it's going to be around 100W gaming load and at the performance of between 970 and 980.

It's going to basically match the RX 480 and beat it in perf/w by 50%.

What this equate to for some of you who are still not appreciating why I have kept on raising perf/w...

THINK about what made the 750Ti and 950/960 so popular. It's not just because it's NV branding. It's because they were very low power. They could go into any crap OEM system and be just fine. No need for a PSU upgrade.

You simply cannot say that about 380 power class that the RX 480 falls into it. It's in another tier higher.

The 1060 will automatically win that mainstream battle because the RX 480 is power hungry for mainstream. It will be the 960 vs 380 situation all over again. Similar perf, but the 960 won that fight due to perf/w.

As for Vega/Polaris, Polaris is IP8, Vega is IP9, a new architecture. Don't draw comparisons to Vega based on Polaris, it will not be accurate. And they also need to hurry up on Vega, once 1060 is out, RX 480 sales will die down and they have to compete against NV that has the mainstream, mid-range and performance segment with strong products.

You are right here, i build pcs and thats what keeps 370 and 380 out of game because the crap generic PSU that comes with the cases cant handle them, but they do play well with up to GTX 960, i was kinda hopping for RX480 to drop in power compared to 380 so i can finally consider them, but no, at this point the 1070 is closer to work with those generic psu than the RX480, and thats is not a good thing.

I hope for RX470 to at least drop to 960 levels.

 

[SimianR]

Was there truly a performance regression with Kepler products? I think Linus made a video with Fermi and Kepler cards and concluded that while drivers releases improved performance over time, they weren't very significant (5-10% at best).

Immature drivers with GCN may be untapped potential, but that's a pretty big gamble and a hard sale for someone looking to get most of his cards now.

I don't think it was a regression in existing titles, it was in newly released games kepler performed worse relative to GCN cards. The performance hierarchy changed up over time.

 

[tviceman]

It's going to basically match the RX 480 and beat it in perf/w by 50%.

With a smaller die, too. And at equal performance, Nvidia will set the MSRP at $250-260 as well, meaning they are making noticeably more money per chip/card sold with higher prices and lower costs. AMD's margins continue to get squeezed and R&D continues to suffer as a result.

It's too late to say woulda coulda shoulda but AMD needs to back off the high end again and bring out a technologically competitive mid-range die that can scale well as an x2 card to compete with Nvidia's high end. Had P10 been within 20% of GP104's efficiency, an RX 480 would be only 20% slower than GTX 1070, could have been priced at $250, and would easily defeat GTX 1080 as a $500 x2 card.

AMD's continued console-update wins will help alleviate the technical hurdles vs. Pascal somewhat, but I don't think Pascal is going to lose ground as much as Maxwell did (and certainly not as much as Kepler). If Pascal holds up well and Nvidia's technical advantages keeps them steam rolling the market, we could be looking at AMD's last node it manufactures discrete GPU's on.

 

[Stormflux]

Was there truly a performance regression with Kepler products? I think Linus made a video with Fermi and Kepler cards and concluded that while drivers releases improved performance over time, they weren't very significant (5-10% at best).

Immature drivers with GCN may be untapped potential, but that's a pretty big gamble and a hard sale for someone looking to get most of his cards now.

No, there was no regression. Just neglect. When the 290X was released, they were going toe-to-toe with 780Tis with games at the time. Now 290Xs are easily up to 20% faster in today's games as an example.

 

[Jaskalas]

Claim: RX 480 fails PCI-E specification

Some power draw / consumption issues on more than one review card.
Maybe the launch is not without a hiccup.

 

[Yakk]

I'll be getting either one of these. Just hope the price isn't above $300. It's odd and may seem stupid but if I'm going to pay over $300 with tax I feel as if I should spend more and get the $399 1070...I know doesn't seem to make sense but oh well.

Yeah I've decided on a AIB version.

http://wccftech.com/amd-rx-480-asus-strix-msi-gaming/

Should still be less than $300 at least I hope so, the Canadian exchange rate is horrible right now which makes anything more a no-go for me even with the mining it's going to pay for itself.

 

[sze5003]

http://pcpartpicker.com/products/video-card/#c=369&sort=a8&page=1 450$ for the cheapest 1070

maybe you should take your own advice and get a new one

Oh what he mentioned was in reference to my post about AIB 480 versions being $300 or a bit more. With tax for me that would come to about $330 maybe $350 depending on initial price. In that case I'd rather get the $399 gigabyte 1070.

 

[felix5]

..

 

[sze5003]

Will I notice an improvement with 8GB over the 4GB version of this card if i am trying to utilize 144hz/freesync at 1080p or 1440p?

The 8gb has actual 8ghz vs the 4gb at 7ghz so more memory would be useful for higher resolution depending on the games but it will definitely be fine at 1080p. The reviews state it does fine at 1440p but not with the same settings on, it does struggle a bit at that resolution.

 

[boozzer]

Oh what he mentioned was in reference to my post about AIB 480 versions being $300 or a bit more. With tax for me that would come to about $330 maybe $350 depending on initial price. In that case I'd rather get the $399 gigabyte 1070.

we know nothing about AIB 480s so far or their prices or their performance.

and where do you shop that you can find a 399$ 1070? I am genuinely curious. and wouldn't shipping and tax apply to 1070 too? no need to include that in your calculations.

 

[tviceman]

deleted