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

[Glo.]

What is becoming apparent is that this GPU is made for poor people.

AMD PR is absolutely terrible. They could have released not buggy driver in the first place, so the GPU performance could go up by 5%. And then they could lower the voltage so that GPU could not get into thermal limit, and draw over 30 watts less, so the general perception of this GPU would have been MUCH, MUCH better.

 

[littleg]

Not sure if it was mentioned. If you read the Computerbase.de article carefully they say:

AMD hatte ursprünglich den Crimson 16.6.2 zum Testen zur Verfügung gestellt. Wie ComputerBase jedoch an diesem Montag erfahren hat, hat dieser mit einem Bug zu kämpfen, der die PCIe-Bandbreite limitiert. AMD hat ComputerBase daraufhin den Crimson 16.20.1035.1001-RC1 zur Verfügung gestellt, der das Problem behebt. Alle Werte wurden daraufhin noch einmal erhoben, denn je nach Spiel steigt die Performance durch den neuen Treiber um bis zu fünf Prozent an. Im Durchschnitt ist die Radeon RX 480 etwa 1,5 Prozent schneller.

They got informed by AMD on Monday, that the 16.6.2 driver has a bug with limits PCIe bandwidth. Therefore they got Crimson 16.20.1035.1001-RC1 which solved the problem and noticed a performance increase of up to 5% compared to 16.6.2.
From what i read, most other reviews just used the buggy 16.6.2 driver.

Why does every AMD launch have to be a massive exercise in facepalming? Surely they tested this before they sent out the drivers?

It's pretty disappointing in terms of power at the moment but that sounds like maybe it's a BIOS issue that they might be able to fix. The throttling and overclocking headroom is a bit of a joke though. 1080 got panned for the same issues and this is just as disappointing.

 

[TeknoBug]

That means, you managed to buy 3 of the worst NV architectures of all time:

FX5800/5900 series = pure trash
GeForce 7 = more trash
Fermi = the very architecture that failed in nearly every metric you keep discussing in modern times (perf/mm2, perf/watt, price/performance)

You forgot the FX5600 and FX5700/LE versions, omg they sucked, they couldn't even draw correctly in some games (walls in some buildings were missing in Star Wars Galaxies for example which prompted me to go out and buy the Radeon X700Pro at the time).

 

[Magic Carpet]

Supposedly the 470 is actually the most efficient, 110W and it's what AMD used for it's claims of up to 2.8x perf/w. Look at their slides again, specifically a 470, not 480.

Since both cards feature a single power plug, I'd say they both can be efficient. Just downclock/undervolt some. Been this way with AMD for as long as I can remember. If you want OOBE, buy Geforce. Simple as.

Great card otherwise :thumbsup:

 

[Keysplayr]

This is how big companies work. There is more to a launch than just a low number of driver bugs.

In AMDs case, though, it's best foot forward time.
And, this really isnt how big companies work.
Extensive testing should have materialized any memory bugs months and months ago. Not after launch day when benchmarks are done.

 

[Janooo]

LOL! Coming from a guy who bought a GTX980 for $550-600 less than 2 years ago. Today, AMD's $199-229 card is within 10-15% of that overpriced garbage 980! Generation after generation these relative concepts seem completely foreign to you.
...

RX480 is almost 10% faster than 980 in DX12!
Custom 480's will be even more.

http://www.hardwarecanucks.com/foru...9-radeon-rx480-8gb-performance-review-24.html

 

[Janooo]

But there are people who think that a 970/980 at the same price as a 480 is a good purchase.
Just because this has forced a price cut by the competition doesn't make those cards good value. They are going to die a fast death just like Kepler did. As of now there's no reason to believe that Pascal's fate will be any different. People who should know better though will keep recommending them.

Let them stay in their mistaken point of view.
There is nothing that can be done about it.

 

[boozzer]

What is becoming apparent is that this GPU is made for poor people.

AMD PR is absolutely terrible. They could have released not buggy driver in the first place, so the GPU performance could go up by 5%. And then they could lower the voltage so that GPU could not get into thermal limit, and draw over 30 watts less, so the general perception of this GPU would have been MUCH, MUCH better.

haha 200-240 market segment is poor people haah

 

[Glo.]

haha 200-240 market segment is poor people haah

Well that is general "melody" that spreads on forums . Not my opinion .

 

[MrTeal]

In AMDs case, though, it's best foot forward time.
And, this really isnt how big companies work.
Extensive testing should have materialized any memory bugs months and months ago. Not after launch day when benchmarks are done.

Assuming that the bug existed the whole time. Is the PCIe bandwidth bug present in earlier drivers, or was it introduced in the June driver drop?

 

[gamervivek]

Videocardz have a collection of overclocked results and point out that overclocking the memory can lead pretty good gains on the card. If AMD sort out the clocking woes and have a GDDR5X controller as well, Polaris 10 can jump 15-20% in performance easily and perhaps be reused as a new card.

The benchmarks are clear: the increase in memory bandwidth by 15% affects in a performance gain 7-11%. This is quite remarkable, especially when you make yourself fully aware that the card would move ahead of some competitors.

http://videocardz.com/61697/how-fast-radeon-rx-480-can-be

 

[iiiankiii]

Videocardz have a collection of overclocked results and point out that overclocking the memory can lead pretty good gains on the card. If AMD sort out the clocking woes and have a GDDR5X controller as well, Polaris 10 can jump 15-20% in performance easily and perhaps be reused as a new card.



http://videocardz.com/61697/how-fast-radeon-rx-480-can-be

Damn, I might be disappointed on the power draw. But, I'm getting excited on grabbing an AIB with better components, cooling and power delivery. I want to see how far I can push this thing. A 1500mhz/9000mhz RX480 @ ~$250 is still pretty appealing regardless of the power draw.

 

[Stuka87]

It really doesn't instill confidence when they still went ahead with their launch knowing their driver had a pci-e bandwidth limiting problem.
You cant tell me that throughout all their testing, preparing this product for a launch, that they didnt know if there was a bug or not. Im guessing there was no bug.

1: Kernel bugs are not always cut and dry, easy to find. Typically they are hard to find and test for. I know, because its part of my job.

2: nVidia has their own launch bugs with Pascal, so its not like AMD is alone in having a bug.

 

[RussianSensation]

RX 480 4GB $259 CDN, now you're talking.

:thumbsup:

I think $199 RX 480 4GB is actually the one I am going to recommend for single GPU use in the mainstream category (until GTX1060 6GB shows up). The 8GB model seems like a questionable way to spend $40 that can be used towards a future GPU upgrade instead. I would also easily recommend for a new PC builder to invest that $40 towards a better PSU or step up from an i5 6500 to i5 6600K or get him/her closer to i7 6700/K from i5 6600K. The investment into a faster CPU or a better PSU is $ well spent in this case imho.

The RX 480 doesn't seem powerful enough to truly benefit from > 4GB of VRAM as its performance falls in the same tier as 390/390X/970/980 level cards. In 2 years, that $40 can be used to step up from a $200-210 to a $240-250 card. I think it'll matter more in that case than the 4GB vs. 8GB today. GTX1060 3GB seems DOA as I won't recommend a 3GB card in 2016 in the $200 price range.

Also, I hate blower cards. It's smart to wait for AIB RX 480s.

- 0dBA idle/desktop use
- cool
- quiet
- more overclocking headroom
- DVI (I like having this option)
- easier to dust clean over time
* Newer Sapphire cards should use double ball bearing fans

No brainer. I don't understand why people have no patience to wait a month for a good product when they intend to use this for 2+ years.

Looks like AMD is finally adopting a logical naming scheme.

RX 1st number = 4 = generation
RX 2nd number = x60, x70, x80, x90 = performance tier
RX 3rd number = revision (we could see refreshes called RX 465, 475, 485, 495, etc.)

haha 200-240 market segment is poor people haah

The difference between objective/informed PC enthusiasts and brand loyalists is that the former recognize that the $200-240 dGPU's target market is unlikely to have CPUs fast enough to take advantage of GTX1070/Fury X/980Ti level of GPU performance without a severe bottleneck. Yet, the rhetoric comparing RX 480 vs. GTX1070 misses this point by a country mile, due to either ignorance, or lack of research.

http://forums.anandtech.com/showpost.php?p=38321281&postcount=15

What are the chances an i7 2600K @ 4.5Ghz<->6800K user will be cross-shopping an $200 RX 480 4GB and a $400 GTX1070? Pure NV viral marketing/brand bias at play/shilling. Yet, I already see on this forum people wanting to buy a GTX1070 with an anemic CPU such as an i5 2400 -- a CPU that's even slower than the i7 960 4Ghz!

 

[sze5003]

:thumbsup:

I think $199 RX 480 4GB is actually the one I am going to recommend for single GPU use in the mainstream category (until GTX1060 6GB shows up). The 8GB model seems like a questionable way to spend $40 that can be used towards a future GPU upgrade instead. I would also easily recommend for a new PC builder to invest that $40 towards a better PSU or step up from an i5 6500 to i5 6600K or get him/her closer to i7 6700/K from i5 6600K. The investment into a faster CPU or a better PSU is $ well spent in this case imho.

The RX 480 doesn't seem powerful enough to truly benefit from > 4GB of VRAM as its performance falls in the same tier as 390/390X/970/980 level cards. In 2 years, that $40 can be used to step up from a $200-210 to a $240-250 card. I think it'll matter more in that case than the 4GB vs. 8GB today. GTX1060 3GB seems DOA as I won't recommend a 3GB card in 2016 in the $200 price range.

I was curious why they chose to use 3gb in the 1060. I also assume it won't be cheaper than a 8gb 480. I want to see the aib models for the 480 and I believe I'll go with an 8gb model. I would think more memory in some games would be useful if you want to turn settings up.

If the AIB 480's fall in the $300+ range I may just spend a bit more and go with the $399 gigabyte 1070.

 

[Magic Carpet]

I was curious why they chose to use 3gb in the 1060.

I suspect, they don't want to have another 3+1 drama

 

[zinfamous]

1: Kernel bugs are not always cut and dry, easy to find. Typically they are hard to find and test for. I know, because its part of my job.

2: nVidia has their own launch bugs with Pascal, so its not like AMD is alone in having a bug.

no way man: ATI/AMD is the only company to ever release buggy drivers, ever! /s

That being said, I would assume that the review cards have been in hands of reviewers for weeks now, correct? The hype train was already leaking various benchmarks and tests with various clocks and a few driver changes.

I would think that the earlier cards reviewers got, running on a few different drivers, may not be completely reflective of the cards released to consumers in the first couple of batches. Or not, who knows. This is why I never buy such things on release because with enough years under your belt buying electronics you eventually learn that it's a fool's game.

Sucks that it may be a "lottery pick" of GPUs according to one reviewer (implying a process issue at the foundry, no?) but if it's mostly a driver/random board issue that sounds at least more fixable. No one knows what kind of chips Apple and other partners received yet, so it's hard to say, imo, where the real problem is here at the moment.

Either way, if AMD released a pile of bad chips to early adopters and this PCIE power issue ends up frying some boards, that is very bad news for them. It doesn't really matter what is causing it for those users, or for AMD's reputation.

 

[Stuka87]

I was curious why they chose to use 3gb in the 1060. I also assume it won't be cheaper than a 8gb 480. I want to see the aib models for the 480 and I believe I'll go with an 8gb model. I would think more memory in some games would be useful if you want to turn settings up.

If the AIB 480's fall in the $300+ range I may just spend a bit more and go with the $399 gigabyte 1070.

Because of the bus width. The number of memory chips has to be matched to the bus width. If it has 192bit memory bus, it has to have 1.5, 3, or 6GB of RAM.

 

[beginner99]

Why wouldn't AMD do the same under-volt and performance increase as stock settings?

Different chips with different qualities. bad ones requiring more voltage and hence one that could run with less still gets the same BIOS with higher voltage setting.

 

[Faulkner]

I've been following all of the announcements and the overall hype about the card (480 that is), and am overall pleased with the outcome. If we exclude the reference design though .

A bit disappointing with the launch prices here in Australia. Kotaku state that it is as low as ~$350 for XFX model but in reality they state:
Stock of the XFX RX 480 4GB model isn’t actually available — it’s said sold out ever since the listing went online — but there is plenty of availability for the other brands.

Actual prices are ~$450 - http://www.shoppingexpress.com.au/buy/asus-rx480-8g-amd-radeon-graphics-card-8gb/RX480-8G

Cant wait for this initial hype to flatten and for AIB partners to do their things regarding the cooling and possibly adding 1x8 pin connector instead of the 1x6.

 

[Glo.]

After the number 9, which indicates that RX 490 will be, there is: >256 bit/4K.

So RX 490 may have wider memory bus, than 256 bit.

 

[HiroThreading]

LOL! Coming from a guy who bought a GTX980 for $550-600 less than 2 years ago. Today, AMD's $199-229 card is within 10-15% of that overpriced garbage 980! Generation after generation these relative concepts seem completely foreign to you.

*snip*

This has got to be one of the greatest posts of all time. Backed up by facts and data and no ad hominem attacks whatsoever.

Back to Polaris, I am now convinced that GloFo's 14nm FF process is responsible for its poor perf/watt performance -- especially if RX 470 or Polaris 11 manage to have significantly better perf/w.

 

[antihelten]

Damn, I might be disappointed on the power draw. But, I'm getting excited on grabbing an AIB with better components, cooling and power delivery. I want to see how far I can push this thing. A 1500mhz/9000mhz RX480 @ ~$250 is still pretty appealing regardless of the power draw.

There's also the fact that the reference version is power throttling (as demonstrated by computerbase.de), AIB version that fix this (by using an 8 pin or 6+8 pin design) should get another 5% performance (this is what computerbase.de got when they undervolted the card).

So take this combined with the PCIe bandwidth fix and we are suddenly looking at 390X performance instead of 390 performance. Factory overclocking (of the core, but perhaps more importantly also of the memory as demonstrated above) could then take it up to roughly R9 Fury level. If we can get that for around $250 then it wouldn't be too shabby.

Fingers crossed.

 

[Michael]

Is there any actual indication of when the AIB boards will be available? The reference card may be cheap enough, but I like to run with three monitors. You basically have to get at least 2 DP monitors and one HDMI or a few active adapters to DVI to reuse older monitors and that adds $15 or so or you pay extra for DP. It is hard to find 23" DP monitors in the same price range as others.

 

[LTC8K6]

Different chips with different qualities. bad ones requiring more voltage and hence one that could run with less still gets the same BIOS with higher voltage setting.

Yes, you sell them as different cards...high performance models...and regular performance models...different BIOS.

Marketing.