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

[PeckingOrder]

Its a reference card which always suck. They are not designed to be overclocked at all. If something breaks, blame yourself.

I'm sure the informed consumer who'll put this card in an OEM system with a shoddy mobo with blame himself

just blame the consumer for faulty products, I like that

For any budget gamers - I found review showing scaling RX 480 with different processors compared to GTX 970. Not many people targeting $200 GPU have i7k so this type of results could be very interesting
This is only quick test, but this website promised much bigger test soon.
http://www.purepc.pl/karty_graficzne/premiera_amd_radeon_rx_480_polaris_test_karty_graficznej?page=0,16

TL;DR - GTX 970 performs much better when paired with weaker CPUs.

yet another reason to get a 2 year old card

 

[OILFIELDTRASH]

I'll be receiving a 480 4gb soon. I'm wondering if I can overclock the 7ghz ram to around 8ghz like the 8gb version. Maybe the ram really isn't slower but they needed something to justify a $40 price difference? Even if it wouldn't pass 8ghz maybe 7.5ghz would work?

 

[railven]

For any budget gamers - I found review showing scaling RX 480 with different processors compared to GTX 970. Not many people targeting $200 GPU have i7k so this type of results could be very interesting
This is only quick test, but this website promised much bigger test soon.
http://www.purepc.pl/karty_graficzne/premiera_amd_radeon_rx_480_polaris_test_karty_graficznej?page=0,16

Well that isn't reassuring for someone who will be a budget gamer.


Overall not a bad card. Definitely not what I expected but I admit last minute the hype got me. Even debated buying one going on the OC claims. I mean, for $250 if it got 80-90% of my GTX 980 Ti, I'd give it a shot.

Oh well. Now the wait for Vega. Something tells me this isn't going to be the runaway hit AMD was expecting.

 

[railven]

Its a reference card which always suck. They are not designed to be overclocked at all. If something breaks, blame yourself.

This is interesting.

My reference HD 7970, hit 1125 on stock volts. Actually hit higher clocks at less volts than my Sapphire Dual-X I bought a few months later for CFX.
My reference GTX 980 Ti, hit 1500 with 106% Power setting, fan had to be upped to 100%, but that was remedied with a different cooler.

 

[Elixer]

I have yet to see any review where they test the h.265 encoder.
Has anyone seen any review that actually does something besides games?

 

[boozzer]

This is interesting.

My reference HD 7970, hit 1125 on stock volts. Actually hit higher clocks at less volts than my Sapphire Dual-X I bought a few months later for CFX.
My reference GTX 980 Ti, hit 1500 with 106% Power setting, fan had to be upped to 100%, but that was remedied with a different cooler.

they were both top of the line, best of their time. while 480 is basically 200$ mainstream. and there is no way a 480 can get near a 980 ti, unless you are talking about a AIB 480 vs a reference 980 ti.

you get what you pay for right?

 

[railven]

they were both top of the line, best of their time. while 480 is basically 200$ mainstream. and there is no way a 480 can get near a 980 ti, unless you are talking about a AIB 480 vs a reference 980 ti.

you get what you pay for right?

Okay...

HD 7850, same price, OC'ed just as good as HD 7970.

Just because this reference model appears to be gimped, doesn't mean a blanket statement that I responded to is supported. AMD clearly cheaped out on this card. Considering their prior reference models were beasts.

I recall before this card even coming out the LOL comments at how many power phases it has reference versus GTX 1080. AMD cut corners, and I can understand why. Claiming because it's a mainstream part is bunk.

 

[krumme]

INBF the ugly duckling will be the 460 which some custom AIBs will make them look like a tiger.

And what do you expect guys? That 480 stomps the 1070?
It gives battle to the 970 with less resources. Also take in care something: the 1060 will have same difficulties to enter on that market... kill me guys, but nVIDIA and AMD screwed up BIG time: both will leave a BIG performance hole between 1060/480 and 1070/490...

Agree. There will be a hole. But only for amd. Nv will price the 1060 at their own will and sell it on brand power and marketing muscle.
Amd will sell very low numbers of vega vecause even if its faster than 1080 it will be far slower than the nv big die. Expensive to produce vs 1080. Stuck in a bad place no matter what.

 

[DownTheSky]

Seems RX 480 is heavily memory bandwidth limited.

Just finished with driver installs. No overclocking, latest driver revisions shows ~24.7MH/s each.

Edit 1: GPU way outpaces the memory on these things. 950mhz GPU clock did not change the performance of the hashing at all.

Can somebody try this with games?

 

[boozzer]

Okay...

HD 7850, same price, OC'ed just as good as HD 7970.

Just because this reference model appears to be gimped, doesn't mean a blanket statement that I responded to is supported. AMD clearly cheaped out on this card. Considering their prior reference models were beasts.

I recall before this card even coming out the LOL comments at how many power phases it has reference versus GTX 1080. AMD cut corners, and I can understand why. Claiming because it's a mainstream part is bunk.

it is a main stream card, they cut corners for more profit. so what. get AIB if you want to oc

those were just super hype, I hope not alot of us bought into it? how good can a 200$ gpu get? as long as reference beats a 970, it is a win remember the stupid hype from last year? how many of us buys into the hype for anything these days?

edit: forgot to ask, how noisy were those cards after oc?

 

[krumme]

OCUK probably sold over 1000 units today. Pending verification of final numbers on Friday by Gibbo (he's traveling right now):
https://forums.overclockers.co.uk/showpost.php?p=29716162&postcount=107

May be able to hear more info about AIB custom cards by then also.

So selling the same numbers of 480 in one day as the entire time of 1080.
Thats what mainstream means. Lol.
Interesting what 470 can sell. I guess the 150 to 250 is the sweetspot by far but we will see.
Perhaps its crazy miners undervolting what not. Its really a lot of cards...

 

[Bacon1]

Okay...

HD 7850, same price, OC'ed just as good as HD 7970.

Man, remember when reviewers would actually get > 60 fps in game settings when testing.. good days back then.

I mean look at that, 100+ FPS in a shooter.

Now we have most reviews testing ridiculous "MAX" settings where you can't even hit 60, hell half of the time its under 45 fps like these tests from here:

http://www.purepc.pl/karty_graficzn...x_480_polaris_test_karty_graficznej?page=0,16

What happened to testing playable settings and not these extremes that no one would enjoy playing at?

What happened to 60 fps being in the minimum for action (FPS, Racing, etc) games? 30-45 FPS is fine for an RTS, but not a FPS.

 

[Armsdealer]

After seeing the RX 480 results, I very unhappily pulled the trigger on a $400 GTX 1070. I absolutely hate getting ripped off, but I need the GPU for VR, and when using the GTX 970, I concluded it was not really sufficient.

That said, if I didn't need the product for VR, I would've had my eye on jet.com and trying to get another $25 off an RX 470, which frankly seems to be the star given how little the RX480 overclocks. Or at minimum $30 off the RX480. IT seems like jet.com will carry them given the listings have shown up and I would be shocked if they try to charge more than $200/$240 or $250 before $30 rebate.

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.

 

[EightySix Four]

I'm sure the informed consumer who'll put this card in an OEM system with a shoddy mobo with blame himself

We're being a bit ridiculous here since we don't have all of the facts yet. The reviewers received custom cards with switches on them that changed the amount and speed of the VRAM as well, so this could have been an issue with just those.

I have three in a mining box and I think I'd be experiencing some problems if they were pulling that kind of power.

 

[maddie]

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?

 

[krumme]

I'll be receiving a 480 4gb soon. I'm wondering if I can overclock the 7ghz ram to around 8ghz like the 8gb version. Maybe the ram really isn't slower but they needed something to justify a $40 price difference? Even if it wouldn't pass 8ghz maybe 7.5ghz would work?

In my experience probably. Let us know !

 

[EightySix Four]

Seems RX 480 is heavily memory bandwidth limited.


Can somebody try this with games?

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.

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?

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.

 

[IEC]

Seems RX 480 is heavily memory bandwidth limited.

Can somebody try this with games?

In the computerbase.de review I linked in the OP, they tested with a memory overclock of 12.5%.

Result was an average 6% performance increase:
https://www.computerbase.de/2016-06/radeon-rx-480-test/12/

So it is not linear scaling, but it does help and leaves some room for AIBs to improve on reference performance numbers.

 

[railven]

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?

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.

 

[railven]

Man, remember when reviewers would actually get > 60 fps in game settings when testing.. good days back then.

I mean look at that, 100+ FPS in a shooter.

Now we have most reviews testing ridiculous "MAX" settings where you can't even hit 60, hell half of the time its under 45 fps like these tests from here:

http://www.purepc.pl/karty_graficzn...x_480_polaris_test_karty_graficznej?page=0,16

What happened to testing playable settings and not these extremes that no one would enjoy playing at?

What happened to 60 fps being in the minimum for action (FPS, Racing, etc) games? 30-45 FPS is fine for an RTS, but not a FPS.

No idea what you're talking about. For years the reviews I've read where always settings that were considered Ultra/Max. Unless you read stuff like HardOCP which I know aimed for highest playable options (I don't read HardOCP and have zero intentions to.)

I usually use ATF and TUP for my go to sties. The rest are just fillers.

 

[IEC]

I have yet to see any review where they test the h.265 encoder.
Has anyone seen any review that actually does something besides games?

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.

 

[railven]

it is a main stream card, they cut corners for more profit. so what. get AIB if you want to oc

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.

those were just super hype, I hope not alot of us bought into it? how good can a 200$ gpu get? as long as reference beats a 970, it is a win remember the stupid hype from last year? how many of us buys into the hype for anything these days?

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.

edit: forgot to ask, how noisy were those cards after oc?

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.

 

[EightySix Four]

I have yet to see any review where they test the h.265 encoder.
Has anyone seen any review that actually does something besides games?

What do you want to know?

http://imgur.com/mjZKhFg

 

[Silverforce11]

The 1 nvidia sponsored dx12 title saw performance regress for everyone, any neutral dx12 games out there?

Forza & Quantum Break are neutral. It's actually in Quantum Break that GCN runs far ahead, punching above it's weight. Remember 390 = 980Ti?

So anyone who says that AMD is only faster in DX12 when they sponsor the game is automatically LYING.

 

[Bacon1]

Well that isn't reassuring for someone who will be a budget gamer.


Overall not a bad card. Definitely not what I expected but I admit last minute the hype got me. Even debated buying one going on the OC claims. I mean, for $250 if it got 80-90% of my GTX 980 Ti, I'd give it a shot.

Oh well. Now the wait for Vega. Something tells me this isn't going to be the runaway hit AMD was expecting.

https://www.computerbase.de/2016-06...diagramm-xcom-2-1920-1080-intel-core-i5-2500k

Seems like the old i5 2500k can handle it fine in most games.