AMD RX 400 series (Polaris) MegaThread! FAQ & Resources

[Robert Cook]

Hello fellow Hardware Enthusiasts!

It is currently June 30th, AMD has just released the RX480 which is selling well. As for the RX470 and RX460, they seem to be simply put, "coming soon". After Nvidia's 16nm Pascal architecture rolled around a month ago, AMD was left with some big shoes to fill, and this thread will hopefully cover their efforts to fill or perhaps even surpass expectations using their 14nm Polaris architecture. Naturally as tech news goes, the rumor mills are running full time, and thus this thread is likely to change if better/ newer reasonably believable information is available.

The current lineup of Polaris cards is divided into Polaris 10 and 11. Like Nvidia, this seems to reflect the lower number = bigger chip idea, thus, Polaris 10 is the higher performance card, while polaris 11 will be for more mid range.Currently AMD has announced 3 cards:
The Rx 480 470 and 460 in lessening price and performance respectively.

So without Further ado...

******************************************************
Polaris 10
******************************************************

Rx480 MSRP:$199 4GB/$239 8GB

The core:
2304 Stream Processors (R9 290X/390X has 2816)
32 ROPs ( Render OutPut Units)
144 TMUs (Texture Mapping Units)
1266 MHz base clock

RAM: 4GB or 8 GB of GDDR5 RAM on a 256 bit memory bus. The 4GB card uses Memory clocked at 7 GHz effective which translates to around a 224GB/s bandwidth, while the 8 GB card ships with an 8GHz effective speed. This should mean a memory bandwidth of 256GB/s

Compute performance: 5.8 Trillion Floating Point Operations Per Second Peak ( R9 290X is 5.6)

Power: 150 W TDP
This power comes from the up to 75W that both the PCIE Slot ( and thus motherboard) and six pin PCIE power cable are rated to supply.
This means that the PCIE slot will have to supply power in order to fully power the card, however, there are current issues that would imply the card is taking excessive power from the PCIE slot as seen in this forum's Rx 480 review

Comparisons are made to the R9 290X, as this is AMD's current non Fiji flag ship card, and now sells for around $290 meaning that being able to beat out a R9 290X or at least tie it will mean that Polaris is indeed a successor in performance and power efficiency. Release day results show the 8GB card performing in between a R9 390 and R9 390X, so it stand to reason, that the 4GB RX480 card is under R9 290X performance which is slightly below R9 390X performance.

For a more complete review of the RX 480 here are a bunch of sites to go binge on...
http://www.tomshardware.com/reviews/amd-radeon-rx-480-polaris-10,4616.html
http://www.anandtech.com/show/10446/the-amd-radeon-rx-480-preview
http://arstechnica.com/gadgets/2016/06/amd-rx-480-polaris-review/
https://www.google.com/webhp?sourceid=chrome-instant&ion=1&espv=2&ie=UTF-8#q=RX+480&tbm=nws

For those who want video:
https://www.youtube.com/watch?v=2GdfDCq86Gk
https://www.youtube.com/watch?v=vazL57c4nq8
https://www.youtube.com/watch?v=lRDgiWBfs5I
https://www.youtube.com/watch?v=-y6Ky_pCawQ
https://www.youtube.com/results?search_query=RX+480

If you find more good sites, please tell me.

**************************************************************
Polaris 10 Pro ( cut down Polaris 10)
**************************************************************

RX470MSRP:$149 4GB/ $179 8GB

The Core:
2048 Stream Processors
32 ROPs
128 TMUs
1206MHz Base clock

RAM: 4GB or 8 GB of GDDR5 RAM, with the same 256 bit memory bus running at 7 GHz effective, which should result in 224GB/s memory bandwidth

Compute Performance: ~4.95 Trillion Floating Point Operations Per Second Peak ( R9 390 is 5.1)

Power: Once again a single six pin is to be used and this time the TDP should be around 110W
As this card seems to be sporting specs similar to the situation with the 970 and 980, it seems the RX 470 should fall between the R9 380X and the R9 390

**************************************
Polaris 11
**************************************

RX460MSRP:$99 2GB/$129 4GB

The Core:
1024 Stream Processors
16 ROPS
64 TMUs
Perhaps a 1400MHz core clock

RAM: 2GB or 4GB of GDDR5 running on a 128 bit memory bus running at 7GHz effective for a 112GB/s bandwidth


RX 480 power debate

Although the RX 480 had a favorable launch, it was soon noticed, that the card did not really follow its TDP recommendations and overdrew from the PCIE slot. For more on that please see here:
http://www.tomshardware.com/reviews/amd-radeon-rx-480-polaris-10,4616-9.html
https://www.youtube.com/watch?v=ZjAlrGzHAkI
While this was not going to set motherboards on fire, it certainly was a setback, so AMD had to make some fixes, as of today, here they are:
http://www.tomshardware.com/reviews/amd-radeon-rx-480-polaris-power-fix,4668.html
The power re routing is most likely possible over a simple driver update, as the board has a very complex power delivery system.

This video here looks at the PCB design
https://www.youtube.com/watch?v=qG2e-v94L4M

While the reference card now draws around 10W less from the PCIE slot, making current owners more pleased, it will be interesting to see what AIB cards can do, as it seems power delivery on the board is a real limit to overclocks.
More updates to follow!

 

[cdrkf]

So I did the driver upgrade here is my results:

Idle power useage went down from 180w to 136w, thats a huge improvement IMO.

Ran Firestrike, gained 70 points over my prior best score (which was done when the card was 4gb), and 90 points over my best score with the card set to 8gb. While my power meter on my UPS doesn't do averages it was clearly biased towards the lower range running around 290 to 340w most of the time with some spikes to 370 and up as high as 408. So total power useage is the same on a spike but generally it seemed to run lower.

Current score: http://www.3dmark.com/fs/9199250
Prior 8gb best score: http://www.3dmark.com/fs/9157036

Temp went from 82 to 81 C.

Seems like its working.

Really small performance penalty, that's probably less than 1fps difference in the real world.

I'm amazed that a graphics driver was able to fix this problem. I would of guessed the only way to fix this is thru a BIOS update.

You're getting in backwards- he's saying he *gained* between 70 and 90 points in FireStrike- his card is running FASTER after the update

Edit: Woo, I hereby officially launch this new thread, may all who comment here gain knowledge and wisdom through it's many pages

 

[AndrewJacksonZA]

You're getting in backwards- he's saying he *gained* between 70 and 90 points in FireStrike- his card is running FASTER after the update

I wonder if that's due to less thermal throttling or just better drivers. What are your thoughts on the matter?

A bit of a technical question: Since we're speaking about drivers affecting power consumption (and possibly reprogramming the on-board power management chip?) does anyone know how, since partner boards may or may not contain the same power management chip, AMD caters for each different AIB's power management chips? The power management chips all expose a common API, perhaps?

 

[Rogue Leader]

So I did the driver upgrade here is my results:

Idle power useage went down from 180w to 136w, thats a huge improvement IMO.

Ran Firestrike, gained 70 points over my prior best score (which was done when the card was 4gb), and 90 points over my best score with the card set to 8gb. While my power meter on my UPS doesn't do averages it was clearly biased towards the lower range running around 290 to 340w most of the time with some spikes to 370 and up as high as 408. So total power useage is the same on a spike but generally it seemed to run lower.

Current score: http://www.3dmark.com/fs/9199250
Prior 8gb best score: http://www.3dmark.com/fs/9157036

Temp went from 82 to 81 C.

Seems like its working.

Really small performance penalty, that's probably less than 1fps difference in the real world.

I'm amazed that a graphics driver was able to fix this problem. I would of guessed the only way to fix this is thru a BIOS update.

You're getting in backwards- he's saying he *gained* between 70 and 90 points in FireStrike- his card is running FASTER after the update

Edit: Woo, I hereby officially launch this new thread, may all who comment here gain knowledge and wisdom through it's many pages

Yeah exactly I think he got it backwards, it works better! Didn't get a chance to game on it. TBH real world 70 points is only 1 or 2 fps, the power issues were really the fix here. I'm thinking next I may try the power conservative mode and see if that changes anything.

Incidentally I hate how much power my PC uses at idle, my wife's i7-6700 (admittedly has an m2., only 1 hd (I have 2), only 2 fans (I have 4), and a GTX 950) sits right around 50w at idle.

Hoping AMD's next processor generation is powerful while being a little more power friendly. I mean power isn't expensive here, but it just seems like a waste, not that I even leave it on unless I'm using it anyway most of the time. However my whole Xeon E3 1231v3 based server folding all day long at 50% uses the same wattage.

You're getting in backwards- he's saying he *gained* between 70 and 90 points in FireStrike- his card is running FASTER after the update

I wonder if that's due to less thermal throttling or just better drivers. What are your thoughts on the matter?

A bit of a technical question: Since we're speaking about drivers affecting power consumption (and possibly reprogramming the on-board power management chip?) does anyone know how, since partner boards may or may not contain the same power management chip, AMD caters for each different AIB's power management chips? The power management chips all expose a common API, perhaps?

I don't think it was thermal throttling, at 1deg C difference the temps are effectively the same. Probably just some optimizations on voltages and clocks at specific states.

 

[17seconds]

Need to test the overclocking on the new driver.

 

[Math Geek]

here's an interesting little snippet speculating on what a 490 card could be http://wccftech.com/amd-rx-490-dual-gpu/

makes a lot of sense that a 490 could be a dual gpu card considering we know that the current 480 gpu is the full spec chip and there is not a bigger/better model available. either way, it's a fun little read

 

[simon12]

A dual 480 I would imagine could compete to hard with Vega for them to release while a dual 470 would be lower power and could maybe be priced under a 1070, either would be interesting. Or it could be Vega is an absolute monster priced well above the 1080 and a dual 480 makes sense against the 1080 though this 490 would leave a big gap between it and the 480. Does anyone have any idea what HBM2 would add to the price and performance of a 1080 (assuming they could get it now). Is HBM2 likely to come to mid range cards soon after Vega or is that years away?

 

[Rogue Leader]

The problem of a dual 480 is of course the power consumption (~300w+), and on top of that dealing with all the issues of Crossfire (which now that I'm done with it, I say again, sucks). IMO its a waste of resources, put them into getting Vega out sooner.

 

[Martell1977]

A dual 480 I would imagine could compete to hard with Vega for them to release while a dual 470 would be lower power and could maybe be priced under a 1070, either would be interesting. Or it could be Vega is an absolute monster priced well above the 1080 and a dual 480 makes sense against the 1080 though this 490 would leave a big gap between it and the 480. Does anyone have any idea what HBM2 would add to the price and performance of a 1080 (assuming they could get it now). Is HBM2 likely to come to mid range cards soon after Vega or is that years away?

I doubt nVidia will redesign the 1080 to accommodate HBM2, it would mean a huge undertaking. Same with AMD cards, it would just be impractical to change he entire design of the memory system. Now, future cards will likely have it, i.e. 1080ti, titan successor and possibly the 490, Fury successor.

Until production costs lower and GPU's in the mid range being powerful enough to benefit from it, there is no reason or need for HBM2 below the top end. Yes, the Fury cards were powerful, but the HBM didn't give it much of an advantage. I doubt even the 1080 would get much of a boost from it at this point.

 

[Martell1977]

The problem of a dual 480 is of course the power consumption (~300w+), and on top of that dealing with all the issues of Crossfire (which now that I'm done with it, I say again, sucks). IMO its a waste of resources, put them into getting Vega out sooner.

Yeah, seems a bit premature to have a dual GPU card. The 490 and Fury replacement hasn't even come out yet. Besides, if they did release a dual 480, I'd expect it to be named something along the lines of RX 480x2, since having the 490 be a dual GPU card then Vega as single GPU's would be confusing to say the least.

 

[simon12]

I know nvidia will not rerelease the 1080 on HBM2 just wanted to speculate what difference it wold make. Anyway the most comprehensive 480 crossfire test I can find is here https://www.techpowerup.com/reviews/AMD/RX_480_CrossFire/
If you don't want to read it all this is the best summary

 

[17seconds]

AMD will have nothing new in the upper range until 2017. Furthermore, their current upper range cards were just obsoleted by the GTX 1060. It's not too outlandish that they would come out with a dual 480 card to fill that space as a stopgap measure. It wouldn't be a very compelling card, but conceding the upper range for the better part of a year is not a good strategy.

 

[BurgerandChips66]

Maybe a dual water-cooled RX 480 with 16GB vram and lots of overclocking potential. But they'll need to make sure those crossfire drivers are up to date to tempt me.

 

[Math Geek]

AMD will have nothing new in the upper range until 2017. Furthermore, their current upper range cards were just obsoleted by the GTX 1060. It's not too outlandish that they would come out with a dual 480 card to fill that space as a stopgap measure. It wouldn't be a very compelling card, but conceding the upper range for the better part of a year is not a good strategy.

kind of what i was thinking. since vega not supposed to be out until next year, a dual 480 would be an interesting stop gap. even if it is called a 490 or 480x2. since amd has already stated this is the full polaris gpu and not a cut down version. if they are to do anything above a 480 before vega it would almost have to be a dual gpu card, unless they have some more optimization they can do to the fury gpu to get one last release out of it.

 

[simon12]

I think your jumping the gun saying " their current upper range cards were just obsoleted by the GTX 1060" when we don't know the performance for sure or if there will be enough on the shelves for months, though it will likely happen sooner rather than later. I would say their current upper range cards were just obsoleted by the RX480 would be more accurate based on price performance.

 

[17seconds]

Problem is, it would be twice the power consumption of the Pascal cards In it's performance range. Also it would only work about 75% of the time, and will introduce frame time stuttering another 25% of the time when it does work, not to mention the noise/heat factor.

 

[Rogue Leader]

Problem is, it would be twice the power consumption of the Pascal cards In it's performance range. Also it would only work about 75% of the time, and will introduce frame time stuttering another 25% of the time when it does work, not to mention the noise/heat factor.

100% agree, using it as a stopgap would be a step backwards. Relying on Crossfire (or SLI for that matter) to make up or a performance deficiency would only bring out a card that in benchmarks performs as well as competitors, but in real life does not nearly. In addition the heat, power needs, and compatibility are all issues.

My RX 480's best 3dmark score is 800 points lower than my overclocked Crossfired R9 280's best score. And yet in terms of (my opinion of course) "rated" FPS etc its lower, it performs FAR better, everything is smoother. Heck watching the 3Dmark tests I was sure based on how it performed the score would be the same or higher and I was extremely surprised it wasn't.

 

[salgado18]

Problem is, it would be twice the power consumption of the Pascal cards In it's performance range. Also it would only work about 75% of the time, and will introduce frame time stuttering another 25% of the time when it does work, not to mention the noise/heat factor.

100% agree, using it as a stopgap would be a step backwards. Relying on Crossfire (or SLI for that matter) to make up or a performance deficiency would only bring out a card that in benchmarks performs as well as competitors, but in real life does not nearly. In addition the heat, power needs, and compatibility are all issues.

My RX 480's best 3dmark score is 800 points lower than my overclocked Crossfired R9 280's best score. And yet in terms of (my opinion of course) "rated" FPS etc its lower, it performs FAR better, everything is smoother. Heck watching the 3Dmark tests I was sure based on how it performed the score would be the same or higher and I was extremely surprised it wasn't.

Agree with all of this, but that didn't stop them from creating the 295x2, right?

I believe they are releasing Polaris as the first product out of the new node, with an upgraded-but-same architecture (GCN). Then, with a more mature process, and probably a redesigned architecture, they can improve power efficiency enough to create a card with double (or more) shaders without being an industrial oven. After all, the 1080 does 180% of the performance of the 480, but only ~130% more power. They can't release a top card to compete that uses +250W again.

 

[Math Geek]

the custom 1080 cards regularly use 250w+ depending on the model. some spike easily past 300w!! with 225w being pretty normal for the low end custom cards. so not like a 250w 480x2 would be WAY out there beyond the 1080 power usage. it's not like anyone is going to want to test it against an FE card where it might actually be pretty comparable

 

[Sakkura]

AMD will have nothing new in the upper range until 2017. Furthermore, their current upper range cards were just obsoleted by the GTX 1060. It's not too outlandish that they would come out with a dual 480 card to fill that space as a stopgap measure. It wouldn't be a very compelling card, but conceding the upper range for the better part of a year is not a good strategy.

Nvidia claims the GTX 1060 is 15% faster than the RX 480, and the price is 25% higher. How is that making the RX 480 obsolete?

 

[Martell1977]

AMD will have nothing new in the upper range until 2017. Furthermore, their current upper range cards were just obsoleted by the GTX 1060. It's not too outlandish that they would come out with a dual 480 card to fill that space as a stopgap measure. It wouldn't be a very compelling card, but conceding the upper range for the better part of a year is not a good strategy.

Nvidia claims the GTX 1060 is 15% faster than the RX 480, and the price is 25% higher. How is that making the RX 480 obsolete?

They said "upper range", The 480 is a mid-range card. The 480 all but killed the market for the 390 and 970, the 1060 (assuming nVidia didn't fudge the numbers) is going to do the same to the 390X, Fury and 980.

 

[17seconds]

^Exactly.

Based on the claims that the GTX 1060 is "faster than the GTX 980", that puts it in the range of the Fury/FuryX. Those cards cost $400-$500 compared to their new competitor in that performance range at $250-$300. And then there's the GTX 1070 also right there in that price range. There's not much reason to go AMD above the RX480 level, and even those are a tough sell.

 

[-Fran-]

Keyword being "claims".

Leave room for doubt, please.

Plus, that article from Semiaccurate left me cold... I mean, Corporations do cut each others throats, so I *can* imagine it being possible, but I don't want to believe it. Anyone with some insight or update?

Cheers!

 

[Ags1]

RX480 has more VRAM, more bandwidth and better DX12 than GTX1060. Because of the power fiasco and inadequate cooling in the RX480 reference we still don't know how well it will OC. I am getting the RX480 for VR and I'm wondering - does the high bandwidth of the RX give it an edge over the GTX when it come to the high frame rates needed in VR? Also with VR being a new area, isn't DX12 disproportionately important as there is no back catalog of DX11 games to think about.

 

[Math Geek]

what insight on if nvidia is trying to pull a very public joke on the world??? i can some that one up real easy. NOT A CHANCE IN HELL THE 1060 IS A JOKE AND NOT ACTUALLY BEING RELEASED.

hope that clears it up for you that and reviewers have already said they have cards in hand and most brands have started teasing cards. i very highly doubt this is some kind of global multi-brand joke.