Polaris Refresh (RX 500 Series) Rumors

[JDG1980]

Since the mods don't want Polaris refresh rumors discussed in the Vega/Navi thread, I decided to open a top-level thread for this discussion. I opened it in the main VC&G board because users may want to discuss comparisons with Nvidia cards and this isn't allowed in the AMD board.

So far, there has been no official announcement (as far as I know) of any RX 500 Series products. But there are various leaks and rumors (here is one of the more comprehensive). Some of what has been suggested so far:

• The original Polaris chips were on the GloFo 14LPE (Low Power Early) process; RX 500 will be on the superior GloFo 14LPP (Low Power Performance) process.
• To indicate the generational change, the chips will be renamed from Polaris 10 and Polaris 11 to Polaris 20 and Polaris 21. (Some drivers suggest Polaris 10XT2 instead.)
• There will be a new Polaris 12 chip, which will be a smaller GPU featuring only 640 shaders. (This one seems to be fairly solidly established.)
• RX 580 (full P10/P20) will have a relatively modest clock bump over RX 480, up from 1266 to 1340 MHz. The RAM will remain at 8Gbps. RX 570 (cut P10/P20) will increase clocks to 1244 MHz. RX 560 will use the full Polaris 11 (or Polaris 21?) chip, with 1024 shaders, and go up to 1287 MHz.
• Launch is said to be around April 18, to coordinate with Ryzen 5.

Thoughts? My biggest concern about the rumors so far is the retention of 8Gbps RAM on the RX 480. Some have suggested this might be a bottleneck. There is conflicting information about whether P10's memory controller supports GDDR5X, but even if it doesn't or that is cost-prohibitive, you'd think they could at least go up to 9Gbps like on the newer Nvidia cards.

The original consumer Polaris release was unquestionably a disappointment in terms of perf/watt. Polaris is capable of doing much better, and the Polaris 10-based Radeon Pro WX 5100 manages to crank out over 2.7x the TFlops of its Bonaire predecesor (FirePro W5100) in the same form factor and 75W TDP. But the original RX 480 cards were pushed pretty hard in terms of clocks to stay on par with Hawaii and GP106, resulting in perf/watt not much better than Nvidia's 28nm Maxwell despite the node jump. Some of this is architecture (tiled rendering is a huge deal, and Polaris lacks it), but a lot of it is what amounts to factory overclocking. The question is whether a Polaris refresh can do any better. Will RX 580 really be able to hold a steady 1340 MHz while staying under 150W? If so, that would be a fairly big jump in perf/watt. The original RX 480 cards often throttled due to TDP, and could not sustain the 1266 MHz boost clock during actual gaming scenarios.

 

[piesquared]

Maybe these RX500's are for mobile, and next generation is Vega top to bottom? That seems to be the trend of the last few years for both AMD and NV, moving the current gen into mobile but advance the name to keep everything consistent.

 

[Mopetar]

Not a terribly inspiring performance bump, but if the chips run at those clocks while drawing less power, I think it could be okay. I think the biggest problem with Polaris was that it was all over the place. For all of the 480's that couldn't OC worth a damn and ran hot, there were just as many that could get a reasonable OC while also dropping the voltage. AMD needed to be less aggressive with their bin and do a better job segmenting the cards that could barely manage to handle the stock settings (or occasionally even failed to do that without throttling) and those that were capable of a substantial OC.

Polaris, the 14 nm process, or some combination of the two weren't meant for high clock speeds. It's a rather capable part in terms of efficiency, but not when pushed to the higher clock speeds it needed to have competitive performance to the 1060. AMD probably would have been better off adding more shaders and lowering the clocks to create a better balance and keeping a separate bin of chips that could OC well to have some third party cards capable of commanding $300.

The only saving grace for Polaris is that it's been obtainable for a reasonable price and sometimes has been an outright steal of a deal. Even if the refresh doesn't bring considerable performance improvements, if it brings a big efficiency improvement it has a place, but I don't think it will be fondly remembered. If Vega turns out to be good, Polaris will probably just feel even more disappointing.

 

[VirtualLarry]

Yeah, I guess I just don't see the excitement. Maybe RX 560, with a full complement of 1024 shaders, and reasonable clocks, and low power consumption, could be a winner, and an improvement over the cut-down RX 460 (896 shaders).

But I don't see what I view as a minor clock bump, to the RX 470 and RX 480, to be worth much extra. Especially if 3rd-party AIB cards are already overclocked to nearly that level. And the "stuck at 8Gbit/sec" GDDR5 is also worry-some. If they are bumping the core clocks, but not the memory clocks, what is the point? Or were existing RX 470 and 480 cards not memory bandwidth constrained? (No performance improvements from overclocking RAM on RX 470/480?)

And the cut-down even more Polaris 12 GPU, with 640 shaders? If the RX 560 can already operate without a 6-pin power connector, then what's the point of cutting the design down even more? The only answer that I can see, is that P12 is intended for mobile applications (laptops, mainly), and not desktop cards. (Unless, they come out with a passive, single-slot, low-profile, card, with HDMI2.0 and maybe mini-DP on it. That could find a niche among upgraders, with SFF Core2/2nd-gen/3-gen Core SFF rigs.)

 

[ConsoleLover]

I guess they can up the clocks a bit and retain the same power consumption and using 6pin. Though I suspect the 500 series may only be AIB cards, so no reference AMD cards, therefore most 580's will have 8pin connectors, so it doesn't even matter if they are able to push 1340mhz and keep 150W.

Thing is even later 480 can retain up to 150W at 1340mhz speeds, but going more significantly increases consumption, at 1400MHz consumption goes up to 190W.

The biggest issue with the reference 480 was that AMD was using 6pin connector and so the boost clocks suffered a lot, because it would often hit 150W at 1270MHz, when AIB partners put an 8pin connector it would automatically have more room for performance. Essentially AMD were obsessed with "looking good" and "only" consuming 150W so they included a 6pin connector, even though an 8pin connector would have served them better.

At release the 480 was 15%-25% slower than the 1060, now its 1%-5% slower in newer games, but the biggest reason is AIB partners offering 8pins and better cooling, as well as better boards that has sustain better voltage, therefore with AIB OC's the 480 was able to close the gap and now with DX12 even get the lead in few games.

 

[Ancalagon44]

Here is some interesting information. Do a google search for RX480 undervolting, and you will see Reddit threads shortly after the release of RX480, in which people noticed that their cards were running at 1.3V under load. A lot of them managed to get their voltage down to 1.05V.

I checked my RX480, and by default, it uses 1.138V. So, down from 1.3V as the default, to 1.138V as the default. This is on automatic settings. I know every card is different, but this means either one of two things:

The manufacturing has improved such that later RX480s can get by on a lower voltage and thus use less power, OR, AMD's driver optimization has allowed them to use tighter tolerances for voltage which has allowed the voltage to reduce, thus saving power.

I'm sure if someone were to re-review RX480 now, they would find it uses a lot less power than it used, and hence has a better performance per watt figure.

Anyway, my point is that RX580 could contain these voltage optimizations and more. For the same power envelope, it might end up delivering a lot more performance.

 

[mohit9206]

The RX 560 better be atleast 25-30% faster than rx 460 for $99.

 

[SpaceBeer]

With ~15% more CUs plus ~10% clock increase it can hardly be 30% faster. But even 20% improvement for the same price ($120 for 4GB model) would be ok

 

[antihelten]

The RX 560 better be atleast 25-30% faster than rx 460 for $99.

It would have to be clocked at about 1310-1365 MHz to achieve that (assuming linear scaling), which doesn't seem totally outrageous given that the RX 580 will apparently be clocked at 1340 MHz (although I wouldn't be surprised if the RX 560 only ends up at somewhere around 1250-1300 MHz either).

The $100 price is more questionable imho (for the 4GB version at least). With the above performance boost, it should more or less be neck and neck with the 1050 Ti, which is currently going for about $140. Whilst I would expect AMD to undercut Nvidia by at least some amount, I think 30% is a bit much, and more realistically it will probably be priced at $120 with perhaps a 2GB version at $100.

 

[nathanddrews]

I don't mind a straight rebrand if it significantly lowers prices while offering the same performance of the original SKU, but it seems more typical that the rebrand launch price is the nearly the same as the original SKU launch price. If they sold straight rebrands for the same price as the average on-sale price of the original SKU, it wouldn't seem like such a scam.

A refresh (improved binning, improved performance, improved features) for the same price or lower is much better, however.

Ideally we'd get a whole new architecture all at once, but I understand why that doesn't always work. Even NVIDIA releases products in stages over an 12-18 month period for each generation.

 

[Headfoot]

I wonder if they're saving GDDR5X for a 5x5 refresh to the rebrand. E.g. in another 6-9 months put GDDR5X on a RX 580 and call it RX 585? I bet GDDR5X hasnt come down enough in price yet to make it to Polaris. Probably same story on the 9Gbps GDDR5

 

[Insert_Nickname]

There will be a new Polaris 12 chip, which will be a smaller GPU featuring only 640 shaders. (This one seems to be fairly solidly established.)

Yes please...!!! We'd finally get a bottom-of-the-barrel replacement for all the old Kepler-, VLIW5- and Oland-based entry level cards. With a modern display controller.

And the cut-down even more Polaris 12 GPU, with 640 shaders? If the RX 560 can already operate without a 6-pin power connector, then what's the point of cutting the design down even more? The only answer that I can see, is that P12 is intended for mobile applications (laptops, mainly), and not desktop cards. (Unless, they come out with a passive, single-slot, low-profile, card, with HDMI2.0 and maybe mini-DP on it. That could find a niche among upgraders, with SFF Core2/2nd-gen/3-gen Core SFF rigs.)

Actually, it'd be useful too for HEDT, Ryzen and people who just need to drive more monitors then their IGP allows. There are plenty of low-end boards which only offer a single VGA, DVI and HDMI port, and not necessarily all of them.

 

[w3rd]

Polaris re-spin is aimed at the $99 ~ $199 market. Faster, cooler & more efficient than current models.

Competing strongly for mainstream consumer dollars. They are not profit driven cards, but market saturation cards for 1080p - 1440p Consumers. As the Radeon RX580 will be better than the RX480, for the same price.


AMD is just using these cards to solidify their tech dominance over nVidia's, in dx12.

 

[VirtualLarry]

Polaris re-spin is aimed at the $99 ~ $199 market. Faster, cooler & more efficient than current models.

Competing strongly for mainstream consumer dollars. They are not profit driven cards, but market saturation cards for 1080p - 1440p Consumers. As the Radeon RX580 will be better than the RX480, for the same price.

AMD is just using these cards to solidify their tech dominance over nVidia's, in dx12.

Yeah, while AMD makes no profit, and thus starving their GPU R&D pipeline, all the while being the DX12 pioneer, and taking all of the arrows in the back for being so, just in time for NV to design and implement a family of "true DX12" cards, when the market is finally ripe for them, in a year or two, when coding a game for DX12 is a primary consideration, rather than an afterthought, or a tech demo. Trust me, when that happens, NV will gobble back up whatever market share AMD has pried from their hands, via their strategy of carpet-bombing the under-$200 video card price bracket with DX12-capable cards. While NV still has the performance crown.


Don't feed the trolls.

AT Moderator ElFenix

 

[w3rd]

No profit..?

Sounds like you were there when AMD signed the fab contract? And you know yields and rate of cut-downs?

And if you are going to play that game of "down the line nVidia will eventually catch" up to AMD (in dx12) on low-budget & mainstream cards... Then I agree.

But people running 1080p - 1440p won't be upgrading their cards in a few years, they will be upgrading their monitors by then, when 4k are $250 bucks.

Odd, that you are a mainstay here and you don't understand that Nvidia will not have the performance crown long.

Pascal is EOL bro.




Additionally, any software company designing for, or using dx11 is already insignificant..!


This isnt' the general AMD v. nVidia trollbait thread of uselessness.

AT Moderator ElFenix

 

[VirtualLarry]

Odd, that you are a mainstay here and you don't understand that Nvidia will not have the performance crown long.

Pascal is EOL bro.

Additionally, any software company designing for, or using dx11 is already insignificant..!

LOL, is all I have to say to that.

Hey, I like AMD, I really, really do.

But as far as video cards, NVidia has shown time and time again, that they can seemingly time the market's demands perfectly. They also make a lot of money from GPUs, and can design and engineer halo cards.

If AMD can actually both out-engineer NVidia, at the very highest end of the market, AND successfully market that card or family of cards to actually SHOW A PROFIT, then I'll be impressed.

Until then, well, we'll see.

And DX11 isn't going anywhere soon.

Edit: You tell me that NV won't have the performance crown for long. Yet, AMD's 14nm-class dGPUs, the highest-end Polaris SKU, the RX 480, is only barely competitive, with NV's mid-range GTX 1060 SKU. What does that tell you about "performance engineering" at the 14nm level?

And the supposed "Better power efficiency" of Polaris, was essentially a lie, because AMD was force to push up the clockspeed of their Polaris GPUs, just to attempt to compete better with NV's cards.

 

[Head1985]

It looks like just rebrands.I was hoping for something like ATI 4870 to ATI 4890 atleast.
4890 have faster memory and 13% faster GPU clock+ it can oc to 1Ghz.It was not only rebrad it was respin.
http://www.anandtech.com/show/2745

Polaris 5xx looks like another boring rebrand.
Ryzen really starved GPU division to zero money.

 

[w3rd]

LOL, is all I have to say to that.

Hey, I like AMD, I really, really do.

But as far as video cards, NVidia has shown time and time again, that they can seemingly time the market's demands perfectly. They also make a lot of money from GPUs, and can design and engineer halo cards.

If AMD can actually both out-engineer NVidia, at the very highest end of the market, AND successfully market that card or family of cards to actually SHOW A PROFIT, then I'll be impressed.

Until then, well, we'll see.

And DX11 isn't going anywhere soon.

Edit: You tell me that NV won't have the performance crown for long. Yet, AMD's 14nm-class dGPUs, the highest-end Polaris SKU, the RX 480, is only barely competitive, with NV's mid-range GTX 1060 SKU. What does that tell you about "performance engineering" at the 14nm level?

And the supposed "Better power efficiency" of Polaris, was essentially a lie, because AMD was force to push up the clockspeed of their Polaris GPUs, just to attempt to compete better with NV's cards.

Nvidia = Jen-Hsun Huang, while AMD has had many CEOs.

So, trying suggest "traditionally/typically" anything about AMD is based on moot speculation. When you have a new leader at the helm. And you are correct, that nVidia has timed the market nearly perfectly, previously. But you and I both know the 1080ti is a tad too late for the current market, even if it's release mirrors the cadence of previous "Ti" releases. (resting on their heels?)

And yes, AMD has better technology than NVidia, just that in the past, they have not penned better deals with Fabs. So their competitiveness has always suffered. None of ANYTHING you have said, matters today.



And yes, AMD has better patents and tech than NVidia. Not sure why you aren't impressed already. Everyone else is eager to get their hands on said tech... even more odd, is that you are focused on being impressed by AMD's profits, but not their cards, or tech.

I suspect, you'll be here posting in thousands of threads, about how impressed you are once AMD stock raises..? Not sure what that does for gamers, but OK.

 

[frozentundra123456]

Nvidia = Jen-Hsun Huang, while AMD has had many CEOs.

So, trying suggest "traditionally/typically" anything about AMD is based on moot speculation. When you have a new leader at the helm. And you are correct, that nVidia has timed the market nearly perfectly, previously. But you and I both know the 1080ti is a tad too late for the current market, even if it's release mirrors the cadence of previous "Ti" releases. (resting on their heels?)

And yes, AMD has better technology than NVidia, just that in the past, they have not penned better deals with Fabs. So their competitiveness has always suffered. None of ANYTHING you have said, matters today.



And yes, AMD has better patents and tech than NVidia. Not sure why you aren't impressed already. Everyone else is eager to get their hands on said tech... even more odd, is that you are focused on being impressed by AMD's profits, but not their cards, or tech.

I suspect, you'll be here posting in thousands of threads, about how impressed you are once AMD stock raises..? Not sure what that does for gamers, but OK.

Yes, that AMD tech lead has given them the power efficiency and absolute performance crown for months, right?


Don't feed the trolls.

AT Moderator ElFenix

 

[VirtualLarry]

And yes, AMD has better technology than NVidia, just that in the past, they have not penned better deals with Fabs. So their competitiveness has always suffered. None of ANYTHING you have said, matters today.

You do realize, that ATI and then AMD, had their chips primarily fabbed by TSMC, the SAME fab that NV has traditionally used. So, fab competitive reasons are not the reason that AMD was behind NV in performance.

 

[frozentundra123456]

You do realize, that ATI and then AMD, had their chips primarily fabbed by TSMC, the SAME fab that NV has traditionally used. So, fab competitive reasons are not the reason that AMD was behind NV in performance.

Every time there is a new or even rehashed AMD product, it is the same story. New poster's with shall we say "enthusiastic" comments come out of the woodwork.

 

[mohit9206]

A faster than GT 730 GDDR5 for $60 would be nice.
The R7 240/250 are too old and not competitive. $50/60 low end gpu are still extremely popular in low income asian, south America countries etc.
Hopefully the rumored Polaris 12 with 640 shaders and 2gb gddr5 ram is the answer but even that will be $80.
Need a fresh new $60 Polaris card.
Sure one can just buy a better used card at that price but that's no excuse for not having some decent new cards at that price.

 

[Aristotelian]

. But you and I both know the 1080ti is a tad too late for the current market, even if it's release mirrors the cadence of previous "Ti" releases. (resting on their heels?)

I don't understand what you mean here. Why was the 1080Ti release 'too late for the current market'? Those who wanted extreme performance had already bought Titan Pascal; there were others who bought 1080s or 1070s depending on resolution. And now Nvidia releases the 1080Ti at a fairly decent price which beats Titan Pascal in many games. Who was the 1080Ti 'too late' for? The 1080Ti is almost constantly sold out in Europe...

Look, I try to be as fair and optimistic as the next guy. I wish that the highest end Vega would be out before the end of this year and stomp the 1080Ti. But I would not even bet on that happening, if the odds given to me were 1000:1 in favour of that happening.

If I was Nvidia I'd be happy with the 1080Ti release, see what Vega looks like when it comes out, and if the performance crown is threatened in any way, do any of the following:

(i) adjust pricing to remain competitive; and/or
(ii) try to push the Volta Titan out.

Nvidia is in a position now where it can 'rest on its heels' and see what AMD can do to respond to its dominant performance position. AMD has to react - and, depending how good that reaction is, Nvidia can respond as well.

Please don't misunderstand me - I want competition because I want to get the most for my money. But when we talk about optimism here I feel like I have very little reason to be as optimistic as you are.

 

[frozentundra123456]

I don't understand what you mean here. Why was the 1080Ti release 'too late for the current market'? Those who wanted extreme performance had already bought Titan Pascal; there were others who bought 1080s or 1070s depending on resolution. And now Nvidia releases the 1080Ti at a fairly decent price which beats Titan Pascal in many games. Who was the 1080Ti 'too late' for? The 1080Ti is almost constantly sold out in Europe...

Look, I try to be as fair and optimistic as the next guy. I wish that the highest end Vega would be out before the end of this year and stomp the 1080Ti. But I would not even bet on that happening, if the odds given to me were 1000:1 in favour of that happening.

If I was Nvidia I'd be happy with the 1080Ti release, see what Vega looks like when it comes out, and if the performance crown is threatened in any way, do any of the following:

(i) adjust pricing to remain competitive; and/or
(ii) try to push the Volta Titan out.

Nvidia is in a position now where it can 'rest on its heels' and see what AMD can do to respond to its dominant performance position. AMD has to react - and, depending how good that reaction is, Nvidia can respond as well.

Please don't misunderstand me - I want competition because I want to get the most for my money. But when we talk about optimism here I feel like I have very little reason to be as optimistic as you are.

Yea, don't understand how the 1080 Ti can be considered "late". I actually am surprised it came out this early, since anything high end is still MIA from AMD. If anyone is late in this segment, it is clearly AMD. Guess they have gone back to their speciality of releasing rebrands.

 

[railven]

Not sure if mentioned, but just reminds me of Hawaii to Grenada. AMD made some fine tune adjustments that basically fixed the initial ailments of Hawaii and put Grenada in a place where consumers felt the product was worth the investment.

A faster, quieter, cooler, less power hungry RX "480" can do the same, if marketed right.

I hope not, though, if AMD's next <$300 chip is going to be a slightly faster RX 480, those that got in on <$200 R9 290X Lightnings or Sapphire Tri-X are going to be possibly put off another cycle of upgrading or worse, changing sides.