Clues Gather Regarding Possible New AMD Polaris (Re)Revision Launch

[Assimilator]

Nobody really knows quite what AMD is gonna do right now.

AMD isn't big enough to execute successfully on more than one front. Currently they're succeeding on the CPU front with Zen, so that's where they're focusing their efforts, and as such graphics isn't going to get any love for a while. Once Zen starts to lose its shine, they'll shift focus back to GPUs - assuming that NVIDIA hasn't sewn up the market by then.

so we are prolly looking at RX585/590/595

Nah, they'll make it the 600 series to make it look new compared to the 500 series and thus get sales from suckers. Rebranding is RTG's forte.

 

[DeathtoGnomes]

AMD isn't big enough to execute successfully on more than one front.

wait, what!!? what fronts are you referring to? video cards? consoles? cpus? server chips? AI cards?

This comments just baffles me.

Nah, they'll make it the 600 series to make it look new compared to the 500 series and thus get sales from suckers. Rebranding is RTG's forte.

very well could be

 

[Vya Domus]

What are the chances of these just being OEM cards RX 500 series.

Going through the trouble of switching to a slightly better node and slightly better performance just for OEMs ? Doubt it.

AMD isn't big enough to execute successfully on more than one front.

Yeah they are, this notion that they are relying exclusively on CPUs is baloney. APUs, HPC and yes gaming, all of those things are part of their portfolio and it makes them money.

Funny, they went from people forgetting they make CPUs to everyone thinking that's all they focus on in less than what, 2 years ?

 

[lexluthermiester]

This comments just baffles me.

Right there with you. Sheer silliness. AMD not only has the resources but has them to spare.

 

[Casecutter]

I find it strange how now the "marketing" to the mid-range enthusiast segment is now needing to be a $300-500 price.

 

[windwhirl]

Revised Polaris, revised. Revision inception?

Unless there is some heavy reengineering, I doubt AMD could get more than a 10% performance increase for Polaris-based GPUs. Hope they prove me wrong, though.

 

[Fluffmeister]

I suspect AMD really want to put at least Vega 56 to bed, having a large HMB2 chip competing against multiple variants of the much smaller GP104 does their margins no favours at all.

 

[Casecutter]

Unless there is some heavy reengineering, I doubt AMD could get more than a 10% performance increase for Polaris

That said if they possibly see a 7-9% in just from the GPU & GDDR memory upticks, then a modest 10% clock increase from the shrink along with some improved boost algorithms all while maintain to a 185W TDP... Would that get it more to the FuryX/980ti level, not a bad position if they hold the price to $230.

AMD really want to put at least Vega 56 to bed,

IDK, At first before miners grabbed them up, it was probably hard when HBM prices escalated well beyond their original "costed" BOM, but the mining "boon" probably saved them. Today if Vega was still underwater I'm sure we would not be seeing pricing at (and below) the original stated MSRP's. Don't worry AMD get's their margins retrospective of where the competition is. As long as the can sell what they produce and keep the channel filled and price equilibrium they're making their nut just fine.

 

[DeathtoGnomes]

I find it strange how now the "marketing" to the mid-range enthusiast segment is now needing to be a $300-500 price.

You have the memory price fixing scandal and mining to thank. Prices should be closer to normal again soon.

 

[Nkd]

how can a mere 2nm node reduction (from 14 nm to 12nm) give an increase of 15% performance.
maybe 5-10% performance increase with AMD pushing the clock rates beyond the limits thereby having like 220W same as rtx 2080( because rx 580 = gtx 1080 in efficiency) power consumption for card only
thereby making it DOA...

12nm TSMC is much more refined. You can't just assume it will be in no way 15% more effecient. They can easily get there with slight node shrink. 12nm has its advantages and it is more efficient than 14nm. I am sure AMD has learned to tweak Polaris overtime as well. Global Foundry wasn't a process designed for high clocks but AMD did their best. I think had they been on TSMC they could likely clock vega and polaris higher even at 14nm. If they are indeed moving to 12nm TSMC then we can definitely see 15% improvement. Even GF 12nm was said to be about 15% performance improvements at the same power but that was best case scenario and I don't really trust GF with high clocks on GPU.

 

[InVasMani]

how can a mere 2nm node reduction (from 14 nm to 12nm) give an increase of 15% performance.
maybe 5-10% performance increase with AMD pushing the clock rates beyond the limits thereby having like 220W same as rtx 2080( because rx 580 = gtx 1080 in efficiency) power consumption for card only
thereby making it DOA...

12 divided by 14 is about 15% difference plus on top of that performance and efficiency aren't linear eventually you reach a point of diminishing returns which can be gradual to severe. Based on that 15% isn't too far fetched actually. Should be interesting if the power comes down slightly along with the extra performance and lower cost to build. It would be cool if they made these PCIE 4.0 compliant. I would be interesting if they could squeeze on another chip or two of memory two as a result of the die shrink as well 10GB/12GB Polaris card perhaps?

 

[eidairaman1]

I find it strange how now the "marketing" to the mid-range enthusiast segment is now needing to be a $300-500 price.

Yeah considering 4 years ago a 290 Sapphire vapor-x 4 gig was 460

 

[Keullo-e]

The original Polaris was already "meh", performance similar to the old R9 390 series, problems with PCI-E specifications (reference design), bad overclockability..

 

[Assimilator]

wait, what!!? what fronts are you referring to? video cards? consoles? cpus? server chips? AI cards?

This comments just baffles me.

I should probably have prefaced that comment with "in the high-end consumer space".

 

[hat]

Well, why not? If they're still producing Polaris, it would make more sense to move to a more efficient node. We get a better product, maybe better prices (yeah right) and it's probably cheaper for them to produce... though calling it RX680 would be a stretch.

 

[Vya Domus]

12 divided by 14 is about 15% difference

That is not how this works at all.

problems with PCI-E specifications (reference design)

An obscure problem blown way out of proportion and a couple of claims with no proof. Not to mention that a crap ton of cards go over the spec anyway.

The original Polaris was already "meh", performance similar to the old R9 390 series

Meh performance but not meh power efficiency, which was the whole point of Polaris.

 

[efikkan]

12 divided by 14 is about 15% difference plus on top of that performance and efficiency aren't linear eventually you reach a point of diminishing returns which can be gradual to severe. Based on that 15% isn't too far fetched actually. Should be interesting if the power comes down slightly along with the extra performance and lower cost to build. It would be cool if they made these PCIE 4.0 compliant. I would be interesting if they could squeeze on another chip or two of memory two as a result of the die shrink as well 10GB/12GB Polaris card perhaps?

It has been pointed out several times in this thread that "12nm" is not a node shrink, it's just a refinement, similar to what Intel did with 14nm+ and 14nm++. The density is not changed, but the thermals and/or yields might be marginally improved.

I don't know where people get this "15%" from. If they do it like you and base it on marketing numbers for nodes, then it's complete nonsense. 15% from a node refinement alone seems highly unusual, somewhere around 5% is more common.

It's highly unlikely that they will "squeeze on another chip or two of memory two as a result of the die shrink".

 

[lexluthermiester]

though calling it RX680 would be a stretch.

Maybe RX585/RX565? LOL!

 

[Assimilator]

Meh performance but not meh power efficiency, which was the whole point of Polaris.

Cool story bro!

https://www.techpowerup.com/reviews/Sapphire/RX_580_Nitro_Plus/35.html

Cons: Very high gaming power consumption

Gaming power draw is really high though; with 214 W, it is 50 W higher than the GTX 1080, which is much faster at the same time, of course. To me, it looks like AMD's GPU fabrication process improvements only help with achieving higher clocks, with no reduction in power consumption. Looking at performance-per-watt numbers, the RX 580 has dropped back to R9 Fury X levels, which makes NVIDIA's GPU more than twice as energy efficient.

 

[Vya Domus]

Cool story bro!

https://www.techpowerup.com/reviews/Sapphire/RX_580_Nitro_Plus/35.html

The original Polaris was already "meh", performance similar to the old R9 390 series

Context is a difficult thing, I know, don't worry bro.

 

[Jism]

how can a mere 2nm node reduction (from 14 nm to 12nm) give an increase of 15% performance.
maybe 5-10% performance increase with AMD pushing the clock rates beyond the limits thereby having like 220W same as rtx 2080( because rx 580 = gtx 1080 in efficiency) power consumption for card only
thereby making it DOA...

Their 15% performance is based on the node and not actual FPS.

So you could say that this new node would offer a 15% better performance in terms of heat, clocks and required voltage compared to the previous node, RX580.

So if it was possible to have the RX680 work at 1500Mhz with GDDR6 then it might be even faster then the 15% performance. It should be in my opinion towards the Vega 56 to be succesfull.

The Polaris was succesfull btw. It is the best card you can get for the money for Full HD gaming. Dont mind the miners buying every card up from AMD; and the online shops putting their margin on top of that.

 

[Casecutter]

4 years ago a 290 Sapphire vapor-x

I don't consider the "Hawaii" when it released 11/13 as the "mid-range enthusiast"... To me while a serious gamer who plays a few times per week, they can't afford the bleeding edge at least when it's first released. A true "Enthusiast" a person playing pretty much daily, and has no qualms in buying the latest high-end price when released, retrospective of actual gaming enhancement or immersion. In 2014 the R9 280(X) that topped out at $300ish would be a card, what I term mid-range enthusiast might have inclined to buy.

It should be in my opinion towards the Vega 56 to be succesfull.

As much as we'd like a Polaris to be at Vega 56 1440p level that a chip that has 3584 SP vs. Polaris at 2304 SP. That a big leap and means a true new Polaris design. I'd say that's not what's happening if there's any truth to his Polaris 30 rumor.

If they re-spin and optimize the current Polaris, go with GDDR6 as a "pipe-cleaner" exercise, get what little extra's they can pull from just a process node shrink, even refined boost algorithms to push past 1500Mhz, all perhaps in a 175W envelope with a $230 price. A card like that has a lot going for it, while there's the geldings (570) versions, and that's where perhaps the best attack on 1080p would bolster the "entry-novice" as that would keep them pulling market share.

 

[yeeeeman]

AMD needs to release a card that is close to 1080Ti with 200W TDP on 7nm. They can strap two of those together on the same card if necessary.

 

[WikiFM]

Clues have been popping here and there regarding a possible new Polaris revision being launched by AMD in the (relatively) near future. Speculation first reared its head regarding a revised "Polaris 30" silicon, allegedly being built for TSMC's 12 nm process - not unlike AMD's 2000-series Ryzen CPUs.
@Raevenlord Ryzen 2000 Series is being built on GF's 12 nm, not on TSMC's 12 nm.

Some publications are pointing towards a 15% performance improvement being achieved on the back of this process change for Polaris - which, if achieved only via a new process implementation, would require clock speed increases that are higher than that. AMD has already launched their revised Polaris 20 RX 500 series, which built upon their RX 400 series (and Polaris 10) by upping the clocks as well. A smaller node would likely be associated with higher yields and decreased costs per finished chip, which would allow AMD to further reduce pricing/stabilize pricing while introducing a new product generation to tide users over until Navi is finally ready.
If Polaris 30 is being built on TSMC's 12 nm then is not comparable to Polaris 10 or 20 which are built on GF's 14 nm instead. The metrics are not compatible between different foundries, we can't be sure if is a smaller node after all, so any performance uplift is uncertain.

 

[medi01]

Remember, the real reason Fury and Vega were coupled to HBM is that its power consumption is so much lower than GDDR

The reason Fury and Vega was coupled to HBM is that AMD, being an underdog with next to no money for R&D had to take the risk and bet on something not traditional to have chances to win.

It worked quite well with Zen (Infinity Fabric), it didn't work with HBM, which nVidai also had project to develop with, just in case and because they sit on heaps of money, (thank you very much for buying overpriced cards), but unlike AMD, had resources for alternative, conventional projects

AMD needs to release a card that is close to 1080Ti with 200W TDP on 7nm. They can strap two of those together on the same card if necessary.

TI is unlikely, but 1080 levels are realistic to expect.