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

[renz496]

don't be surprise if Vega end up being manufactured by TSMC though

 

[cdrkf]

don't be surprise if Vega end up being manufactured by TSMC though

I must admit I thought it was pretty much confirmed it is going to be a TSMC part? GloFo have only just started manufacturing GPU parts for AMD- they don't have the experience imo for the really big parts TSMC have been doing.

 

[Zaxx420]

Wasn't there talk of a lower performing 'affordable' HBM in the works? Can't recall the deets...

 

[Robert Cook]

Wasn't there talk of a lower performing 'affordable' HBM in the works? Can't recall the deets...

yeah it was in between HBM 1 and 2 and cost less than 1. but it sounds like it is not ready yet.

 

[renz496]

don't be surprise if Vega end up being manufactured by TSMC though

I must admit I thought it was pretty much confirmed it is going to be a TSMC part? GloFo have only just started manufacturing GPU parts for AMD- they don't have the experience imo for the really big parts TSMC have been doing.

i'd suspect Vega probably will end up with TSMC. but one day GF must also do big GPU for AMD? might as well let Vega be their first attempt on it instead just let TSMC do the big GPU

 

[renz496]

Wasn't there talk of a lower performing 'affordable' HBM in the works? Can't recall the deets...

but also slower than HBM2. heck might even be slower than HBM 1. at that point both GDDR5X and GDDR5 will become stiff competitor as well. need to look back the info on it

 

[TMTOWTSAC]

don't be surprise if Vega end up being manufactured by TSMC though

I must admit I thought it was pretty much confirmed it is going to be a TSMC part? GloFo have only just started manufacturing GPU parts for AMD- they don't have the experience imo for the really big parts TSMC have been doing.

i'd suspect Vega probably will end up with TSMC. but one day GF must also do big GPU for AMD? might as well let Vega be their first attempt on it instead just let TSMC do the big GPU

Heh, why not? Pressure from Pascal at the top end, limited Polaris availability, 1050 rumors, looming debt repayments, and Volta on the horizon. Not to mention everything riding on the Zen release. I doubt AMD wants to be making any other huge gambles right now, they've already got all their chips on the table, so to speak.

 

[renz496]

well i have this idea. i was suspecting that AMD having problems with GF14nm hence they only make Polaris 10 around 232mm2. but then again if AMD having issues with GF 14nm causing them unable to make bigger GPU than 232mm2 then why not make even bigger polaris at TSMC if GF cannot do it for them? they should know about this issue during evaluation time (to use TSMC or GF). the answer probably lies with PS4 Pro? it is possible that Polaris 10 is the side product they get after developing PS4 pro APU. developing new GPU will not be a cheap effort for AMD. hence we saw tons of "rebrands" after 7k series. so to save R&D AMD probably develop PS4 Pro APU and then use the GPU portion for the APU design for Polaris 10 as well.

so if Polaris 10 design was limited due to PS4 pro design that's mean it is possible that GF 14nm process still can handle larger gpu design from AMD? anyway people expecting Vega will be 500-600mm2 part. but i would not be surprise if Vega end up with 400mm2-500mm2 size.

 

[Martell1977]

well i have this idea. i was suspecting that AMD having problems with GF14nm hence they only make Polaris 10 around 232mm2. but then again if AMD having issues with GF 14nm causing them unable to make bigger GPU than 232mm2 then why not make even bigger polaris at TSMC if GF cannot do it for them? they should know about this issue during evaluation time (to use TSMC or GF). the answer probably lies with PS4 Pro? it is possible that Polaris 10 is the side product they get after developing PS4 pro APU. developing new GPU will not be a cheap effort for AMD. hence we saw tons of "rebrands" after 7k series. so to save R&D AMD probably develop PS4 Pro APU and then use the GPU portion for the APU design for Polaris 10 as well.

so if Polaris 10 design was limited due to PS4 pro design that's mean it is possible that GF 14nm process still can handle larger gpu design from AMD? anyway people expecting Vega will be 500-600mm2 part. but i would not be surprise if Vega end up with 400mm2-500mm2 size.

Are we sure it's an APU? I say this because the PS4 Pro seems to use the same CPU but higher clocked and what looks like either an OC'd 470 or heavily underclocked 480. Looking at the TFlops,:

PS4 Pro = 4.2
RX 480 = 5.1
RX 470 = 3.8

I'm wondering if Sony didn't go for a 2-piece option this time to save development time and costs. Plus, this would explain the supply shortages, as many of the chips would be sent to Sony.

 

[Sakkura]

That HBM alone could end up the card costing and arm and leg? Well at least that is my impression with AMD Fury series last year. Also it will be interested what changes can be seen in Vega. Will Vega still largely GCN or something new entirely? Papermark mentioned Vega will be better in terms of power efficiency. Will they achieve that with major architecture changes or did they intend to gain efficiency through HBM like Fury?

What? The Fury cards used a GPU that was the same size and on the same node as the Titan X and GTX 980 Ti. The Fury X was the uncut GPU, and came with AIO watercooling. It was much cheaper than Nvidia's Titan X, and cost the same as Nvidia's cut-down GPU without AIO watercooling. AMD's cut-down Fury and Nano cards were cheaper.

So obviously, whatever cost the HBM added, it did not show up in the price tag at all - in fact the HBM-equipped cards were cheaper than comparable hardware from Nvidia.

As for Vega, why are you wondering if it's GCN? It's been confirmed all along that it's GCN 4, just like Polaris. It's just a different macroarchitecture, the microarchitecture is the same.

 

[Zaxx420]

so if Polaris 10 design was limited due to PS4 pro design that's mean it is possible that GF 14nm process still can handle larger gpu design from AMD? anyway people expecting Vega will be 500-600mm2 part. but i would not be surprise if Vega end up with 400mm2-500mm2 size.

Are we sure it's an APU? I say this because the PS4 Pro seems to use the same CPU but higher clocked and what looks like either an OC'd 470 or heavily underclocked 480. Looking at the TFlops,:

PS4 Pro = 4.2
RX 480 = 5.1
RX 470 = 3.8

I'm wondering if Sony didn't go for a 2-piece option this time to save development time and costs. Plus, this would explain the supply shortages, as many of the chips would be sent to Sony.

Type-o...the 470 comes in at 4.8/4.9...

An o/cd 470 would probably cause cooling issues unless they really upped the case airflow I'm thinkin'. If the scorpio comes in at 6 TF...it'd almost have to be something besides Polaris...the 480 uses all the horsepower available on that chip iirc.

 

[Martell1977]

Type-o...the 470 comes in at 4.8/4.9...

An o/cd 470 would probably cause cooling issues unless they really upped the case airflow I'm thinkin'. If the scorpio comes in at 6 TF...it'd almost have to be something besides Polaris...the 480 uses all the horsepower available on that chip iirc.

I got that number from:
http://www.tomshardware.com/reviews/amd-radeon-rx-470,4703.html

If the 470 was a 4.8, that would imply an underclocked 470 un the PS4 Pro.

Page 1, 4th paragraph under the "Meet Radeon RX 470 section. Unless that number has changed or Tom's made an error in the review.

I seems the 480 is pretty much topped out as OC potential seems very limited. I guess the PS4 Pro could be a custom chip, Just seems odd that they would design a new APU with the same CPU cores and all new GPU side. A lot of what I have read says that the CPU side has been more a limiting factor for the PS4 than the GPU.

Either way, it's seems like that either the chips were divided between Sony and the market or the production capability was split between the custom APU and Polaris GPU's.

 

[renz496]

That HBM alone could end up the card costing and arm and leg? Well at least that is my impression with AMD Fury series last year. Also it will be interested what changes can be seen in Vega. Will Vega still largely GCN or something new entirely? Papermark mentioned Vega will be better in terms of power efficiency. Will they achieve that with major architecture changes or did they intend to gain efficiency through HBM like Fury?

What? The Fury cards used a GPU that was the same size and on the same node as the Titan X and GTX 980 Ti. The Fury X was the uncut GPU, and came with AIO watercooling. It was much cheaper than Nvidia's Titan X, and cost the same as Nvidia's cut-down GPU without AIO watercooling. AMD's cut-down Fury and Nano cards were cheaper.

So obviously, whatever cost the HBM added, it did not show up in the price tag at all - in fact the HBM-equipped cards were cheaper than comparable hardware from Nvidia.

As for Vega, why are you wondering if it's GCN? It's been confirmed all along that it's GCN 4, just like Polaris. It's just a different macroarchitecture, the microarchitecture is the same.

then look at the launch price of Fury X. titan x was at 1k. back then some people expecting 980Ti will be in the $800 range. but suddenly nvidia launch 980ti ahead of amd fury x at $650. that is even cheaper than 780ti debut price. if we look at AMD history they usually undercut nvidia price even if the performance is end up on the same level (7970 is the only exception). 290X was priced at $550 when GTX780 was $600. 390X 8GB was priced at $420 vs 980 at $550. then why Fury X did not undercut 980Ti pricing but instead have the same price at $650? that's a sign that AMD probably did not want to price Fury X lower than that but they also cannot price it higher because competition from 980Ti. and then MSRP for Nano at launch also $650. the card is slower than Fury X, no AIO cooler and yet why AMD want to sell them at such high price?

http://www.anandtech.com/show/9621/the-amd-radeon-r9-nano-review

they were probably cheap from customer point of view when compared to nvidia product but it doesn't mean it is also cheaper for AMD to make them. there are early rumors indicating that AMD intend to sell Fury X in the range of $800-$900 competing directly with nvidia titan x. just like titan is nvidia premium product it seems AMD also intend Fury to be their premium product. hence the Fury naming instead include the card in the R 300 series.

about Vega some people are speculating that Vega will be entirely new architecture instead of another GCN iteration just like Polaris.

 

[BurgerandChips66]

The PS4 Pro still uses an APU: - http://www.gamespot.com/articles/ps4-pro-specs-release-date-and-price-confirmed/1100-6443352/

 

[Sakkura]

then look at the launch price of Fury X. titan x was at 1k. back then some people expecting 980Ti will be in the $800 range. but suddenly nvidia launch 980ti ahead of amd fury x at $650. that is even cheaper than 780ti debut price. if we look at AMD history they usually undercut nvidia price even if the performance is end up on the same level (7970 is the only exception). 290X was priced at $550 when GTX780 was $600. 390X 8GB was priced at $420 vs 980 at $550. then why Fury X did not undercut 980Ti pricing but instead have the same price at $650? that's a sign that AMD probably did not want to price Fury X lower than that but they also cannot price it higher because competition from 980Ti. and then MSRP for Nano at launch also $650. the card is slower than Fury X, no AIO cooler and yet why AMD want to sell them at such high price?

http://www.anandtech.com/show/9621/the-amd-radeon-r9-nano-review

they were probably cheap from customer point of view when compared to nvidia product but it doesn't mean it is also cheaper for AMD to make them. there are early rumors indicating that AMD intend to sell Fury X in the range of $800-$900 competing directly with nvidia titan x. just like titan is nvidia premium product it seems AMD also intend Fury to be their premium product. hence the Fury naming instead include the card in the R 300 series.

So the Fury X cost an arm and a leg because it cost the same as the 980 Ti that was surprisingly cheap, while the Fury X included AIO watercooling?

The Fury X was indeed AMD's Titan X. Same size chip, and added cost from watercooling and HBM. Yet it cost far less, and matched Nvidia's surprisingly cheap cut-down card, in fact undercut it once you account for the watercooling.

The R9 Nano used binned chips and was in a niche market segment where there was no relevant competition from the Fury cards or 980 Ti etc. So AMD could charge a premium. At least they could for a little while, but it's a tiny niche and they soon had to cut the price.

And yes, the Fury cards were probably costly for AMD to make. But as I said, that didn't show up in the price tag. So why are you saying it will definitely show up in the price tag now? If anything, history teaches us that it will not.

 

[Embra]

then look at the launch price of Fury X. titan x was at 1k. back then some people expecting 980Ti will be in the $800 range. but suddenly nvidia launch 980ti ahead of amd fury x at $650. that is even cheaper than 780ti debut price. if we look at AMD history they usually undercut nvidia price even if the performance is end up on the same level (7970 is the only exception). 290X was priced at $550 when GTX780 was $600. 390X 8GB was priced at $420 vs 980 at $550. then why Fury X did not undercut 980Ti pricing but instead have the same price at $650? that's a sign that AMD probably did not want to price Fury X lower than that but they also cannot price it higher because competition from 980Ti. and then MSRP for Nano at launch also $650. the card is slower than Fury X, no AIO cooler and yet why AMD want to sell them at such high price?

http://www.anandtech.com/show/9621/the-amd-radeon-r9-nano-review

they were probably cheap from customer point of view when compared to nvidia product but it doesn't mean it is also cheaper for AMD to make them. there are early rumors indicating that AMD intend to sell Fury X in the range of $800-$900 competing directly with nvidia titan x. just like titan is nvidia premium product it seems AMD also intend Fury to be their premium product. hence the Fury naming instead include the card in the R 300 series.

So the Fury X cost an arm and a leg because it cost the same as the 980 Ti that was surprisingly cheap, while the Fury X included AIO watercooling?

The Fury X was indeed AMD's Titan X. Same size chip, and added cost from watercooling and HBM. Yet it cost far less, and matched Nvidia's surprisingly cheap cut-down card, in fact undercut it once you account for the watercooling.

The R9 Nano used binned chips and was in a niche market segment where there was no relevant competition from the Fury cards or 980 Ti etc. So AMD could charge a premium. At least they could for a little while, but it's a tiny niche and they soon had to cut the price.

And yes, the Fury cards were probably costly for AMD to make. But as I said, that didn't show up in the price tag. So why are you saying it will definitely show up in the price tag now? If anything, history teaches us that it will not.

^ Well stated Sakkura. What hurt the Fury X's value was the lack of OCing, is was an amazing design at launch. Sill a good card.

 

[cdrkf]

then look at the launch price of Fury X. titan x was at 1k. back then some people expecting 980Ti will be in the $800 range. but suddenly nvidia launch 980ti ahead of amd fury x at $650. that is even cheaper than 780ti debut price. if we look at AMD history they usually undercut nvidia price even if the performance is end up on the same level (7970 is the only exception). 290X was priced at $550 when GTX780 was $600. 390X 8GB was priced at $420 vs 980 at $550. then why Fury X did not undercut 980Ti pricing but instead have the same price at $650? that's a sign that AMD probably did not want to price Fury X lower than that but they also cannot price it higher because competition from 980Ti. and then MSRP for Nano at launch also $650. the card is slower than Fury X, no AIO cooler and yet why AMD want to sell them at such high price?

http://www.anandtech.com/show/9621/the-amd-radeon-r9-nano-review

they were probably cheap from customer point of view when compared to nvidia product but it doesn't mean it is also cheaper for AMD to make them. there are early rumors indicating that AMD intend to sell Fury X in the range of $800-$900 competing directly with nvidia titan x. just like titan is nvidia premium product it seems AMD also intend Fury to be their premium product. hence the Fury naming instead include the card in the R 300 series.

So the Fury X cost an arm and a leg because it cost the same as the 980 Ti that was surprisingly cheap, while the Fury X included AIO watercooling?

The Fury X was indeed AMD's Titan X. Same size chip, and added cost from watercooling and HBM. Yet it cost far less, and matched Nvidia's surprisingly cheap cut-down card, in fact undercut it once you account for the watercooling.
.
The R9 Nano used binned chips and was in a niche market segment where there was no relevant competition from the Fury cards or 980 Ti etc. So AMD could charge a premium. At least they could for a little while, but it's a tiny niche and they soon had to cut the price.

And yes, the Fury cards were probably costly for AMD to make. But as I said, that didn't show up in the price tag. So why are you saying it will definitely show up in the price tag now? If anything, history teaches us that it will not.

Well the pricing will be determined by where it sits relative to the competition. AMD try to be *a bit* cheaper- but I'd hardly call the Fury X cheap. It's more a case of nVidia taking the **** with the Titan prices that made Fury X appear 'cheap'. Not saying it was overpriced either- I think Fury X was priced where it needed to be given the chip size, interposer, HBM and water cooling.

As for Vega- if it performs as fast or faster than nVidias best at the time- expect it to be priced high. If it performs between the 1070 and 1080 then it will be priced in that range imo. If it's faster than the 1080ti, well- at best expect it to cost the same. AMD are a business after all.

 

[BurgerandChips66]

AMD is planning to replace Polaris 10 with Vega 11 next year. At least that is the latest rumour.

http://videocardz.com/63700/exclusive-first-details-about-amd-vega10-and-vega20

Apparently there will be Vega 10, 11 and 20.

 

[renz496]

then look at the launch price of Fury X. titan x was at 1k. back then some people expecting 980Ti will be in the $800 range. but suddenly nvidia launch 980ti ahead of amd fury x at $650. that is even cheaper than 780ti debut price. if we look at AMD history they usually undercut nvidia price even if the performance is end up on the same level (7970 is the only exception). 290X was priced at $550 when GTX780 was $600. 390X 8GB was priced at $420 vs 980 at $550. then why Fury X did not undercut 980Ti pricing but instead have the same price at $650? that's a sign that AMD probably did not want to price Fury X lower than that but they also cannot price it higher because competition from 980Ti. and then MSRP for Nano at launch also $650. the card is slower than Fury X, no AIO cooler and yet why AMD want to sell them at such high price?

http://www.anandtech.com/show/9621/the-amd-radeon-r9-nano-review

they were probably cheap from customer point of view when compared to nvidia product but it doesn't mean it is also cheaper for AMD to make them. there are early rumors indicating that AMD intend to sell Fury X in the range of $800-$900 competing directly with nvidia titan x. just like titan is nvidia premium product it seems AMD also intend Fury to be their premium product. hence the Fury naming instead include the card in the R 300 series.

So the Fury X cost an arm and a leg because it cost the same as the 980 Ti that was surprisingly cheap, while the Fury X included AIO watercooling?

The Fury X was indeed AMD's Titan X. Same size chip, and added cost from watercooling and HBM. Yet it cost far less, and matched Nvidia's surprisingly cheap cut-down card, in fact undercut it once you account for the watercooling.

The R9 Nano used binned chips and was in a niche market segment where there was no relevant competition from the Fury cards or 980 Ti etc. So AMD could charge a premium. At least they could for a little while, but it's a tiny niche and they soon had to cut the price.

And yes, the Fury cards were probably costly for AMD to make. But as I said, that didn't show up in the price tag. So why are you saying it will definitely show up in the price tag now? If anything, history teaches us that it will not.

by any mean $650 for a graphic alone is not "cheap" either. that $650 for fury x indicate amd does not want to sell them for lower than that but they have no other choice because of competition. heck they maybe have intention to give some headroom for Fury X hence the talk about "overclockers dream" at first. but it turns out just to be competitive with vanilla 980ti already push Fury X to it's limit. AMD end up cutting the price of nano most likely because the same are not picking up. it is better be able to sell the hardware with mininal profit (or none at all) rather than let the stock unsold. because at this point the sluggish sales will also going to affect their board partner revenue.

and this time if there is no pressure from nvidia (just imagine if they have something faster than 1080) they probably will price them accordingly with performance. right now AMD flagship sits at $650 (MSRP). depending on how nvidia priced their product AMD might increase that to $700-$750.

 

[renz496]

AMD is planning to replace Polaris 10 with Vega 11 next year. At least that is the latest rumour.

http://videocardz.com/63700/exclusive-first-details-about-amd-vega10-and-vega20

Apparently there will be Vega 10, 11 and 20.

so the competition is being heat up

 

[Martell1977]

AMD is planning to replace Polaris 10 with Vega 11 next year. At least that is the latest rumour.

http://videocardz.com/63700/exclusive-first-details-about-amd-vega10-and-vega20

Apparently there will be Vega 10, 11 and 20.

The Vega10 information is pretty much what was rumored.

The Vega20 information is very interesting, it seems that AMD is expecting an explosion in VRAM demand or it is a workstation card primarily. Plus, being PCIe 4.0 and 7nm, I expect it to be a 2H 2017 or 2018 card at best, which makes sense if they really did push Navi to 2019.

Polaris 10/Vega 11 replacement I find odd. I don't see why they would look at replacing Polaris 10. They seem to be selling plenty of them and are a solid card. Now I could see them making a 485 and squeeze in Vega 11, or maybe changing their naming to allow it to the the 490. If they replace Polaris 10 with Vega 11 and it changes performance, a LOT of customers will feel burned and that could do a lot of harm.

For some reason people hold things against AMD but let nVidia slide. i.e. If AMD had pulled the 970 3.5+0.5 VRAM thing or selling a cut down GPU and calling it the same name as the full one(1060 3gb vs 6gb), there would be picketing outside AMD's headquarters.

 

[renz496]

Lol i also see opposite things happen. Remember how people say fermi is hot and going to increased your power bill? Back then many said efficiency is vert important. But when the tables got turned around we see people claiming that power consumption does not matter only real performance does. In short it happens on both sides actually.

 

[IAEInferno]

What would be the use of 32gb of Vram for?

 

[manleysteele]

What would be the use of 32gb of Vram for?

Any large data set that changes over time. Dynamic systems such as weather are very hard to model in a useful timeframe. Aerodynamic info, ballistic characteristics of projectiles in atmosphere. There are plenty of problem domains that use large memory footprints.

 

[ex_bubblehead]

What would be the use of 32gb of Vram for?

Any large data set that changes over time. Dynamic systems such as weather are very hard to model in a useful timeframe. Aerodynamic info, ballistic characteristics of projectiles in atmosphere. There are plenty of problem domains that use large memory footprints.

And for CAD/CAM:
[strike]Large[/strike] Humongous texture maps.
Room for 100's of thousands/millions of objects.

The ability of Photoshop to do transforms and such on much larger images.

The ability to keep much more detailed scenery, covering a much larger portion of the earth, in VRAM when flying in X-Plane (or other similar software), thus eliminating many loads that introduce stuttering/pausing while flying.