12nm Polaris Refresh Coming in October

[xkm1948]

Source: wjm47196 from Chiphell, as usual

https://translate.google.com/transl...om/thread-1911969-1-1.html&edit-text=&act=url

Definitely not gonna be a challenger for RTX2080/Ti. But probably gonna bring something new to the table for lower end product.

It is very much likely to be Polaris 30.

Once again take these rumors with your preferable measurement of salt.

 

[HTC]

Source: wjm47196 from Chiphell, as usual

https://translate.google.com/translate?sl=auto&tl=en&js=y&prev=_t&hl=en&ie=UTF-8&u=https://www.chiphell.com/thread-1911969-1-1.html&edit-text=&act=url

Definitely not gonna be a challenger for RTX2080/Ti. But probably gonna bring something new to the table for lower end product.

It is very much likely to be Polaris 30.

Once again take these rumors with your preferable measurement of salt.

Seriously doubt even a Vega 64 refresh on 7nm would be enough against RTX 2080 Ti: for that, most likely they'll need a new arch, IMO. Unless ofc it offers performance higher then the 1080 Ti @ a much smaller price, which i even more seriously doubt due to HBM2 prices forcing AMD to price any such cards quite high.

 

[xkm1948]

I think Polaris re-refresh is looking more to compete with 1060 6GB to 1070Ti.

 

[cucker tarlson]

Come on,Polaris refresh .......

 

[HTC]

I think Polaris refresh is looking more to compete with 1060 6GB to 1070Ti.

@ least 1070 should be quite reachable: dunno about 1070 Ti. If it were in 7nm, then 1070 Ti would be in "danger" i think, depending on price, ofc.

 

[RejZoR]

The problem with Polaris is, if they make it "too good" it'll start to interfere with Vega 56. So, AMD is kinda in a weird position with Polaris.

@HTC
Actually, a properly refreshed Vega would still sell. It doesn't have to be better than RTX whatever. RTX cards have such ridiculous price it doesn't even matter and if there will be a shortage of Pascal cards anytime soon, AMD could use this to their advantage. Question is, when will they launch such Vega refresh and for how much. If they make Vega as fast as 1080Ti and sell it for as much, it could work. RTX 2080 is like 1000€ and is as fast as GTX 1080Ti. So, if AMD gets Vega there and sells it for 200€ less, it's gonna be a win. Not everyone needs king of the hill card and Polaris proved that on and on. Some for the mining, but if they were available for gaming, they'd go like hot cakes just the same.

 

[HTC]

The problem with Polaris is, if they make it "too good" it'll start to interfere with Vega 56. So, AMD is kinda in a weird position with Polaris.

@HTC
Actually, a properly refreshed Vega would still sell. It doesn't have to be better than RTX whatever. RTX cards have such ridiculous price it doesn't even matter and if there will be a shortage of Pascal cards anytime soon, AMD could use this to their advantage. Question is, when will they launch such Vega refresh and for how much. If they make Vega as fast as 1080Ti and sell it for as much, it could work. RTX 2080 is like 1000€ and is as fast as GTX 1080Ti. So, if AMD gets Vega there and sells it for 200€ less, it's gonna be a win. Not everyone needs king of the hill card and Polaris proved that on and on. Some for the mining, but if they were available for gaming, they'd go like hot cakes just the same.

Agree: i was just referring performance. If such a Vega refresh had ... say ... 20% less performance then RTX 2080 Ti but sold for ... say ... $800, nVidia would be forced to change prices of their RTX line. Not that we would mind that scenario ...

 

[xkm1948]

Vega cost of manufacturing is too high due to interposer and HBM2. A shrinked Polaris makes a whole lot more sense as the cost to make them is very low.

 

[HTC]

Vega cost of manufacturing is too high due to interposer and HBM2. A shrinked Polaris makes a whole lot more sense as the cost to make them is very low.

Actually, the interposer is quite cheap because it has virtually no transistors in it and it is produced @ a bigger node than Vega itself. IIRC, Vega's interposer is fabricated in 65nm node, for example.

It's the HBM2 that drives Vega card prices high.

 

[HD64G]

Some of us have posted about a Polaris refresh on 12nm with GDDR6 a few months ago as a possibility for AMD to try and stay relevant in GPU market until Navi is out. Might work for them to keep some piece of the mid-performance market theirs until 7nm matures for consumer-grade GPUs. They cannot get much higher performance than RX580 though. About 20% max imho. And that is enough to just reach 1070 for well optimised DX11 games and surpass on DX12-Vulkan ones. But TDP will remain higher for sure. 14 to 12nm cannot do much as RTX 20X0 GPUs showed clearly.

 

[Joss]

The problem with Polaris is, if they make it "too good" it'll start to interfere with Vega 56

If that's the case they can just scrap Vega 56 and position Vega 64 right above RX 680 (or whatever it's called).

 

[R0H1T]

The problem with Polaris is, if they make it "too good" it'll start to interfere with Vega 56. So, AMD is kinda in a weird position with Polaris.

@HTC
Actually, a properly refreshed Vega would still sell. It doesn't have to be better than RTX whatever. RTX cards have such ridiculous price it doesn't even matter and if there will be a shortage of Pascal cards anytime soon, AMD could use this to their advantage. Question is, when will they launch such Vega refresh and for how much. If they make Vega as fast as 1080Ti and sell it for as much, it could work. RTX 2080 is like 1000€ and is as fast as GTX 1080Ti. So, if AMD gets Vega there and sells it for 200€ less, it's gonna be a win. Not everyone needs king of the hill card and Polaris proved that on and on. Some for the mining, but if they were available for gaming, they'd go like hot cakes just the same.

Probably when Pascal stocks deplete enough, so about a quarter to three from now. This is assuming there is a Vega refresh coming for consumers & AMD haven't thrown in the towel waiting for Navi.

 

[xkm1948]

I can totally see RTG stop production of gaming Vega(64 & 56) and focus solely on Polaris re-refresh until Navi comes along

 

[chaosmassive]

Yes Polaris is great, and Vega too
but its about time AMD move on to new uarch
you dont want same uarch rebranded three times again, just like HD 7000 series...
where is Navi by the way?

 

[xkm1948]

Yes Polaris is great, and Vega too
but its about time AMD move on to new uarch
you dont want same uarch rebranded three times again, just like HD 7000 series...
where is Navi by the way?

Navi is still GCN. The same design that has been in use since before 2011. So nope, don't even count on Navi if you want something completely new.

 

[RejZoR]

Lol, people still keep on doing this, they look at GCN and it's nope, can't be good no matter what. It's quite funny actually. GCN in HD7000 or GCN in Vega is like two radically different things. But both are GCN. Basically, it's like saying 4 cylinder engine can't ever be good because it's 4 cylinder. We know they can make monster 4 cylinder engines. GCN isn't much different. GCN by itself doesn't define much or anything. It's just a 3 letter name.

 

[ppn]

what is refresh good for. Up the units by 50%. 2034 to 3456 even 4096(the limit of GCN) and have a GDDR6 card ready asap. How hard could it be to just copy pase some units and give us 50% bigger chip, Cmon.

 

[xkm1948]

what is refresh good for. Up the units by 50%. 2034 to 3456 even 4096(the limit of GCN) and have a GDDR6 card ready asap. How hard could it be to just copy pase some units and give us 50% bigger chip, Cmon.

Very very hard compared to what our couch GPU experts' might think

It is not a simple scale up or down. Everything has to be designed first by engineers at RTG. Then back and forth with TSMC/GloFo/Samsung to fine tune the prototype chip before molds are made for mass production. Even after that it takes a while to improve yields and fine tune eletronic properties. Process can easily take several years for a single GPU design. For a cash stripped RTG, it makes sense to refresh a GPU as much as possible.

 

[ppn]

I dont believe it. Digital foundry showed the photos, what AMD did is double the CU count of the playstation PRO and they didn't even look copy pasted, but different shape and size.

 

[MrGenius]

you dont want same uarch rebranded three times again, just like HD 7000 series...

I get what you're saying. But, as a blanket statement, it's not really true. The HD 7000 series is actually comprised of 3 different µarch(4 if you count the HD 7000 TeraScale 3 IGPs, Devastator and Scrapper). TeraScale 2(Caicos, Cedar, Turks), GCN 1(Cape Verde, New Zealand, Pitcairn, Tahiti), and GCN 2(Bonaire). Some of which were already rebranded(Caicos, Cedar, Turks). One of which wasn't rebranded afterwards(Turks). Only one of which was rebranded 3 times(Pitcairn HD 7850/7870, HD 8870, R7 265, R9 270(X), R7 370, R9 370X). And two of which were actually rebranded 4 times(Bonaire HD 7790, HD 8770, R7 260(X), R7/R9 360, RX 455 & Cape Verde HD 7730/7750/7770, HD 8730/8760, R7 250(E,X), R7 350, R7 450).

 

[Frick]

Would be nice, but as long as the prices are what they are I'm not interested. €200 is pretty much my pain threshhold for new GPUs.