What are AMD's plans next year to counter Pascal?

[poohbear]

What is the roadmap for Radeon cards from AMD? This year seemed very disappointing with mostly rebrands. Is there a new architecture coming next year that well counter Nvidia's Pascal? Or is it just gonna be another rebrand?

 

[.vodka]

New architecture, Arctic Islands.
New node, 16/14nm FF.
HBM -> HBM2.

It should be enough to make actual 28nm GCN 1.x parts look like toys. So will nV's Pascal + the same 14/16nm fabrication node relative to their 28nm Maxwell cards.

So yeah, start saving to upgrade because next year is going to be a fun one for GPUs.

 

[tviceman]

2016-2018 should be a great time span for new, fast GPU's. Prices though.... ehhh....

 

[Mondozei]

2017 is likely to be a dud year. Normally we'd see a node shrink but nodes now take much longer than they used to. The only way out would be for NV to go for GP104 in 2016 and GP100 in 2017 for consumers(for enterprise and prosumers, GP100 would be available). And then go to 10 nm in 2018 with Volta.

If they go GP100 right ahead, there's literally nothing they can offer in 2017, so that's unlikely.

 

[poohbear]

Cool. 2016-2018 is 2 years though, I'm just talking 2016.

How would prices be different? I paid $400+ tax for my G1 gaming 970 when it was first released, prices haven't dropped that much on it. I usually look @ the 350-400 price bracket, so I imagine the pascal mid highend would be in that price range too.

 

[Silverforce11]

I'm expecting a shift upwards of around ~$50 on each SKU up to enthusiast grade.

Mid-range of $330-350 USD will move towards $400. High end of $650 will move towards $750.

Artic Island vs Pascal is the real DX12 battle since by then there should be many DX12 games out.

 

[poohbear]

Cool. 2016-2018 is 2 years though, I'm just talking 2016.

How would prices be different? I paid $400+ tax for my G1 gaming 970 when it was first released, prices haven't dropped that much on it. I usually look @ the 350-400 price bracket, so I imagine the pascal mid highend would be in that price range too.

Why would they suddenly do that for pascal? Because it's a new architecture? Or just for inflation?

 

[Silverforce11]

Why would they suddenly do that for pascal? Because it's a new architecture? Or just for inflation?

Because:

1. Currently 28nm is very mature, high yielding, is cheaper than 14/16nm ff process.
2. GDDR5 is a very established vram with great volume/yields. HBM2 is anything but not that.

These companies don't want to go backwards on margins, thus, the only way is up in price.

 

[VirtualLarry]

Because:

1. Currently 28nm is very mature, high yielding, is cheaper than 14/16nm ff process.
2. GDDR5 is a very established vram with great volume/yields. HBM2 is anything but not that.

These companies don't want to go backwards on margins, thus, the only way is up in price.

You forgot: Discrete GPU sales are falling. Thus, in order to support future R&D, prices on each individual unit must necessarily increase.

 

[Seba]

Unless the cause of falling sales is that the prices are to high.

 

[ShintaiDK]

Unless the cause of falling sales is that the prices are to high.

Unlikely as financial results show. Those lost are due to IGPs. Specially in mobile.

 

[Mopetar]

I think sales have been falling off simply because more and more people are holding off and waiting for the next generation of GPUs. The performance gain from refinements to architectures coupled with the massive density gain from going to 28 nm to 14/16 nm FF is going to be massive as we've been nearly 3 years without a node change from either AMD or NV.

I don't think the yields will be too bad to have a substantial impact on price as both Apple and Samsung have been manufacturing their SoCs using 14/16 nm nodes and TSMC/Samsung are likely getting yields to improve.

 

[Kenmitch]

Both camps will milk the node for all it's worth most likely.

Expecting about 15-20% performance gain in all tiers. Good stuff probably held back for 2nd's and 3rd's.

 

[MrTeal]

Both camps will milk the node for all it's worth most likely.

Expecting about 15-20% performance gain in all tiers. Good stuff probably held back for 2nd's and 3rd's.

I have a hard time believing that from either vendor. They likely won't go for big dies right off the bat and there will be continual process improvements on that node over time, but neither can afford to really hold back significantly. AMD's shown they are willing to throw down with nVidia and try to take the performance crown. They didn't quite get there with Fiji, but there's no reason to think they would try to slow roll 16nm to keep something in reserve when they desperately need major wins now just to remain viable. Depending on how Arctic Islands and Zen launch, there might not be a viable company to launch products in two years if they dick around too much.

nVidia has an easier time as for them even an effective tie is still a dominating position over AMD, but even they can't risk a piddly 20% performance gain. If GP104 was even just as powerful as the 980Ti (which is more like 25% faster than the 980), it's very likely Arctic Islands will beat them by a commanding margin. nVidia might have a smaller die and better cost structure if that's the case, but losing the halo would be a big blow to them. They haven't really been outclassed in the high end since AMD hit with the 9700 Pro and nVidia blew it with FX.

 

[Hugo Drax]

If AMD goes belly up, 2016-2017 will be interesting that is for sure. Not sure if AMD has more than 2 years left.

 

[tential]

If AMD goes belly up, 2016-2017 will be interesting that is for sure. Not sure if AMD has more than 2 years left.

Since you do some type of financial work, based on what measures do you believe amd doesn't have 2 more years left?

 

[Despoiler]

If AMD goes belly up, 2016-2017 will be interesting that is for sure. Not sure if AMD has more than 2 years left.

AMD has at least until 2020 before they run out of money.

 

[Headfoot]

15 whole posts until an "AMD is doomed in the next few years" post. You guys are slipping


OP: Rebrands make it MORE likely AMD has something in the kitty next, not less likely. They wouldn't do very minor refreshes if they weren't trying to hold off for something, of if they were actually going bankrupt which despite much noise to the contrary, doesn't appear to be imminent.

Arctic Islands has to come out before Pascal IMO or they will just get preempted and lose mindshare and thus market share again. Even if its slightly worse.

 

[Kenmitch]

I have a hard time believing that from either vendor. They likely won't go for big dies right off the bat and there will be continual process improvements on that node over time, but neither can afford to really hold back significantly. AMD's shown they are willing to throw down with nVidia and try to take the performance crown. They didn't quite get there with Fiji, but there's no reason to think they would try to slow roll 16nm to keep something in reserve when they desperately need major wins now just to remain viable. Depending on how Arctic Islands and Zen launch, there might not be a viable company to launch products in two years if they dick around too much.

nVidia has an easier time as for them even an effective tie is still a dominating position over AMD, but even they can't risk a piddly 20% performance gain. If GP104 was even just as powerful as the 980Ti (which is more like 25% faster than the 980), it's very likely Arctic Islands will beat them by a commanding margin. nVidia might have a smaller die and better cost structure if that's the case, but losing the halo would be a big blow to them. They haven't really been outclassed in the high end since AMD hit with the 9700 Pro and nVidia blew it with FX.

Time will tell as it could go either way.

Setting the bar too high tends to bring disappointment based on history. Trying to predict the future on history tends to be a crap shoot at best.

At launch both camps will be targeting the holdouts most likely. The enthusiasts will play the waiting game.

2nd guessing or possibly 3rd guessing the competition could lead to substantial performance gains from the get go I guess....Back to time will tale.