AMD publishes Polaris Architecture microsite

[CAT-THE-FIFTH]

Fiji has things like fine grained pre-emption that the GM200 lacks,so there are one or two things Fiji might have over it for things like VR.

[kalniel]

Fiji has things like fine grained pre-emption that the GM200 lacks,so there are one or two things Fiji might have over it for things like VR.

It's better, certainly, but still not enough power really. The Pro Duo on the other hand..

[shaithis]

Nvidia's Pascal chips have a lot of questions hanging over them and I've got a feeling the new chips might still be very DX11 focussed rehash of the failed 20nm architecture

There is nothing inherently wrong with nVidia dx12 implementation, so talking about their cards being 'DX11 focused' is very misleading. The problem is that most DX12 code out there is GCN optimised. Your looking at the problem from the completely wrong angle.

I doubt very much that AMD saw this one coming either but it may be their best chance to leverage market share before nVidia find some way to stop this proliferation (i.e. GameWorks DX12!) or DX12 fails on PC due to running slower on 60% of PCs.

[kalniel]

There is nothing inherently wrong with nVidia dx12 implementation, so talking about their cards being 'DX11 focused' is very misleading. The problem is that most DX12 code out there is GCN optimised. Your looking at the problem from the completely wrong angle.

I doubt very much that AMD saw this one coming either but it may be their best chance to leverage market share before nVidia find some way to stop this proliferation (i.e. GameWorks DX12!) or DX12 fails on PC due to running slower on 60% of PCs.

My fear is that leveraging DX12 is going to be too expensive for many devs to programme close to the metal, leaving it the province of only rich studios or those with a huge amount of on-going support from a vendor. But it's hard to say. The Ashes devs said something like it only took a weekend to implement async, but I get the feeling that game engine started off focused towards GCN hardware in the first place. Whether the console wins will mean more devs do the same I'm not sure - certainly for ports it would suggest it may be a default position until otherwise persuaded by the greens.

[shaithis]

My fear is that leveraging DX12 is going to be too expensive for many devs to programme close to the metal, leaving it the province of only rich studios or those with a huge amount of on-going support from a vendor.

That's exactly what I am starting to expect now but as more games are released in DX12 on XB1 the devs (and lets face it, PC ports have been terrible for years, the devs aren't allowed to spend the necessary time on a release that sells 10 times less than the platform it's being ported from) will just use their hugely-GCN-optimised-code. If recent AAA games like Batman:AK and The Division can't be bothered to support SLI/CFX on PC, what chance have we that they will re-write a large amount of code to take into account all the different GPU architectures that the PC has?

I am starting to think that the best way forward is for Windows DX12 titles to sell terribly and devs go back to DX11 for PC.......but I fear that they won't bother as once they have DX12 XB1 code, the work to port that to DX11 is probably going to be larger than optimising their DX12 code for different GPU uarchs.


We all thought the current-gen consoles and DX12 were going to be great, so far all they have done is fractured the PC market, proliferated terrible ports and thrown even bigger doubts over the performance of future PC titles

[DanceswithUnix]

My fear is that leveraging DX12 is going to be too expensive for many devs to programme close to the metal, leaving it the province of only rich studios or those with a huge amount of on-going support from a vendor. But it's hard to say.

The port of The Talos Principle to Vulkan seems to have been the effort of their one graphics engine guy and sounds like it was a bit of a sideline. That was for a game engine that still is DX11 optimised. Over time I'm sure the skills for the new close to the metal APIs will become more commonplace, wrinkles will get ironed out, and DX12/Vulkan will become the new normal. I think TTP is a good example though, the Vulkan mode is there to try if you want, but after years of DX11 work the game is faster in DirectX and will no doubt be faster for some time because potential for performance is not the same as where we are now.

Mind you, I played quite a lot of that game in Linux so I could avoid having to reboot into Windows despite the OpenGL performance being way worse. The performance only has to hit "good enough".

[jigger]

There is nothing inherently wrong with nVidia dx12 implementation, so talking about their cards being 'DX11 focused' is very misleading. The problem is that most DX12 code out there is GCN optimised. Your looking at the problem from the completely wrong angle.

I doubt very much that AMD saw this one coming either but it may be their best chance to leverage market share before nVidia find some way to stop this proliferation (i.e. GameWorks DX12!) or DX12 fails on PC due to running slower on 60% of PCs.

It's not misleading at all. The problem is exactly what I pointed out. Nvidia have many questions hanging over them and the time frame seems way to short.

[shaithis]

It's not misleading at all. The problem is exactly what I pointed out. Nvidia have many questions hanging over them and the time frame seems way to short.

Go do some research.

[jigger]

Go do some research.

You too.

[DanceswithUnix]

I hope Polaris is better value than this article suggests, 390 performance for a bit less than the cost of a 390.

http://www.fudzilla.com/news/graphic...90-performance

[kalniel]

I hope Polaris is better value than this article suggests, 390 performance for a bit less than the cost of a 390.

http://www.fudzilla.com/news/graphic...90-performance

Early adopter price vs card that's been on market for some time. Polaris should be WAY more efficient/cooler too.

If you really want some confusion, the 390 is likely to drop in price when Polaris releases, while stores clear inventory. But I'd still rather have a Polaris 10.

[CAT-THE-FIFTH]

£250 for a Polaris 10,which is around R9 390 level is a bit of a fail,especially if Nvidia has faster options out there.

AMD said they were trying to improve value for money in a BIG WAY. A few dollars cheaper with slightly better performance is not going to cut it. Sure,it will be more efficient than an R9 390 and have much shorter cards,but the GTX970 already has done that.

It will make AMD look like they are playing catch up to Maxwell.

They should be pricing this more closer to £200 to encourage GTX960,R9 380 and R9 380X owners to upgrade.

AMD,does not seem to repeatedly get that Nvidia has stronger brand marketing that they do not have,and you can see what happened when they launched the Fury cards - only once the Fury and Fury Nano started dropping in price did people start considering them more seriously.

It leaves them open to the GTX1060/GTX1060TI which Nvidia is also launching,a 192 bit GP104 derivative.

They are being outsold 4 to 1 now,and even though they had improvements in sales in the last 6 months,they really need another HD3870 or HD4870 now.

Not a card which is barely better value for money.

The old ATI understood things better - it is why ATI did not go down to 20% of cards sold even with the HD3000 series when they could not beat the 4th fastest Nvidia card.

[DanceswithUnix]

Early adopter price vs card that's been on market for some time. Polaris should be WAY more efficient/cooler too.

If you really want some confusion, the 390 is likely to drop in price when Polaris releases, while stores clear inventory. But I'd still rather have a Polaris 10.

Well yes Polaris will be better in other ways, but in the really key metric of "how many frames can it deliver per second for my money" it has to be better or people will point and laugh.

Meanwhile, Charlie says the lack of Pascal demos is down to packaging problems again: http://semiaccurate.com/2016/05/03/t...00-pascal-gpu/

Kind of odd that Charlie has paywalled off that article, he usually likes to put the boot in as publicly as possible when it comes to Nvidia.

[CAT-THE-FIFTH]

Also,this is significant - it has been over a decade since Nvidia has beaten ATI or AMD or launched a new GPU on a new node around the same time as them.

Nvidia is already making an event of them sending out review samples.

FFS,AMD at least if you have some OEM wins with Polaris officially announce them,even if you are trying to size up the competition first!!

[jigger]

£250 for a Polaris 10,which is around R9 390 level is a bit of a fail,especially if Nvidia has faster options out there.

AMD said they were trying to improve value for money in a BIG WAY. A few dollars cheaper with slightly better performance is not going to cut it. Sure,it will be more efficient than an R9 390 and have much shorter cards,but the GTX970 already has done that.

It will make AMD look like they are playing catch up to Maxwell.

They should be pricing this more closer to £200 to encourage GTX960,R9 380 and R9 380X owners to upgrade.

AMD,does not seem to repeatedly get that Nvidia has stronger brand marketing that they do not have,and you can see what happened when they launched the Fury cards - only once the Fury and Fury Nano started dropping in price did people start considering them more seriously.

It leaves them open to the GTX1060/GTX1060TI which Nvidia is also launching,a 192 bit GP104 derivative.

They are being outsold 4 to 1 now,and even though they had improvements in sales in the last 6 months,they really need another HD3870 or HD4870 now.

Not a card which is barely better value for money.

The old ATI understood things better - it is why ATI did not go down to 20% of cards sold even with the HD3000 series when they could not beat the 4th fastest Nvidia card.

I suppose it depends on what kind of wattage range AMD plan to have and how power efficient Nvidia are. If AMD have a 100 watt card with R9 390 performance for £250 and it retains the DX12/Vulkan performance lead then it would be a solid card, but it will do very little push gaming forward or open high end performance to that much of a bigger market. I think OEM's would love it though.

[CAT-THE-FIFTH]

http://www.guru3d.com/news-story/amd...o-r9-390x.html

Hmm,supposedly the R9 480 non-X is around R9 390X level performance??

Edit!!

Rehashed,earlier leak.Meh.

I suppose it depends on what kind of wattage range AMD plan to have and how power efficient Nvidia are. If AMD have a 100 watt card with R9 390 performance for £250 and it retains the DX12/Vulkan performance lead then it would be a solid card, but it will do very little push gaming forward or open high end performance to that much of a bigger market. I think OEM's would love it though.

I think Polaris is very much an OEM focussed product so hopefully there is good traction for AMD there!!