New AMD Radeon "Polaris" GPU Details Revealed

[phoronix]

Phoronix: New AMD Radeon "Polaris" GPU Details Revealed

Just ahead of CES, AMD/RTG is allowing more details on their next-generation Polaris GPU architecture to be revealed...

http://www.phoronix.com/scan.php?page=news_item&px=Radeon-Polaris-Details

[Nille_kungen]

Waiting for Kano to ask the usual question

[zanny]

So announce this month, release next month, reasonable performance in the OSS driver in 2018, reasonable stability in Catalyst never.

AMD, hire another half dozen developers for Gallium and go punch Intel in the nads so they start using it and I'll buy a bakers dozen of new GPUs from you.

[theriddick]

Sounds interesting, really hope AMD release good drivers once Vulkan comes out (before Polaris).

[kaprikawn]

Wow, an actual improvement. It's been a while.

However, you know that Nvidia at 14nm will outperform AMD at 14nm. AMD needs the process reduction to improve performance per watt, Nvidia will just pull ahead when their 14nm appears that has the same technology which served Maxwell so well. Both technologies in unison will no doubt outdo AMD.

I will buy AMD for Linux when there's a good reason because of open source, but my Windows machine will be staying with the green team.

[bug77]

Sounds interesting

Sounds like catching up to nvidia to me (perf/watt improvements, memory compression - see Maxwell). And it's more than half a year away, I'm not going to read every presentation they will release between now and then. Hard launch or gtfo (and that applies to nvidia, too).

[Ansla]

Still no VP9 support? That's quite disappointing.

[boffo]

Will there be day one support in kernel & mesa or is it a big gap and we will need to wait a long time?

[sabun]

- Polaris GPUs will be on a 14nm FinFET process.
- DisplayPort 1.3 for support to drive 4K @ 120Hz.
- HDMI 2.0a support will finally be found for Radeon GPUs.
- 4K HEVC/H.265 encode/decode support.
- Big improvements in performance-per-Watt.
- New hardware features like a primitive discard accelerator, instruction pre-fetch, memory compression, and hardware scheduler.

The important question to be asked here is how many of these features and functionality will actually reach Linux users within 2016?

They can list all the best looking and highest buzzing words for new features, but if none of it reaches us in a timely and usable manner then there is no point to get excited.

I really doubt the performance-per-Watt will improve for us Linux users at all. My hopes are for marginal gains at least, and hopefully not regressions. We already consume more electricity and get less performance in the Linux driver compared to the Windows driver.

[theriddick]

Well AMD GPU's have proven to be better in DX12 and probably Vulkan API. So I don't think its a issue with NVIDIA being better, the main issue is if AMD can sort their driver issues out in time, I really don't think AMD can afford another BAD PR run with terrible driver issues, in Windows and Linux.