The Radeon RX 590 Is Finally Running Strong On Linux

[M@GOid]

There is some big especulation on AMD's 7nm products next year. If this table bellow is true, my body is ready for that 3600G APU on a very compact ITX case:

[ms178]

Not that AMD had any trouble selling CPUs the last couple of months.

[dungeon]

Cool that is working now correctly

At least if it is faster than RX 580 with that black magic avfs cks-off voltages

[dungeon]

All reviews I can find about this GPU basically boil down to "there is zero reason to buy this, get a 580 instead."

AMD marketing claim up to 27% performance over initial referent RX 480 and since this RX 590 model is also a bit OC card, one could count up to 30% and that is not nothing

[smitty3268]

What's the difference with Padoka PPA and Oibaf PPA?

Mostly the LLVM version. Padoka is more bleeding edge, Oibaf is more stable.

[illwieckz]

Is this card good for OpenCL image processing with GEGL or Darktable using the open-source AMDGPU stack? (What does clinfo say about image support?)

Mesa OpenCL (clover) does not have image support yet for AMD, but ROCm has. The good thing is that you can install both. rocm is buggy with luxrender but clover is blazing fast and not buggy, rocm is having image support hence it works with darktable while clover does not. If you install both because you need clover for other tasks, darktable will just ignore clover and picks rocm since it's the only one implementing requirements. Not that building rocm may be boring because the provided packages want to build modules (and fail to) you may not need if your distro is recent enough.

The other solution is the opencl stuff shipped within amdgpu pro but without hacks they require old kernels from Ubuntu LTS, this piece of software covers all the usages. This is proprietary but it may be ok for some people to use it while waiting for free opencl implementation getting better.

[TemplarGR]

There is some big especulation on AMD's 7nm products next year. If this table bellow is true, my body is ready for that 3600G APU on a very compact ITX case:

If that table is true, i really hate AMD now. They know that most people don't need plenty of cores, so they artificially lower the clocks of 6 core "entry level" skus in order to make higher end skus more attractive. This is EVIL and anti-consumer. Even Intel when it was all powerful, still made the dual cores to run at high clock speeds...

[mitch074]

I'm pretty sure this is AMD's attempt at phasing out 14nm GPUs - this 590 replaces the 14nm 580, and if successful, a 12nm 580 2048 replaces the 570 as a recycling path for faulty 590s. The 570 simply goes away, or if demand remains, we get a "2nm 570 1792" - or a 12nm 570 that can't handle overclocks.
In the meantime, I'll keep rocking this continued support for my reference RX480 8Gb, and I'll try to overclock it. But as of now, even with a new cooler, I can't push it higher than a 2% frequency hike.

[dungeon]

If that table is true, i really hate AMD now. They know that most people don't need plenty of cores, so they artificially lower the clocks of 6 core "entry level" skus in order to make higher end skus more attractive. This is EVIL and anti-consumer. Even Intel when it was all powerful, still made the dual cores to run at high clock speeds...

That is mainstream table, from $99 to $499. 3/5/7/9

Bellow that are Pentium/Celeron and Athlon/Sempron

[M@yeulC]

If that table is true, i really hate AMD now. They know that most people don't need plenty of cores, so they artificially lower the clocks of 6 core "entry level" skus in order to make higher end skus more attractive. This is EVIL and anti-consumer. Even Intel when it was all powerful, still made the dual cores to run at high clock speeds...

Keep in mind that they have to play the silicon lottery, and that their infinity fabric interposer is especially adapted to hardware binning, unlike Intel's (former?) monolithic approach. This reads to me as if they packaged the slower chips together at a lower cost, and the faster ones at a higher cost.

Unlike Intel, AMD is not known for artificially disabling some of their chips functionality to segment the market. I love how the performance steadily increases with the price on these new chips (perf/price is nonlinear, and last I checked, the more expensive the CPU, the more bang for your buck), and how you get the same number of PCIe lanes, and features such as ECC, virtualisation, etc...
I feel like providing a low core-count, high-frequency CPU wouldn't be so great from a consumer perspective, as they would need to know their workload's details in advance... And that would almost seem wasteful, especially as new workloads can exploit more cores better and better...