AMD Kaveri Gets A Big Performance Boost With Mesa 18.2 & AMDGPU DRM

[Kemosabe]

The software industry is lucky that you are not a manager of software development.

Funny, because I am ;-)
And yes, this person would face serious consequences in my company.

[jo-erlend]

This is really good news. I still love my passively cooled A10-7850K desktop and now I guess I'll get to love it a little more.

[ruthan]

Hmm, even new machines are disaster.. I never had such problems with proprietary Nvidia drivers.. Look at this Manjaro thread even new Ryzen APU are still not working even with experimental kernel how long after launch 3 months?
https://forum.manjaro.org/t/any-supp...linux/38079/41

Im Mint user (only 4.15.20 kernel) and have one Ryzen 2200G, so i have to use Windows 10 (yeah nice thing ignore even Windows 7/8)..

[fuzz]

Hmm, even new machines are disaster.. I never had such problems with proprietary Nvidia drivers.. Look at this Manjaro thread even new Ryzen APU are still not working even with experimental kernel how long after launch 3 months?
https://forum.manjaro.org/t/any-supp...linux/38079/41

Im Mint user (only 4.15.20 kernel) and have one Ryzen 2200G, so i have to use Windows 10 (yeah nice thing ignore even Windows 7/8)..

Proprietary Nvidia drivers don't bother working with the community to include open source code in the kernel and userland.

But I guess you can go compromise morals for some capitalist agenda.

EDIT: I'd say your mentality would work with BSD but good luck running new hardware there.

[bridgman]

Hmm, even new machines are disaster.. I never had such problems with proprietary Nvidia drivers.. Look at this Manjaro thread even new Ryzen APU are still not working even with experimental kernel how long after launch 3 months?
https://forum.manjaro.org/t/any-supp...linux/38079/41.

NVidia does not make APUs (other than Tegra, I guess) and so doesn't need to resolve SBIOS and chipset issues, which seem to have been the biggest factor here. AFAICS some of the issues were driver-related but getting the latest SBIOS / AGESA into the motherboards seems to have been a key part of the solution as well.

If you read through the thread the pattern seems to be:

- latest kernels weren't available in distro builds
- once kernels with HW support were being used stability issues remained
- once latest SBIOSes were released & adopted most stability problems went away
- a couple of issues remained and could then be isolated and addressed with GPU driver fixes

If you want to draw conclusions about graphics drivers it's probably best to compare apples to apples, or in this case dGPUs to dGPUs.

I'm not saying the other issues were not real problems for our customers, just that trying to draw conclusions about GPU drivers by lumping in all of the CPU/chipset/SBIOS issues in for one vendor but not the other isn't likely to give you a valid result.

EDIT - this is really wierd... whenever I try to edit this post the content gets back-dated to a much older version.

[artivision]

NVidia does not make APUs (other than Tegra, I guess) and so don't need to deal with SBIOS and chipset issues, which seem to have been the biggest factor here. AFAICS some of the issues were driver-related but getting the latest SBIOS / AGESA into the motherboards seems to have been a key part of the solution as well.

If you want to draw conclusions about graphics drivers it's probably best to compare apples to apples, or in this case dGPUs to dGPUs.

I wanted to ask how we call a company that cannot keep up with their own word. Do you remember "the north-bridge will be inside the apu/cpu" or "there will be no special platform for over-clocking". Today we cannot even find an A320 Mini ITX.

[bridgman]

I wanted to ask how we call a company that cannot keep up with their own word. Do you remember "the north-bridge will be inside the apu/cpu" or "there will be no special platform for over-clocking". Today we cannot even find an A320 Mini ITX.

Not sure I understand the relationship between your sentences.

The northbridge (DRAM controller & high speed bus interface) has been inside the APU/CPU for a decade or more. The "chipset" you see today is the southbridge, albeit with narrow PCIE buses partially or completely replacing older PCI buses.

What is the connection between that and availability of specific chipset / mobo form-factor combinations ?

I don't remember hearing any statements about "there will be no special platform for over-clocking" although I believe that is pretty much the case already - most of the chipsets have similar support for overclocking although one (A320) does not while three do. Strictly speaking 2 chipsets do not and 4 do, but as far as I can see the 2 smallest chipsets (X300/A300) are not being used much yet.

Can you point me to whatever statement you think we are not keeping up with ? Is your concern just that you are not able to buy a mini-ITX board that does not support overclocking ?

[artivision]

Not sure I understand the relationship between your sentences.

The northbridge (DRAM controller & high speed bus interface) has been inside the APU/CPU for a decade or more. The "chipset" you see today is the southbridge, albeit with narrow PCIE buses partially or completely replacing older PCI buses.

What is the connection between that and availability of specific chipset / mobo form-factor combinations ?

I don't remember hearing any statements about "there will be no special platform for over-clocking" although I believe that is pretty much the case already - most of the chipsets have similar support for overclocking although one (A320) does not while three do. Strictly speaking 2 chipsets do not and 4 do, but as far as I can see the 2 smallest chipsets (X300/A300) are not being used much yet.

Can you point me to whatever statement you think we are not keeping up with ? Is your concern just that you are not able to buy a mini-ITX board that does not support overclocking ?

The problem is that B350 Mini ITX costs 100+ like Intel's Zxxx Mini ITX, wile other Intel Mini ITX like H110 cost only 50. So for my pc its ok but when i need 10 of them guess what i will buy.

[nanonyme]

So instead of fixing radeon, they crippled the performance ... also a solution ...

I suspect the plan is just to use the brokenness as a reason to migrate from radeon to amdgpu instead where you will get the full performance.

[bridgman]

The problem is that B350 Mini ITX costs 100+ like Intel's Zxxx Mini ITX, wile other Intel Mini ITX like H110 cost only 50. So for my pc its ok but when i need 10 of them guess what i will buy.

Ah, OK. Isn't some of the low price from the H110 mobo a function of it being last-generation though ? You can get some low-priced last-gen AMD mini-ITX boards as well. My understanding was that H310 was the current low-end Intel chipset and those seem to be equally rare. I do see what you mean about the pricing differences though.

I suspect another factor is that even with higher end chipsets the (CPU+mobo) price ends up being similar for comparable performance. Usually you don't see vendors putting much emphasis on lower-end products until price pressure starts to erode sales of the higher end parts.