Intel Haswell HD Graphics 4600 vs. AMD Radeon Graphics On Linux

[brosis]

So it looks like for an open source solution, the Ivy "R" CPUs might be the best choice overall, with the 5200.

SB-600 shader optimizer not used. For AMD (Michael uses Intel CPU) - ondemand governor must be tuned. Kernel must be patched for AMD APUs on desktop to allow walking past TDP clocks. And so on.
Radeon driver currently has a lot of options not set from default.
And ofc dynamic power management is absent for AMD. So, yes, I would dump up the opensource performance a bit(+20%?) but due to PM, going with Intel in midrange is a wiser choice.

Great article! Its nice to see how the OSS radeon gfx are fast approaching speed parity with the closed-source solution. One thing that would be interesting is a compatibility test between OSS & Closed Drivers on Radeon. On Ubuntu 13.04, I tried the OSS drivers on my 4670 card, and was very dismayed that many games were just not compatible with it (ie. would crash or just not load), especially many of the HiB games as well as many Wine games.

Oh, Ugh-buntu. You would need to build MESA and radeon from GIT, build kernel at least 3.10, play with options a bit, don't forget to install S3TC and then your 4670 with OSS will have same performance as catalyst. Still... if you have chance, better go with Intel 4600 CPU or higher. But again, a lot will be needed to be built manually in any case due to urge for the lastest code, Intel will need less fiddling and will have full automatic "power gear transmission" . Or wait 1 year+ till Ubuntu catches up. But then, there gonna be Mir and a lot of breakage. So, do something good today - change the distro

[brosis]

AMD, where is your dyn PM ?

[artivision]

You just prove that AMD steeling FPS. An HD4870(1200Gflops) card has the same FPS with an HD6570(600Gflops) card. And an HD7970(3.8Tflops@32bit) card has the same FPS with an GTX680(3.2Tflops@64bit or 6.4Tflops@32bit). And the thing is that all that happened in just 1-3 Catalyst releases.

[smitty3268]

You just prove that AMD steeling FPS. An HD4870(1200Gflops) card has the same FPS with an HD6570(600Gflops) card. And an HD7970(3.8Tflops@32bit) card has the same FPS with an GTX680(3.2Tflops@64bit or 6.4Tflops@32bit). And the thing is that all that happened in just 1-3 Catalyst releases.

The games michael tests don't usually require any gflops at all, they are largely streaming tests to see how fast you can transfer data from the cpu to the gpu - at least for the faster gpu hardware.

[smitty3268]

Michael, can i ask why you didn't just set the proper environment vars for Unigine?

Since you clearly know what they are? It's a fair comparison to Intel, since those are just the same thing the intel driver does by default when it detects a Unigine app running.

[Nille]

You just prove that AMD steeling FPS.

No, the Game Engines are only old.

[twriter]

If only AMD would send out the hardware...

Michael, I tried to PM you about this but you've exceeded your storage quota! I'll e-mail you directly instead.

Tim

[Michael]

Michael, I tried to PM you about this but you've exceeded your storage quota! I'll e-mail you directly instead.

Tim

Thanks. You can email me at Michael at Phoronix.com. For some reason my forum account always says full even though its empty.

[remm]

SB-600 shader optimizer not used. For AMD (Michael uses Intel CPU) - ondemand governor must be tuned. Kernel must be patched for AMD APUs on desktop to allow walking past TDP clocks. And so on.
Radeon driver currently has a lot of options not set from default.
And ofc dynamic power management is absent for AMD. So, yes, I would dump up the opensource performance a bit(+20%?) but due to PM, going with Intel in midrange is a wiser choice.

Cool ... So I own a HD7870, nobody with an ounce of sanity should buy AMD, period.

[pingufunkybeat]

Cool ... So I own a HD7870, nobody with an ounce of sanity should buy AMD, period.

Nonsense.

People have different requirements and expectations and the free Radeon drivers have come a very long way and are nowadays seriously approaching feature-completeness.