Ahead Of Radeon RX Vega, AMDGPU+RadeonSI Is Offering The Most Competitive Performance Yet Against NVIDIA On Linux

[dungeon]

You know what is funny? Apparently the Nvidia driver did not regret, but have a nice performance bump since the last comparative Michael did:

Then: http://www.phoronix.com/scan.php?pag...si-1june&num=4

Now: http://www.phoronix.com/scan.php?pag...onsi-384&num=4

These systems are not the same one, one has i7 7700K and this is with 7740K... and If you bench on AMD Ryzen you can get entirely something else fps wise, etc...

So as always "grain of salt" is always there, put the same card on different system you will get different CPU boundwared results

[NihilMomentum]

These systems are not the same one, one has i7 7700K and this is with 7740K... and If you bench on AMD Ryzen you can get entirely something else fps wise, etc...

So as always "grain of salt" is always there, put the same card on different system you will get different CPU boundwared results

So that increase for Nvidia could be just the use of a stronger CPU, since the 7740k is *just* an slightly overclock 7700k with some cache changes...

[Shevchen]

The difference between the Nvidia results are way too big to just put the blame on the CPU. Its not like you are comparing a 3 GHz I3 vs an 4+ GHz overclocked I7. (which could explain the difference in a CPU-bound scenario) Something else is going on and it doesn't look good for team green.

Aside from that, the results from team red are amazing. Some games/synths (ad victoriam!) still need some work - but overall there is no holding back point anymore, if you want to buy an AMD card for gaming on Linux.

Also for the ones trying to compare it versus Windows: Have in mind most games here are ports. They have naturally a lower performance than native (DX) games and thus the comparison is a little flawed.

A nice bench would be comparing native to native Software - but I can't remember such a thing out of the head. A game that has the same OpenGL core on Windows and Linux and is halfway "up-to-date" using the latest hardware-arch capabilities from both AMD and Nvidia?

One could look at Vulkan, but RADV is still not finished and the comparison would be kinda unfair then. In any case, my bets are on RADV getting competitive so I can enjoy the Vega 64 on Linux playing Star Citizen, once their Vulkan renderer is in place. (which may take another year *sigh*)

[tomtomme]

Who would have thought we will see this LOL
AMD killing it in openGL and struggling with Vulkan

this is because Marek and his team are the most awesome programers a graphics driver can have.
fury @ 1070 level in most games is sooo out of line. if this would have been a Fury X - ok - but this?!? Thank you Marek and the whole Team!

[Wielkie G]

So that increase for Nvidia could be just the use of a stronger CPU, since the 7740k is *just* an slightly overclock 7700k with some cache changes...

No cache changes since it uses the same die. Just an overclock.

[Azpegath]

Wow, amazing results from AMD! Numbers like these was a pipe-dream a couple of years ago...

But there seems to be something weird going on with DOTA, what could be the issue there? All the cards seems to be vastly under-performing, considering the results in the other tests. marek, any thoughts or ideas on this? Is the game just badly tuned for AMD cards?

[marek]

Wow, amazing results from AMD! Numbers like these was a pipe-dream a couple of years ago...

But there seems to be something weird going on with DOTA, what could be the issue there? All the cards seems to be vastly under-performing, considering the results in the other tests. marek, any thoughts or ideas on this? Is the game just badly tuned for AMD cards?

I think there is a performance regression in Mesa affecting DOTA 2.

[Apopas]

It would be nice to have a second comparison between lower performance cards, ie rx 550, rx 460, rx 560 and gt 1030, 1050, 1050 ti, with these new drivers.

[mphuZ]

AMDGPU is the DRM kernel driver, RadeonSI is the user-space OpenGL driver

Always there is confusion.
And the confusion arises because of the initials SI - is it not possible to rename the driver to something "universal" ? E.g. "RadeonVideo" / "RadeonGL" ?

[schmidtbag]

All of this and yet Marek is still discovering performance tweaks, as has been posted earlier today. This is exciting times. At this rate, radeonsi might out-pace the Windows drivers by the end of this year.

Very cool to see the Fury exceed the GTX 1080 sometimes. If the non-X Fury can do that great, I'm intrigued to see what Vega can accomplish. The weird thing is (to my recollection) even the Fury X is consistently worse than a 1080 in Windwos.