The OpenGL Speed & Perf-Per-Watt From The Radeon HD 2000/3000 Series Through The R9 Fury

[phoronix]

Phoronix: The OpenGL Speed & Perf-Per-Watt From The Radeon HD 2000/3000 Series Through The R9 Fury

What's the best way to beat the winter blues? Benchmarking, of course! For starting off our 2016 of graphics card benchmarking under Linux, I've been working on a large round-up of re-testing AMD Radeon graphics cards from the HD 2900XT (R600) graphics card through the latest R9 Fury (Fiji) graphics card while running Ubuntu and using the very latest open-source graphics driver stack. Here's an interesting look at how the OpenGL graphics performance has evolved on the AMD side over the past decade while also looking at the performance-per-Watt.

http://www.phoronix.com/vr.php?view=22694

[andrei_me]

Why Fury runs great on GpuTest's Furmark but not on games?

[Michael]

Will also be running a similar NVIDIA comparison but with using their current/legacy drivers. Should be done next week. HOping to go back to the GeForce 6 series.

[bridgman]

Why Fury runs great on GpuTest's Furmark but not on games?

First guess would be that the GPU utilization isn't quite high enough for dpm to spin the clocks up.

IIRC there is a sysfs option to force the card to use high clocks for testing, will try to get it to Michael.

[whitecat]

Where are the R600 numbers?

[Michael]

Where are the R600 numbers?

Did you read the article? The HD 2900XT regressed to an unworkable state.

[ruthan]

I have note to benchmark design, it doesnt make a sense use for FPS games such big resolution and quality setting where even best card has only 30 FPS, i would be better to use less quality settings, were best card has around 60 FPS, there would be probably so big gaps between best and other cards, because these are not designed for 2560x1600, especially in Linux usage.
I would like to see also Windows results for at least one card as etalon (standard of quality) .

[Veto]

Great article!
Nice to see how the 285 is shaping up nicely. The 370 is performing impressively for its price and seems to be the best optimized card of the bunch. Counter-strike performance seems great for most cards - I wonder how it compares to Windows?

Also impressive performance leap for Fury in Furmark - although there is clearly some work left to unleash its potential. Also GCN still need some optimization for TF2 and Metro especially considering their popularity.

[stevenc]

So I'm quite happy with my HD6950. If AMD hopes I'll buy a new one, they really better help the open source driver work properly on newer cards.

[smitty3268]

Why Fury runs great on GpuTest's Furmark but not on games?

Just guessing, but I bet the core shader clock is fine and the memory clocks are slow.

Furmark was designed to run very differently from games, and *i think* it does that by hammering very hard on the shader hardware, without doing anything else like memory accesses. And i think ancient stuff like OpenArena is heavily limited by memory speed, at least on good graphics drivers.