NVIDIA GeForce GTX 760/960/1060 / RTX 2060 Linux Gaming & Compute Performance

[phoronix]

Phoronix: NVIDIA GeForce GTX 760/960/1060 / RTX 2060 Linux Gaming & Compute Performance

The NVIDIA GeForce RTX 2060 is shipping today as the most affordable Turing GPU option to date at $349 USD. Last week we posted our initial GeForce RTX 2060 Linux review and followed-up with more 1080p and 1440p Linux gaming benchmarks after having more time with the card. In this article is a side-by-side performance comparison of the GeForce RTX 2060 up against the GTX 1060 Pascal, GTX 960 Maxwell, and GTX 760 Kepler graphics cards. Not only are we looking at the raw OpenGL, Vulkan, and OpenCL/CUDA compute performance between these four generations, but also the power consumption and performance-per-Watt.

http://www.phoronix.com/vr.php?view=27413

[tildearrow]

Typos:

when going from the GTX 760 to RTX 2070 was a 4x

Warhammer II with the Vulkan-pwoered Linux port was able to run at 60 FPS

here is a look at the haromic mean of all the frame-rate gaming benchmarks

[birdie]

For Pascal owners the RTX 3060 on 7nm will be a GPU to buy, the RTX 2060 not so much. However if you come from earlier generations (like Maxwell) the RTX 2060 is an amazing card, albeit a little bit too expensive. That's all granted NVIDIA will continue to support the Linux kernel and graphics stack which is not a given.

[cybertraveler]

Really nice benchmarks Michael thanks. Also, thanks again for including Counter Strike.

One question for Michael or anyone really. Are the frame-time graphs mostly just useful for spotting upward spikes? I ask because obviously, the normal FPS bar charts give us the average (and often the min and max).

Upward spikes on the frame-time graph could signify micro stutters right?

[xorbe]

For Pascal owners the RTX 3060 on 7nm will be a GPU to buy, the RTX 2060 not so much. However if you come for earlier generations (like Maxwell) the RTX 2060 is an amazing card, albeit a little bit too expensive. That's all granted NVIDIA will continue to support the Linux kernel and graphics stack which is not a given.

Yeah but:
960 - $199
1060 - $249 (+$50)
2060 - $349 (+$100)
3060 - $499 (+$150)

[birdie]

Yeah but:
960 - $199
1060 - $249 (+$50)
2060 - $349 (+$100)
3060 - $499 (+$150)

The price of the RTX 3060 is not known at the moment but I believe it'll be below $350 (the 7nm die should be quite smaller thus cheaper to produce and also the GDDR6 memory must become cheaper in the future when technology matures and more IC manufactures start producing it).

[coder]

Yeah but:
960 - $199
1060 - $249 (+$50)
2060 - $349 (+$100)
3060 - $499 (+$150)

I see what you did, there. But we have to wait and see how Nvidia reacts to AMD's Navi. Presumably, between that and GTX 1060 stock drying up, there could be some price cuts of the RTX 2060.

Also, Intel's Xe might be in the game, by the time RTX 3060 launches, or shortly thereafter.

[coder]

Anyway, these results are almost too predictable. With AMD GPUs, there are usually at least a couple surprises.

[Brisse]

Upward spikes on the frame-time graph could signify micro stutters right?

Exactly that. The physical quantity is the same, but expressed in a different measurement. Mathematically speaking, frame time is just the inverse of frame rate, but when talking about frame rate one usually talks about an average over time, while frame time usually indicates an instantaneous value, not an average over time. Average over time can obscure micro stuttering and other frame timing issues.

[Jabberwocky]

Exactly that. The physical quantity is the same, but expressed in a different measurement. Mathematically speaking, frame time is just the inverse of frame rate, but when talking about frame rate one usually talks about an average over time, while frame time usually indicates an instantaneous value, not an average over time. Average over time can obscure micro stuttering and other frame timing issues.

+1

I find 1% Low and 0.1% Low average FPS very useful for benchmarks, especially when compared to average FPS. Gamer's Nexus has used this measurement on their graphs for many years. It is useful to have instantaneous values like frame-time, but can be misleading when the entire event-loop is not taken into consideration (something that cannot be done by looking at a graph).