NVIDIA GeForce GTX 1080 On Linux: OpenGL, OpenCL, Vulkan Performance

[deppman]

Completely nothing to do with efficiency. Async_compute means to do compute upon and during shading (aka complex shaders). Standard compute means to do compute after shading on the entire shader. This is not a hardware nor an Api future, it's developers thinking. Nvidia hardware runs this but doesn't accelerate it (the gain in beauty of the field compensates for the loss anyway). With Amd there is no loss at all. Don't find it strange if a future FX (a modern Motion Blur for example) cannot even run on an Nvidia Gpu at all and give something like 40ms latency. Amd has this magnificent blend threading but also good connection-compression of information. In Ashes TitanX is just as fast as a 390x and TitanX cheats with snow and other FX.

Of course this has everything to do with efficiency. Pascal provides the ability to process compute and shader simultaneously, as did Maxwell. However, pascal has added a) dynamic load balancing and b) pixel-level preemption instead of preemption at draw call boundries. So now compute and shading can conclude with much higher resource utilization in a much shorter time period (a) and asynchronous (e.g "real-time") tasks can be completed with much lower latency (b).

Check out the attached slides. It seems to me that NV delivers much of what AMD promises. Low latency? Got it. Compute + shading at delivered concurrently? Yup. Much better resource allocation? Got it.

If AMD's version of " async_compute (tm)" is so substantially superior and important then answer me this: when is the last time it made anything anywhere substatially better, much less on Linux? Now remove "AOTS on Windows 10". Got anything left? Right.

See the slides below.
http://cdn.wccftech.com/wp-content/u...-Balancing.jpg
http://cdn.wccftech.com/wp-content/u...on-635x307.jpg

[Asariati]

Phoronix: NVIDIA GeForce GTX 1080 On Linux: OpenGL, OpenCL, Vulkan Performance

$699 USD is a lot to spend on a graphics card, but damn she is a beauty. Last month NVIDIA launched the GeForce GTX 1080 as the current top-end Pascal card and looked great under Windows while now finally having my hands on the card the past few days I've been putting it through its paces under Ubuntu Linux with the major open APIs of OpenGL, OpenCL, Vulkan, and VDPAU. Not only is the raw performance of the GeForce GTX 1080 on Linux fantastic, but the performance-per-Watt improvements made my jaw drop more than a few times. Here are my initial Linux results of the Gigabyte GeForce GTX 1080 Founder's Edition.

http://www.phoronix.com/vr.php?view=23259

Great article!

Bought Premium to support you making more of those

[artivision]

Of course this has everything to do with efficiency. Pascal provides the ability to process compute and shader simultaneously, as did Maxwell. However, pascal has added a) dynamic load balancing and b) pixel-level preemption instead of preemption at draw call boundries. So now compute and shading can conclude with much higher resource utilization in a much shorter time period (a) and asynchronous (e.g "real-time") tasks can be completed with much lower latency (b).

Check out the attached slides. It seems to me that NV delivers much of what AMD promises. Low latency? Got it. Compute + shading at delivered concurrently? Yup. Much better resource allocation? Got it.

If AMD's version of " async_compute (tm)" is so substantially superior and important then answer me this: when is the last time it made anything anywhere substatially better, much less on Linux? Now remove "AOTS on Windows 10". Got anything left? Right.

See the slides below.
http://cdn.wccftech.com/wp-content/u...-Balancing.jpg
http://cdn.wccftech.com/wp-content/u...on-635x307.jpg

You have mixed some things incorrectly. When we say that a TitanX is just as fast as a 390X on Ashes 4K, we mean: a) That both GPUs running at full TDP (efficiently). b) Ashes is not a full "async" game, if it was, things would be bad for Nvidia. c) TitanX cheats on snow and other FX. Its not good to replace things with your own, if I was a game developer or GTX owner, I would sue them because I can't reproduce my content correctly. Imagine if this was happening with MPEG4 players and movies or music. d) Nothing of all this has changed for GTX1080. On Ashes, two RX480 with 85% utilization =9.5TFlops is 10% faster than a GTX1080 at 100% =9TFlops. The +50% advantage that Nvidia had per Flop comparison is now lost and they will lose more as time goes. To bad because this advantage was less cheat and more "invisible" units like SFUs, extra Load/Store, extra pipeline for 64bit that was faster on 32bit vs native 32bit units. All this cost room and energy. e) GCN GPUs are Shader Model 5.1 (different thinking) not SM5.0.

That I meant to say is that we Linux users have Vulkan and Vulkan has this future. Eventually we will have this with the open source drivers, when most of the games will be made with this. With reasonable prices like 200 bucks. With free FX from AMD for developers to use and experiment upon.

[deppman]

You have mixed some things incorrectly ...

You seem awfully convinced that ATOS on DX12 / Windows 10, an AMD sponsored game with over a year of optimization for GCN architecture is pertinent to a Linux user. What matters to my children is can they play F1 2015 @3440x1440 and 90FPS today on daddy's computer? Or war thunder at 120 FPS? Or shadows of mordor? Or tomb raider? Thanks to the Nvidia GTX 980 Ti, they can. It works virtually flawlessly with a 20% over clock. Nothing from AMD comes close to providing this level of care-free performance on Linux. And when your only OS is Linux, that counts for a lot.

... Linux users have Vulkan and Vulkan has this future ...

I gave up on the "AMD/ATI future" about 10 years ago now and have been much happier as a result. Every year they promise to fix their broken drivers and every year we are disappointed. I really hope I am wrong this year, but I'm not going to waste another minute of my life wrestling with Radeon graphics drivers until they get their act together.

I'm excited about Vulcan, as it can unify development across the world's most popular OS's.

[artivision]

You seem awfully convinced that ATOS on DX12 / Windows 10, an AMD sponsored game with over a year of optimization for GCN architecture is pertinent to a Linux user. What matters to my children is can they play F1 2015 @3440x1440 and 90FPS today on daddy's computer? Or war thunder at 120 FPS? Or shadows of mordor? Or tomb raider? Thanks to the Nvidia GTX 980 Ti, they can. It works virtually flawlessly with a 20% over clock. Nothing from AMD comes close to providing this level of care-free performance on Linux. And when your only OS is Linux, that counts for a lot.


I gave up on the "AMD/ATI future" about 10 years ago now and have been much happier as a result. Every year they promise to fix their broken drivers and every year we are disappointed. I really hope I am wrong this year, but I'm not going to waste another minute of my life wrestling with Radeon graphics drivers until they get their act together.

I'm excited about Vulcan, as it can unify development across the world's most popular OS's.

OK and how many AAA games can you play with Nvidia OpenGL on Linux? I forgot how many I have played with AMD and Gallium Nine. And why you use Linux with closed core code in the first place? If I could accept that, I would use Windowz and have it all.

[deppman]

I use Linux because that is where I deploy my software and I can be vastly more productive with it vs a closed toy OS like Windows spyware. But I'm not afraid of buying closed software when it makes me more happy or more productive. I contribute $$ to OSS projects if I use the product a lot - like this site. I also pay for plenty of proprietary software while also sharing a lot of OSS. While all things being equal I'd prefer if NV were more open, the choice I have today is AMD where support has featured major suckage forever, or NV where everything just works beautifully.

It seems a lot of people who are super religious about Linux and OSS tend to dual boot. I'm not one of those hypocritical hippies. The reality is people spend a lot of money creating good hardware and software and need to be compensated for it, including myself. Being OSS is very desired attribute of software, but it's not the only one. Others, such as "is it good" and "does it work" are just as important. Paying good money for tools that work in my preferred and much more open ecosystem is an excellent way to spur adoption and compensate those who deserve it.

[GreekGeek]

Just pre-ordered an EVGA Founders Edition, ex Amazon. I am looking forwards to speed, stability and no heat dumped into my case-which I can finally close again. At 4K, this thing will be a total Bad Ass Beast. It is, I gotta say the "Muhammad Ali" of video cards, at this point in time.

Did I mention stable drivers?

Drums fingers waiting for dispatch.

Cool link deppman.

GreekGeek :-)

[Passso]

OK and how many AAA games can you play with Nvidia OpenGL on Linux? I forgot how many I have played with AMD and Gallium Nine. And why you use Linux with closed core code in the first place? If I could accept that, I would use Windowz and have it all.

This "all or nothing" way of thinking is pathetic. Do you refuse to run closed source softwares like games other than OpenArena or Tux racer too?
Some people use a Linux system because they like it and run closed code/drivers on it because they want/need it.

This is called freedom.

[mitcoes]

I would love to read bare metal MS WOS vs QEMU KVM GPU pass through - especially with the arch (AUR) linux-vfio packages - MS WOS and GNU/Linux vs WINE vs bare metal GNU/Linux benchmarks.

[efikkan]

OK and how many AAA games can you play with Nvidia OpenGL on Linux? I forgot how many I have played with AMD and Gallium Nine. And why you use Linux with closed core code in the first place? If I could accept that, I would use Windowz and have it all.

If you are experiencing allergic reactions to proprietary code, be careful not to open any web-browser! Most web pages are proprietary.