Doom Vulkan vs Open GL performance

shankly1985 · 13 Jul 2016 at 12:06

Rroff said:
Windows 10 store is one of the worst implementations of a store/repository like feature I've seen - how can a big company with MS's experience get it so wrong?

True hopefully this big change coming 2nd August will be good step up.

Rroff · 13 Jul 2016 at 12:11

shankly1985 said:
True hopefully this big change coming 2nd August will be good step up.

I hope so - there is so much squandered potential in the OS - so many features I want to love and should be a step up from 7 but the implementation leaves me wondering how MS could have so utterly failed so badly to read the last 20 odd years of what worked and didn't in previous OSes.

Orangey · 13 Jul 2016 at 12:19

Rroff said:
Yeah good post(s) and much more accurate than most people - though I'd slightly disagree on a couple of points - it is possible to get gains from Maxwell - just takes a lot more thought and depends more on the nature of the workload compared to GCN which you can load up far more indiscriminately - as he/she said you are over-riding the architecture's native functionality to a degree so you really need to know how the card works with Maxwell to know if doing things differently will have a benefit. Which kind of explains the delay in support with Doom.

I was amazed to find someone on the internet who actually doesn't believe async is a generic magic speedup.

Async compute itself is NOT a magic bullet, and must be done properly other avoid stalls and can have disastrous effects if done poorly. It's a tool to offload compute tasks to reduce frame times, and like any other tool it can be a detriment if done wrong.

People treating it like CUDA before anyone knew what it actually did.

Doom112 · 13 Jul 2016 at 12:19

Another benchmark of Doom Vulkan.

https://www.youtube.com/watch?v=WOaHpZjQ73M

I wonder how Computerbase did their benchmark.

shankly1985 · 13 Jul 2016 at 12:26

Look forward to digital foundry results. I dislike using charts because they don't show the full picture.

External capture is the only way to benchmark a game and showcase it side by side in a video.

A chart is only good for a reference point.

TheRealDeal · 13 Jul 2016 at 12:29

Orangey said:
I was amazed to find someone on the internet who actually doesn't believe async is a generic magic speedup.

People treating it like CUDA before anyone knew what it actually did.

What is your actual point. Console developers have used it and proclaimed the gains are good. In every game that it's used on the pc the gains are good. ROTTR has just added in Async and the gains are good. It might not be magic but it seems as a tool the gains are there to be extracted when used properly.

Dygaza · 13 Jul 2016 at 12:32

Game doesn't have proper benchmark, so everyone tests from their own savegames. Which naturally means that different scenes favor different architechtures differently. Best way is to threat all reviews legit.

I've paid attention that most people who has posted videos 1070/80 not getting benefits has 6 or 8 core cpu, and people who has benefits has 4C/8T cpu's. (Remember when I7 4C/8T cpu overall usage exceeds 50%, it's performance scaling drops as performance delta comes from hyperthreading mostly).

For example just by looking at this guy's cpu usage on OpenGL, it's just no wonder he gets benefits from Vulkan.

https://www.youtube.com/watch?v=pw9Q4FhBKd0

Edit. It's shame to ask you Gregster to do more tests, as you have already done so much. But could you do some quick tests by disabling few cores from your system and seeing if you get performance upgrade when you tone down your cpu to match traditional I7? - Thanks.

SiDeards73 · 13 Jul 2016 at 12:45

TheRealDeal said:
What is your actual point. Console developers have used it and proclaimed the gains are good. In every game that it's used on the pc the gains are good. ROTTR has just added in Async and the gains are good. It might not be magic but it seems as a tool the gains are there to be extracted when used properly.

Just ignore him, he lost at life so decides to come here and be a miserable so and so and pee in everyones cornflakes.

The gains are there and explained in a very well and easy to read format, honestly theres a lot of people in this thread that are either totally blinded by one side or the other or are desperately trying to prove the rest of the internet wrong.

At the base level, AMD gain hugely from Vulkan / DX12 / Async due to their hardware finally being leveraged correctly, NVidia do not gain as much as their hardware is already being used correctly and efficiently, thats it, end of story really.

Lokken86 · 13 Jul 2016 at 12:50

Doom112 said:
Another benchmark of Doom Vulkan.

https://www.youtube.com/watch?v=WOaHpZjQ73M

I wonder how Computerbase did their benchmark.

Haha exactly the increase I posted about 4k, 37 - 52fps. I'm sure I can get to 60 by enabling the shaders and a small oc.

Also it's as I suspected that the high end Nvidia cards are seeing more cpu bottleneck at 1080p, which is why they don't see much gain after this.

Curious as to why the Fury's are not much ahead of the 390's. 4gb?

Lokken86 · 13 Jul 2016 at 12:52

Dygaza said:
Edit. It's shame to ask you Gregster to do more tests, as you have already done so much. But could you do some quick tests by disabling few cores from your system and seeing if you get performance upgrade when you tone down your cpu to match traditional I7? - Thanks.

He loves it.

Dygaza · 13 Jul 2016 at 12:58

Doom112 said:
Another benchmark of Doom Vulkan.

https://www.youtube.com/watch?v=WOaHpZjQ73M

I wonder how Computerbase did their benchmark.

Take a look at the power effieciency graphs. My head just can't get through with the math that how does AMD not use more power in Vulkan, when it's card really starts to work. I can think only for a reason to be that the card is running idle on high power under OGL / DX11, instead of shutting down shaders (or parts in general) not in use.

Rroff · 13 Jul 2016 at 13:00

Dygaza said:
Take a look at the power effieciency graphs. My head just can't get through with the math that how does AMD not use more power in Vulkan, when it's card really starts to work. I can think only for a reason to be that the card is running idle on high power under OGL / DX11, instead of shutting down shaders (or parts in general) not in use.

Parts will be in use just not fully utilised - doesn't mean that they can necessarily be power gated, etc. as that might cause a penalty to performance if they are in frequent use but under-utilised.

TheRealDeal · 13 Jul 2016 at 13:10

SiDeards73 said:
Just ignore him, he lost at life so decides to come here and be a miserable so and so and pee in everyones cornflakes.

The gains are there and explained in a very well and easy to read format, honestly theres a lot of people in this thread that are either totally blinded by one side or the other or are desperately trying to prove the rest of the internet wrong.

At the base level, AMD gain hugely from Vulkan / DX12 / Async due to their hardware finally being leveraged correctly, NVidia do not gain as much as their hardware is already being used correctly and efficiently, thats it, end of story really.

+1 To me it's very obvious.

Doom112 · 13 Jul 2016 at 13:26

Lokken86 said:
Haha exactly the increase I posted about 4k, 37 - 52fps. I'm sure I can get to 60 by enabling the shaders and a small oc.

Also it's as I suspected that the high end Nvidia cards are seeing more cpu bottleneck at 1080p, which is why they don't see much gain after this.

Curious as to why the Fury's are not much ahead of the 390's. 4gb?

When you are posting the video. It will be really fun to benchmark.

Doom112 · 13 Jul 2016 at 13:30

gamegpu Doom Vulkan benchmark

http://gamegpu.com/action-/-fps-/-tps/doom-api-vulkan-test-gpu

shankly1985 · 13 Jul 2016 at 13:32

Doom112 said:
gamegpu Doom Vulkan benchmark

http://gamegpu.com/action-/-fps-/-tps/doom-api-vulkan-test-gpu

Again you can go and dig the net up all you want, Chart Benchmarks are OLD school.. They is so much wrong with them its pointless.

Use the already available video benchmarks on youtube for a much better understand whats happening.

Dygaza · 13 Jul 2016 at 13:49

Just to note gamegpu ain't running settings that enable async compute. So AMD cards currently in that test are missing that ~10% performance.

layte · 13 Jul 2016 at 14:01

shankly1985 said:
Again you can go and dig the net up all you want, Chart Benchmarks are OLD school.. They is so much wrong with them its pointless.

Use the already available video benchmarks on youtube for a much better understand whats happening.

Remember boys and girls. The only valid results, are those that fit your bias.

And everyone lived happily ever after, well apart from the GPU subsection trolls, who grumpily stalked back to their bridges muttering about charts and videos.

Griffildur · 13 Jul 2016 at 14:02

layte said:
Remember boys and girls. The only valid results, are those that fit your bias.

And everyone lived happily ever after, well apart from the GPU subsection trolls, who grumpily stalked back to their bridges muttering about charts and videos.

can't say I disagree with you tbh ...

shankly1985 · 13 Jul 2016 at 14:08

layte said:
Remember boys and girls. The only valid results, are those that fit your bias.

And everyone lived happily ever after, well apart from the GPU subsection trolls, who grumpily stalked back to their bridges muttering about charts and videos.

A video tells a much big picture than some chart. I have posted videos that come up in my Youtube feed from channels I follow. I have not went searching for anything.

We in a time now that charts should just be scrapped completely they are worthless!! Raw Footage captured from game and side by side is the only way to really showcase 1. how they benched the game and 2. shows the real performance mins and avg and Like DF adds frame rate latency KEY to how smooth a game plays.

A chart showing Doom 1440p min 60, max 120, avg 80 Is pointless. This isn't the first time I have banged on about this.