Rise of the Tomb Raider DX12 Performance Benchmarks

[csbin]

http://techbuyersguru.com/first-look-dx12-performance-rise-tomb-raider?page=1

 

[Bacon1]

For some reason the CPU usage goes to 100% in DX12 vs half that in DX11 while also being slower... Seems there is something causing the CPU usage to go nuts?

http://imgur.com/a/3kwp6

 

[Silverforce11]

For some reason the CPU usage goes to 100% in DX12 vs half that in DX11 while also being slower... Seems there is something causing the CPU usage to go nuts?

Somebody forgot to tell those guys that DX12 is supposed to reduce CPU overhead, so you know, it achieves higher performance while using less CPU resources...

Why would they release DX12 patch in such a broken state?!

At least those Ark guys had the right idea to delay it.

 

[Game_dev]

So the 970 is on par with a 390X in directx 12.

 

[Maverick177]

Weird, I had a 14fps drop switching from DX11 to DX12.

 

[Bacon1]

Somebody forgot to tell those guys that DX12 is supposed to reduce CPU overhead, so you know, it achieves higher performance while using less CPU resources...

Why would they release DX12 patch in such a broken state?!

At least those Ark guys had the right idea to delay it.

Even stranger - http://i.imgur.com/bc5vHti.png

Keeping the game open between runs removed the huge CPU spike during start of each area in the benchmark. CPU usage got less severe each time it ran.

 

[Bacon1]

Weird, I had a 14fps drop switching from DX11 to DX12.

Those are all slower runs not faster... Looks like they have a lot of work to do to get it working right.

 

[Bacon1]

So the 970 is on par with a 390X in directx 12.

You mean when DX12 is broken and all cards lose FPS instead of gaining.. sure. Glad you like being king of broken games

 

[tential]

You mean when DX12 is broken and all cards lose FPS instead of gaining.. sure. Glad you like being king of broken games

This is a prime example of DX12. A GTX 970 is on par with an R9 390x.
Yes, all cards lose FPS in this DX12 title. But this is also a game where Nvidia is doing well. So this game is representative of all DX12 games.

 

[Ma_Deuce]

So the 970 is on par with a 390X in directx 12.

You mean 380X. you must have fat fingered the keyboard.

 

[Game_dev]

NVIDIA looks to be ahead in directx 12. After seeing this and Gears of War. Their full support of directx 12.1 is paying off. That's why I only look at benchmarks from released games.

 

[Mahigan]

Ahh, I see it, The CPU spikes just as the RAM usage rises.

It looks like the CPU spikes due to a lack of framebuffer (spilling over the 4GB frame buffer and into Dynamic Memory).

That's just like what the review showed. The DX12 patch is broken thus far. Needs better memory management.

 

[Bacon1]

Ahh, I see it, The CPU spikes just as the RAM usage rises.

It looks like the CPU spikes due to a lack of framebuffer (spilling over the 4GB frame buffer and into Dynamic Memory).

That's just like what the review showed. The DX12 patch is broken thus far. Needs better memory management.

Whats odd, is there was no difference in the FPS when keeping the game open, but that huge CPU usage went away between runs - http://i.imgur.com/bc5vHti.png

So it must be loading up every texture it possible can, even though they aren't used yet.

 

[Mahigan]

Even stranger - http://i.imgur.com/bc5vHti.png

Keeping the game open between runs removed the huge CPU spike during start of each area in the benchmark. CPU usage got less severe each time it ran.

Not weird,

The CPU spikes you saw were copy commands. Copying textures onto the system memory and framebuffer.

Once they're copied, the CPU spikes won't reappear.

It's what I mentioned elsewhere. This game needs Async Compute and parallel copy commands to the graphics and rendering work.

The devs will likely work on that next. Basically they need to better manage texture streaming so that only the required textures are streamed for each frame instead of streaming them all in and over flowing the framebuffer.

This is affecting all GPUs with under 8GB of framebuffer (including the GTX 980 Ti).

 

[Mahigan]

Whats odd, is there was no difference in the FPS when keeping the game open, but that huge CPU usage went away between runs - http://i.imgur.com/bc5vHti.png

So it must be loading up every texture it possible can, even though they aren't used yet.

Well your image shows that there is no Multi-threaded rendering going on as well.

13, 0, 3, 2, 0, 2, 2, 28. In other words, Rise of the Tomb Raider DX12 only scales across two cores.

I find it odd that two NVIDIA sponsored games now show no implementation of multi-threaded rendering.

1. I don't want to put a conspiracy hat on, but so far it looks like two NVIDIA sponsored titles are maintaining an API over head on AMD GPUs.

2. Also two titles now, sponsored by NV, with horrible memory management.

Maintaining an API overhead hurts all GCN GPUs.
Bad memory management hurts Fiji (but also older GCN GPUs with less than 8GB of framebuffer).

That's quite interesting... If this is maintained then "Gimpworks" could be truer than ever.

 

[Bacon1]

Well your image shows that there is no Multi-threaded rendering going on as well.

13, 0, 3, 2, 0, 2, 2, 28. In other words, Rise of the Tomb Raider DX12 only scales across two cores.

I find it odd that two NVIDIA sponsored games now show no implementation of multi-threaded rendering.

1. I don't want to put a conspiracy hat on, but so far it looks like two NVIDIA sponsored titles are maintaining an API over head on AMD GPUs.

2. Also two titles now, sponsored by NV, with horrible memory management.

Maintaining an API overhead hurts all GCN GPUs.
Bad memory management hurts Fiji (but also older GCN GPUs with less than 8GB of framebuffer).

That's quite interesting... If this is maintained then "Gimpworks" could be truer than ever.

Oh don't use the numbers at the end, thats the current usage (after testing), just use the graphs. So it does spread usage well, but well... so does DX11 version.

There is definitely some ugliness going on though, no reason all this hardware should be losing framerate in an API designed to improve it

 

[tential]

NVIDIA looks to be ahead in directx 12. After seeing this and Gears of War. Their full support of directx 12.1 is paying off. That's why I only look at benchmarks from released games.

Given the 2 games that are indicative of DX12, Gears of War and Rise of the Tomb Raider.
Are you excited for DX12?
I mean, Rise of the Tomb Raider, DX12 hurts the GTX 970 (even though it's faster than the R9 390x) and Gears of War is extremely buggy with DX12.

What's happening and wrong with DX12 games?

 

[Mahigan]

Oh don't use the numbers at the end, thats the current usage (after testing), just use the graphs. So it does spread usage well, but well... so does DX11 version.

There is definitely some ugliness going on though, no reason all this hardware should be losing framerate in an API designed to improve it

Ahh gotcha,

That would have been mighty interesting.

 

[tential]

NVIDIA looks to be ahead in directx 12. After seeing this and Gears of War. Their full support of directx 12.1 is paying off. That's why I only look at benchmarks from released games.

http://techbuyersguru.com/first-look-dx12-performance-rise-tomb-raider?page=1

Since Nvidia loses performance in DX 12, I guess this is indicative of DX 12 games. I guess we'll continue to see the GTX 970 lose performance in DX12 paths compared to DX11 paths. I guess the GTX 970 is a 1080p/DX11 card.

 

[IllogicalGlory]

NVIDIA looks to be ahead in directx 12. After seeing this and Gears of War. Their full support of directx 12.1 is paying off. That's why I only look at benchmarks from released games.

How did you come to the conclusion that this has anything to do with feature levels? Do Tomb Raider or Gears of War use 12_1 features?

I guess by this logic, AMD is ahead in DX11 because they support 11_3 and Kepler doesn't. Feature level support is just some magical number that automatically means your cards run better in the API now.

By my reckoning, Hitman is a released game that NV sucks dirt in. Oh that's not valid because it's Gaming Evolved? Ok, Tomb Raider isn't valid because it's Gameworks.

I don't see NV getting any better in Ashes between now and its March 31st release date either.

 

[Bryf50]

How did you come to the conclusion that this has anything to do with feature levels? Do Tomb Raider or Gears of War use 12_1 features?

I guess by this logic, AMD is ahead in DX11 because they support 11_3 and Kepler doesn't. Feature level support is just some magical number that automatically means your cards run better in the API now.

By my reckoning, Hitman is a released game that NV sucks dirt in. Oh that's not valid because it's Gaming Evolved? Ok, Tomb Raider isn't valid because it's Gameworks.

I don't see NV getting any better in Ashes between now and its March 31st release date either.

He's trolling. Don't give him the attention and replies he's seeking.

 

[chuy409]

I tested it with amd cpus and it increased the performance on both. I tried the fx 8120 oc to 4.77ghz and phenom 960t with all 6 cores unlocked oc to 4.22ghz. I saw about a 10 fps increase on both. They are running with a gtx 980 btw.

http://m.imgur.com/a/Zv2L6

 

[Silverforce11]

I tested it with amd cpus and it increased the performance on both. I tried the fx 8120 oc to 4.77ghz and phenom 960t with all 6 cores unlocked oc to 4.22ghz. I saw about a 10 fps increase on both. They are running with a gtx 980 btw.

http://m.imgur.com/a/Zv2L6

That's some funky data, DX12 implementation only helps AMD CPUs? Someone forgot to tell these console devs about Intel on the PC.

 

[Osjur]

Did some testing with Fury X and yep, what a nice DX12 patch. I lost around 7fps in soviet installation and it stutters when running in dx12. I can even hear the stutter because my card has some slight coil whine and it stops whining every time the game stutters.

That is with everything maxed out expect Texture Quality which was set to High and 1600p resolution.

But I also lost some DX11 performance with this patch at the same spot in soviet. Nixxes, when you patch a game, people usually expect to get perf. increase, not decrease.

Edit: ran the in-game benchmark, lost 7fps in DX12, it stuttered and there were graphic glitches.

 

[DarkKnightDude]

Seems like DX12 is losing performance for all?