DX12 Multi-GPU, live and well in Shadow Of The Tomb Raider

noko

Supreme [H]ardness
Joined
Apr 14, 2010
Messages
7,262
Using two 1080 Ti's, the configuration is in signature. Just using the typical canned benchmark to see if scaling works with two GPU's. Stock clocks, CPU is actually 4.6ghz and is not reported correctly by benchmark.

First test is all max settings with SMAA, 2x 1080 Ti'x at 3440x1440 resolution (giving some love to the 3440x1440 crowd). TAA gives a driver error and I don't think will work with two GPU's:

MGPUMaxSMAA.jpg
Average FPS is 110 FPS
Now using again max settings but with 4xSMAA with 2x 1080 Ti's, really putting some load on both GPU's:

MGPUMax1.jpg MGPUMax2.jpg
Average FPS is now 70 FPS
Going down to a single 1080 Ti, same settings as directly above, max setting with 4xSMAA

SingleGPUMax.jpg
Average FPS is 42 FPS
I am seeing scaling for two GPU's in this game on this benchmark at 1.67x or 67% increase in performance.

First benchmark or configuration maybe good for those 120 hz Gsync 3440 x 1440 monitors, while I will probably use the Max settings with 4xSMAA using adaptive Sync on a 60 hz monitor which shows a virtually constant 60 FPS experience.

Later I will test two VegaFE's in Multi-GPU configuration if the drivers work with this game, not sure if the Radeon Pro Adrenalin Gaming 18.19.1 are out or will ever be, currently on 18.17.1. This will probably be tomorrow.

So is it still worth getting two GPU's, especially on the Nvidia side with the hike in price?
 
I would like to know as well if Turing or the 2080Ti/80 scales better in DX 12 using Multi-GPU. Are they better DX 12 cards and the more the better result? While SLI is dying, especially over 2 cards but how about 3 cards using Multi-GPU? Just another aspect that I believe very few will cover, especially most reviewers.
 
Presumably they’ll scale worse cause they’re faster and would hit the cpu bottleneck earlier.
 
Presumably they’ll scale worse cause they’re faster and would hit the cpu bottleneck earlier.
A well OC 8600K running at 4K, I would think that would not be the case. This game can bring a 1080 Ti down pretty low at 3440x1440 when maxed out, at 4K it would be terrible but then I would not bother with 4xSMAA at 4K.

I should test at 4K as well if time permits.
 
I struggled to get 60fps in 4K on RotTR, even with 2 Vega 64's. Crossfire did have decent scaling, but the game is intensive. Had to use medium settings and limited AA.
 
Presumably they’ll scale worse cause they’re faster and would hit the cpu bottleneck earlier.
The point of DX12 is the remove the CPU bottleneck so I wouldn't expect that to be the case unless it was an extreme circumstance.
 
The point of DX12 is the remove the CPU bottleneck so I wouldn't expect that to be the case unless it was an extreme circumstance.

Yeah, but he is saying that More graphics power is morel likely to get bottlenecked than less.
 
Yeah, but at higher res (4K, multi-monitor) you will still be GPU limited, even with 2 cards.
 
Yeah, but he is saying that More graphics power is morel likely to get bottlenecked than less.

As true as this is, we'd have to see where the bottlenecks hit. Hard part is that this is not just per title, but per system per settings per scene...

Still, I don't think the CPU will come into play if DX12 mGPU is well implemented, at least for games like this that have linear dynamics and no multiplayer. Give me a proper BF4 sequel with massive destruction and let's say up it to 128 players and that might actually be an issue.
 
1080 Ti DX 12 Multi-GPU 4K results below. As a note: Vega FE with current available drivers and current version of Shadows Of The Tomb Raider does not support DX 12 Multi-GPU, neither the Pro drivers or the newest 18.8.1 gaming drivers work with Multi-GPU, single GPU yes but with dismal performance at 4K.

First I did a quick test using Nvidia settings for 4k in Shadow Of The Tomb Raider to see how a factory OC card, EVGA SC Black performs. Highest preset setting with TAA:

TAAHighest2.jpg TAAHighest.jpg
Average FPS 44

NVtesting.png
So in this title the 2080 47 fps is faster then an OC version of the 1080 Ti. I could OC the EVGA card and get higher numbers but not sure how well the 2080 will OC. Anyways back to Multi-GPU.

Shadow Of The Tomb Raider is utterly stunning at 4K, jaw dropping at that. Using max settings with SMAA at 4K, single GPU:

4KSgMax.jpg 4KSgMax2.jpg
Average FPS 43
Upping the ante with two 1080 Ti's, DX 12 Multi-GPU, max settings with SMAA at 4K:

4kMaxSMAA2.jpg 4KMaxSMAA.jpg
Average FPS 82
The scaling with two GPU's, 82/43 = 1.91 or 91% increase in performance, is OUTSTANDING! Max settings and getting way over 60 FPS. Now I wish all titles scaled like this, my hat is off to the developers.
 
Last edited:
Nice work, thanks.
It is a very pleasant surprise and looks to be some hope for multiple graphics card which will not only push fast refresh rates for 4K but should allow 8K as well as next generation VR to perform well in.
 
nice! can you show us a bench of two 2080 rtx at max?
thank you :)
I think Brent and Kyle will be able to do that since they have 2x for both the 2080 and 2080 Ti. Also testing those PriMax 8x (real 4k panels being rendered at 4k each eye), just need some DX 12 VR games.
 
The scaling with two GPU's, 82/43 = 1.91 or 91% increase in performance, is OUTSTANDING! Max settings and getting way over 60 FPS. Now I wish all titles scaled like this, my hat is off to the developers.

It's a good indicator- but what kills multi-GPU performance is not the average framerates, but the maximum frametimes and frame pacing. Those have shot way up in the past, particularly older implementations of Crossfire, such that the experience was actually worse than using a single card.

Call that worst case; I don't think that that's what we're seeing here, but we need to use the appropriate measure when we get the chance :).
 
SLi and crossfire are effectively dead which is bad news as an eyefinity user. I really hope dx12 mutliGPU becomes a thing devs actually care about.
 
It's a good indicator- but what kills multi-GPU performance is not the average framerates, but the maximum frametimes and frame pacing. Those have shot way up in the past, particularly older implementations of Crossfire, such that the experience was actually worse than using a single card.

Call that worst case; I don't think that that's what we're seeing here, but we need to use the appropriate measure when we get the chance :).
In the past with DX 11 titles that was indeed an issue, especially with older AMD drivers. Looking at Shadow Of The Tomb Raider canned benchmark and the added data besides average frame rates shows something really good I think, only down side is this is only one game:
  • The 4K Single Gpu with Max settings using SMAA had a 95% percentile of 38 FPS, meaning 95% of all frames were 38 FPS or faster
  • With Multi-GPU that 95% percentile jumps to 74 FPS, meaning 95% of all frames rendered time wise were 74 FPS or faster!
  • With DX 12 Multi-GPU
    • Two cards gave 74/38 = 1.95x or increase of 95% performance increase over a single card for 95% of the frames rendered! It does not get much better than this.
The graphs for frame times seem to be too limited to tell if there are any significant pacing issues, I cannot discern any at this point but then I am not playing the game yet (let down on this) and just going by the canned benchmark. Brent, the top professional on GPU game testing I hope will do a Shadow Of The Tomb Raider break down. Actual game play, experience, settings significance out weigh just the numbers overall.
 
Presumably they’ll scale worse cause they’re faster and would hit the cpu bottleneck earlier.
Well on second thoughts, I may indeed be CPU limited in this title:
  • The canned benchmark indicates 100% GPU bound when at 4K Max settings using SMAA
  • With Multi-GPU it shows 17% GPU limited
  • Scaling from 1 to 2 GPU's was 91% for the average, 95% for 95 percentile so I am confused what the benchmark is indicating here on being GPU bound
    • I will need to see effects of clock speed for the 6700K has on this, lowering should be easy, not sure how high I can push it (never pushed it since previous cooler would not allow it)
 
  • The 4K Single Gpu with Max settings using SMAA had a 95% percentile of 38 FPS, meaning 95% of all frames were 38 FPS or faster
  • With Multi-GPU that 95% percentile jumps to 74 FPS, meaning 95% of all frames rendered time wise were 74 FPS or faster!
  • With DX 12 Multi-GPU
    • Two cards gave 74/38 = 1.95x or increase of 95% performance increase over a single card for 95% of the frames rendered! It does not get much better than this.

While I absolutely appreciate the data breakout, I'm specifying 'maximum frametimes' for a reason: 95% at 38FPS or faster- meaning 19/20 frames are rendered in at least 26ms- leaves 5%, or 1/20 frames, that could be absolute trash.

If one out of every twenty frames is excessively long, the user is going to experience that as stutter.

Realistically we need 99.9% of frametimes to be fast, so bad frames would only be experienced as minor hitching, and we need to see the data to verify that. This is why Techreport, for example, posts the percentile numbers, the percentile graphs, and the 'time spent above n-ms' graphs.
 
  • Like
Reactions: noko
like this
In the past with DX 11 titles that was indeed an issue, especially with older AMD drivers. Looking at Shadow Of The Tomb Raider canned benchmark and the added data besides average frame rates shows something really good I think, only down side is this is only one game:
  • The 4K Single Gpu with Max settings using SMAA had a 95% percentile of 38 FPS, meaning 95% of all frames were 38 FPS or faster
  • With Multi-GPU that 95% percentile jumps to 74 FPS, meaning 95% of all frames rendered time wise were 74 FPS or faster!
  • With DX 12 Multi-GPU
    • Two cards gave 74/38 = 1.95x or increase of 95% performance increase over a single card for 95% of the frames rendered! It does not get much better than this.
The graphs for frame times seem to be too limited to tell if there are any significant pacing issues, I cannot discern any at this point but then I am not playing the game yet (let down on this) and just going by the canned benchmark. Brent, the top professional on GPU game testing I hope will do a Shadow Of The Tomb Raider break down. Actual game play, experience, settings significance out weigh just the numbers overall.

That is impressive, but really needs frame time charts to convince me. 0.1% of frames spiking can ruin it.

Hopefully PCPER or [H] revisit SLI soon. They usually do frame time charts.

Thanks for the data you’ve given. I was convinced SLI was dead but apparently it might not be!
 
  • Like
Reactions: noko
like this
I would like to know as well if Turing or the 2080Ti/80 scales better in DX 12 using Multi-GPU. Are they better DX 12 cards and the more the better result? While SLI is dying, especially over 2 cards but how about 3 cards using Multi-GPU? Just another aspect that I believe very few will cover, especially most reviewers.

I think nvidia and AMD no loger support tri-gpu configuration for some time.
 
I love the active nay sayers that SLI and Crossfire are dead. Yes, why yes, AMD CROSSFIRE is dead. Not supporting developers does that to your card and company.

SLI is on a resurgence and why SoTTR has great support. Nvidia unlike AMD is active with developers on Ray Tracing, which requires more horsepower to run at higher resolutions. So while helping developers with RT, in goes SLI support to actually be able to play higher resolutions with higher frame rates. Now Nvidia sees, CHAAAAAAAAAAAACHING! Sell more $20-80 SLI bridges, more cards and get Ray Tracing out today by powering it with 2 cards.

Also, another bonus for Nvidia with resurgence of SLI=clear old inventory faster. As much as I can't stand them at times, Gosh damn them bastards are brilliant.
 
works with my 2 rx 480s pretty good.

any AA mode that uses the previous frame though (like TAA) looks a little strange because of the microstutter that occurs at times. as long as the FPS are high enough though it isn't too distracting.

mGPU/D3D12AFR also works in Strange Brigade too.

I hope this sort of thing catches on because the results are awesome
 
I love the active nay sayers that SLI and Crossfire are dead. Yes, why yes, AMD CROSSFIRE is dead. Not supporting developers does that to your card and company.

SLI is on a resurgence and why SoTTR has great support. Nvidia unlike AMD is active with developers on Ray Tracing, which requires more horsepower to run at higher resolutions. So while helping developers with RT, in goes SLI support to actually be able to play higher resolutions with higher frame rates. Now Nvidia sees, CHAAAAAAAAAAAACHING! Sell more $20-80 SLI bridges, more cards and get Ray Tracing out today by powering it with 2 cards.

Also, another bonus for Nvidia with resurgence of SLI=clear old inventory faster. As much as I can't stand them at times, Gosh damn them bastards are brilliant.

SLI and Crossfire only lead to one thing.. Disappointment and frustration. On a very rare occasion it works like it should and the moments are great but are far too rare to make it worth it anymore. Also your guessing with two cards working with Ray Tracing as Nvidia has said nothing of the sort. Vulkan might save multi gpu but even that is unlikely unless more developers support it.
 
SLI and Crossfire only lead to one thing.. Disappointment and frustration. On a very rare occasion it works like it should and the moments are great but are far too rare to make it worth it anymore. Also your guessing with two cards working with Ray Tracing as Nvidia has said nothing of the sort. Vulkan might save multi gpu but even that is unlikely unless more developers support it.
No disappointment here on either, one just need to know the limitations. Most games I tend to play work well. Plus this has nothing to do with SLI/CFX - this is DX 12 and Multi-GPU. Less driver and profile based and more API and developer made.
 
works with my 2 rx 480s pretty good.

any AA mode that uses the previous frame though (like TAA) looks a little strange because of the microstutter that occurs at times. as long as the FPS are high enough though it isn't too distracting.

mGPU/D3D12AFR also works in Strange Brigade too.

I hope this sort of thing catches on because the results are awesome
Your talking about Shadow Of The Tomb Raider in DX 12, Multi-GPU? Good news if the case.
 
SLI and Crossfire only lead to one thing.. Disappointment and frustration. On a very rare occasion it works like it should and the moments are great but are far too rare to make it worth it anymore. Also your guessing with two cards working with Ray Tracing as Nvidia has said nothing of the sort. Vulkan might save multi gpu but even that is unlikely unless more developers support it.

Always has.

Ever since the first days of Voodoo 2 SLI it has always been under supported. Sure a few games work and just enough to make people think it might be cool but if you've ever actually owned an SLI or Crossfire setup you can't appreciate how often it doesn't.

Been there, done that. Not going back there again.

Maybe if RT catches on the graphics pipeline will be easier to make parallel between video cards for all games. But that's how the arguments have always gone. "Next time... next gen... next driver release... next big game release..."

Sure, you can get awesome scaling on SoTR but what about allllllll the other games that wont? Even if you could somehow find a couple of used 1080Ti cards at $400 each, $800 total you'd be AT the price of a 2080. You can either have 90+% scaling on this ONE game, or you can overclock the 2080 as far as it'll easily go and probably have minimum 30% and up to 50% more than the Ti card on ALL new titles and not just the ones that happen to scale.

And as always, twice the video card, twice the power (bigger power supply needed), twice the heat, maybe a bigger case, more cooling hardware and one video card that spends most of its life doing absolutely nothing while you are on the desktop or playing games that don't SLI.

It's just not fun to own.

EDIT- but the OP is correct, it is AWESOME scaling in a game.
 
Your talking about Shadow Of The Tomb Raider in DX 12, Multi-GPU? Good news if the case.

yup! I've got a 1060 coming in tomorrow from evga bstock and am considering trying to pair it with another 1060 - obviously no SLI but what about mgpu? couldn't hurt to try just to see what happens :D
 
  • Like
Reactions: noko
like this
Always has.

Ever since the first days of Voodoo 2 SLI it has always been under supported. Sure a few games work and just enough to make people think it might be cool but if you've ever actually owned an SLI or Crossfire setup you can't appreciate how often it doesn't.

Been there, done that. Not going back there again.

Maybe if RT catches on the graphics pipeline will be easier to make parallel between video cards for all games. But that's how the arguments have always gone. "Next time... next gen... next driver release... next big game release..."

Sure, you can get awesome scaling on SoTR but what about allllllll the other games that wont? Even if you could somehow find a couple of used 1080Ti cards at $400 each, $800 total you'd be AT the price of a 2080. You can either have 90+% scaling on this ONE game, or you can overclock the 2080 as far as it'll easily go and probably have minimum 30% and up to 50% more than the Ti card on ALL new titles and not just the ones that happen to scale.

And as always, twice the video card, twice the power (bigger power supply needed), twice the heat, maybe a bigger case, more cooling hardware and one video card that spends most of its life doing absolutely nothing while you are on the desktop or playing games that don't SLI.

It's just not fun to own.

EDIT- but the OP is correct, it is AWESOME scaling in a game.
In the end it becomes personal choice and great to have options. I have been CFX AND SLI for awhile now. Sometimes it is the only option to push high refresh rates and 4K if one is a total geek about settings. Is it needed? I would say hardly but then again a lot of what we do here at [H]ardOCP is not needed but we do know how to have fun and how to push boundaries.

CFX/SLI is nearly dead, here comes mgpu.
 
Last edited:
Been there, done that. Not going back there again.

Actually had a pretty good experience with GTX670SLI and then GTX970SLI. The GTX670s replaced HD6950 Crossfire, before AMD got called to carpet for their shit-tastic mGPU implementation, and man was the Nvidia setup an improvement.

And while it does depend on developers being moderately cooperative, I found it worked well nearly all the time. The biggest break in support was the DX12 transition, and it looks like we're over that hump now too.
 
yup! I've got a 1060 coming in tomorrow from evga bstock and am considering trying to pair it with another 1060 - obviously no SLI but what about mgpu? couldn't hurt to try just to see what happens :D

My understanding is this is doable (not done it so don't quote me!). If mGPU can support an AMD/RTG and nVidia card trying it (in supported games) with two 1060s should work.
 
Hey, I certainly hope so. It would be nice to have universal mGPU happen finally. How much of a benefit would that be to PC gaming in general if gamers, especially the ones with less money to throw at graphics cards could ACTUALLY get consistent improvement in all games by putting together a couple of mid range cards later in their life cycle? I don't think it would hurt sales of new cards so much as it would put more otherwise useless cards to work with people who normally wouldn't buy new cards.

The result would be MORE gamers with higher-end graphics horsepower. That would benefit all of us.
 
  • Like
Reactions: noko
like this
Does anyone know if DX12 multi-GPU still adds 1 frame of input lag or has any issues with micro stutter? I remember when I was playing Witcher 3 with 2x 970 in SLI and moving to a single 980 Ti just made it feel both smoother and more responsive.
 
Back
Top