alanm:

Only with Turing and AMD.

Yes that is true, but at least at some point Nvidia users don't have to avoid DX12 anymore, and I think that's a plus.

GREGIX:

Which only shows how fake/crippled was dx12 on pascals. Something like 970 and Has 4gb ram

It's actually the opposite, it shows that Turing has idle execution resources in DX11. Turing lacks the ability to issue a second instruction from a thread in a single clock cycle. Turing requires two cycles to execute an instruction but it can issue instructions every cycle, including independently to both FP/INT. DX12 allows developers to more optimally schedule these instructions and fill the pipeline. They're basically trading efficiency for flexibility.

Yeah Dx12 does give better performance, least it still runs fine for me in Dx11.

goodly guide. 🙂

alanm:

So with Turing, Nvidia finally gets proper DX12 performance (well, at least this title), something AMD had done well with earlier.

With Pascal too, it's showing more consistent frame times and slightly higher fps, but Hilbert only tested the GTX 1070 at 1440p at between 40-50 fps, which is not the right conditions to highlight big DX12 advantage, so I'd expect Pascal to show bigger performance differences at the higher fps levels at lower resolutions or settings, just like is seen with Turing at higher fps. Turing also shows only minimal performance differences in DX12 in the 50 fps zone where it was tested at 4K in this article - an increase from 56 to 59fps. Based on that pattern I'd expect Pascal to get a decent bump from DX12 at settings that allow for 80+ frames per second.

karma777police:

I tested the game under Windows 7 x64 and 1080ti performs the same as on Windows 10 under DX12 where DX11 under Windows 10 is broken in performance terms and I noticed that for time now.

And this is why I stick to win7, despite recent mb/cpu change from z97+broadwell@4,1Ghz to z370 taichi and 8086@5Ghz. Which gave me boost like 10+% and OC mem to 4200cl17 another 30% in some titles(especially in low/minimum FPS). Fun thing is I did not had to reinstall my OS, just refresh chipset/usb drivers. Ah, have to redo MS activation too

Denial:

It's actually the opposite, it shows that Turing has idle execution resources in DX11. Turing lacks the ability to issue a second instruction from a thread in a single clock cycle. Turing requires two cycles to execute an instruction but it can issue instructions every cycle, including independently to both FP/INT. DX12 allows developers to more optimally schedule these instructions and fill the pipeline. They're basically trading efficiency for flexibility.

Which is why they ought to seperate the development of consumer and pro cards. Everything about turing was developed for the pro market, it doesn't favor dx11 gaming (hence smallest performance increase in ages from a new gen), which nearly all games still use. And i find the whole dx12 situation quite troublesome, cause as you say, it allows the DEVELOPERS to more optimally schedule these instructions and fill the pipeline, not nvidia - but do the developers have the knowledge of how to do so, and more importantly, are the developers willing to put the effort into specifically optimize their games for 1 architecture ? I think not, unless they are sponsored by nvidia...

Dragam1337:

Which is why they ought to seperate the development of consumer and pro cards. Everything about turing was developed for the pro market, it doesn't favor dx11 gaming (hence smallest performance increase in ages from a new gen), which nearly all games still use. And i find the whole dx12 situation quite troublesome, cause as you say, it allows the DEVELOPERS to more optimally schedule these instructions and fill the pipeline, not nvidia - but do the developers have the knowledge of how to do so, and more importantly, are the developers willing to put the effort into specifically optimize their games for 1 architecture ? I think not, unless they are sponsored by nvidia...

Well on DX12 most games are inherently optimized for AMD since most of them are being directly ported from AMD hardware on consoles - this benchmark shows that AMD gets a pretty large benefit from DX12. So clearly they are optimizing for multiple architectures which is kind of something you have to do in DX12 regardless to whether its Turing or Pascal or AMD. The difference is with Pascal the best a dev can do is extract parity with Nvidia's DX11 performance, with Turing the flexibility allows them to theoretically go past what Nvidia could ever whip up in the driver with the downside being that DX11 performance isn't as optimal given the hardware. On the flipside there is obviously a cost trade off in both GPU size and power consumption as dedicated hardware scheduling is not only back but the number of dispatch units is now doubled for every CUDA core. This is one of the reasons why 2080Ti has only 20% more CUDA cores but is 65% larger in GPU size. So I guess the question is does the added flexibility in scheduling offset the increased cost/complexity to the GPU vs's a GPU that's just Pascal scaled up? No idea. How large would a Pascal scaled up be? No idea. How much is Turing's cost (2080Ti) attributed to die size vs Nvidia's desire for larger margins? No idea. I slightly agree that I'd like to see the architecture split and continue with a gaming variant lacking the flexibility with an unknown decreased cost but I also admit that I have no idea if Nvidia weighed the cost of doing that decided for some weird reason that it wasn't worth it. Perhaps Pascal just doesn't scale up that well. Perhaps developers were indicating an increased desire for mixed FP/INT workloads in games and Nvidia foresees this architecture being more future proof. Perhaps Nvidia fears Intel's arrival into the GPU space and decided that value-add differentiation with machine learning/RTX was the best way to combat that? Too many unknowns to know why they choose to go this route. I personally don't mind the route, technologically speaking, the only issue is the cost for me. I'm not spending $1200 for 25-30% performance improvements. I'm not spending $1200 for a gamble on whether their value-add features will be in the games I want to play. I personally thought this entire launch was an embarrassment. But the marketing/pricing aspect of Nvidia is a separate thing from the engineering - I still think the stuff they are doing under the hood is really neat. Time will tell if the neat tech pays off or falls to the wayside in the presence of more traditional architectures without all the fancy pants AI stuff and crazy scheduling.

BReal85:

It's nice to see Vega 56 match the 1070Ti performance in NV sponsored titles and beat the 1080 in AMD sponsored ones. So get a 2400-2500$ GPU pair to get stable 60 fps in 4K. Nothankyou. Check gamegpu for the SotTR benchmarks, where you can see the DX11 and DX12 results. In DX12 both AMD and NV get huge boost compared to DX11.... This is how DX12 implementation should be done.

You speak as if i don't have the game nor any of the mentioned gpu's - clearly you haven't been following the shadow of the tomb raider thread. In situations where i am not cpu limited, the fps is roughly the same for me in dx11 / dx12... but dx12 has much more uneven frametimes. Dx11 https://i.imgur.com/LwH7EpZ.jpg Dx12 https://i.imgur.com/PNoz8De.jpg No doubt that dx12 is alot better for turing than dx11, but who cares about that overpriced POS... and obviously amd favors dx12, as always.

Denial:

Well on DX12 most games are inherently optimized for AMD since most of them are being directly ported from AMD hardware on consoles - this benchmark shows that AMD gets a pretty large benefit from DX12. So clearly they are optimizing for multiple architectures which is kind of something you have to do in DX12 regardless to whether its Turing or Pascal or AMD. The difference is with Pascal the best a dev can do is extract parity with Nvidia's DX11 performance, with Turing the flexibility allows them to theoretically go past what Nvidia could ever whip up in the driver with the downside being that DX11 performance isn't as optimal given the hardware. On the flipside there is obviously a cost trade off in both GPU size and power consumption as dedicated hardware scheduling is not only back but the number of dispatch units is now doubled for every CUDA core. This is one of the reasons why 2080Ti has only 20% more CUDA cores but is 65% larger in GPU size. So I guess the question is does the added flexibility in scheduling offset the increased cost/complexity to the GPU vs's a GPU that's just Pascal scaled up? No idea. How large would a Pascal scaled up be? No idea. How much is Turing's cost (2080Ti) attributed to die size vs Nvidia's desire for larger margins? No idea. I slightly agree that I'd like to see the architecture split and continue with a gaming variant lacking the flexibility with an unknown decreased cost but I also admit that I have no idea if Nvidia weighed the cost of doing that decided for some weird reason that it wasn't worth it. Perhaps Pascal just doesn't scale up that well. Perhaps developers were indicating an increased desire for mixed FP/INT workloads in games and Nvidia foresees this architecture being more future proof. Perhaps Nvidia fears Intel's arrival into the GPU space and decided that value-add differentiation with machine learning/RTX was the best way to combat that? Too many unknowns to know why they choose to go this route. I personally don't mind the route, technologically speaking, the only issue is the cost for me. I'm not spending $1200 for 25-30% performance improvements. I'm not spending $1200 for a gamble on whether their value-add features will be in the games I want to play. I personally thought this entire launch was an embarrassment. But the marketing/pricing aspect of Nvidia is a separate thing from the engineering - I still think the stuff they are doing under the hood is really neat. Time will tell if the neat tech pays off or falls to the wayside in the presence of more traditional architectures without all the fancy pants AI stuff and crazy scheduling.

Legit points, but personally i think they decided to just cater to the pro market, and develop for their needs, and then shovel whatever they made to the gaming market, saving any development costs, seeing as there is no competition in the gaming market atm. I think all this RTX bs they came up with, was just a way to justify all the hardware that would otherwise just have been dead weight on the gpu for consumers. If it was really something they had been working on for many years, it would have been ready at launch - i think they came up with it when they decided to just shovel the same turing gpu's to the pro and consumer market, hence why it seems so rushed, despite them having had literally 2,5 years since the launch of pascal. And yeah, it should come as no surprise that i think a scaled up pascal, with a bigger bus width, would have fared much better.

Dragam1337:

Legit points, but personally i think they decided to just cater to the pro market, and develop for their needs, and then shovel whatever they made to the gaming market, saving any development costs, seeing as there is no competition in the gaming market atm. I think all this RTX bs they came up with, was just a way to justify all the hardware that would otherwise just have been dead weight on the gpu for consumers. If it was really something they had been working on for many years, it would have been ready at launch - i think they came up with it when they decided to just shovel the same turing gpu's to the pro and consumer market, hence why it seems so rushed, despite them having had literally 2,5 years since the launch of pascal.

The glass half empty side of me agrees with you. It definitely does look that way.

Guys, dont forget Vulkan. Wolfenstein even more dramatic gains for Turing. I think we all knew that Pascals successor would have to be far better with DX12 and Vulkan APIs. Look at the jumps from Maxwell to Pascal to Turing, even at the higher resolutions. https://www.guru3d.com/articles-pages/geforce-rtx-2080-ti-founders-review,21.html and... Star Wars Battlefront II (DX12) https://www.guru3d.com/articles_pages/geforce_rtx_2080_ti_founders_review,16.html

valrond:

Can you make the Pascal GPUs with drivers 399.24? The 411 is known to cripple 1080 Ti results to that RTX 2080 looks better.

My results are slightly higher with 411 compared to 399... https://i.imgur.com/2q7KULH.jpg https://i.imgur.com/Heba5IZ.jpg

valrond:

Can you make the Pascal GPUs with drivers 399.24? The 411 is known to cripple 1080 Ti results to that RTX 2080 looks better.

"known" in your're dizzy head?

Great review, thanks Hilbert.

For what it's worth, I've has stability issues with this game on an RTX 2080 Ti. I've seen several threads in the Steam forums with very similar problems. Device driver "stuck" crashes with no rhyme or reason to them. I have no idea if it's a particular setting (I've tried a bunch of different ones) or a driver thing, but it's frustrating to say the least. In terms of performance, the RTX 2080 Ti definitely can't max everything and still hit 60fps (with vsync), but it can get pretty close. The key that prevents it seems to be the highest shadow setting. That one setting knocks my performance down by 15-20%. The higher AA settings can have a major effect, too.

I just did a double pass with my GTX980 @1450 (normal OC) with the exact same settings as guru did their review on and instead of 52fps i have 67! I wonder what kind of Maxwell gpus is guru using??? Are they reference cards or underclocked or what??? Also the 980Ti should be above 80fps with the same settings as well! I call this review at least inaccurate.

Dj_ALeX:

I just did a double pass with my GTX980 @1450 (normal OC) with the exact same settings as guru did their review on and instead of 52fps i have 67! I wonder what kind of Maxwell gpus is guru using??? Are they reference cards or underclocked or what??? Also the 980Ti should be above 80fps with the same settings as well! I call this review at least inaccurate.

It depends on cpu speed, cores, memoryfrequency, gpu memoryfrequency and gpu frequency. You need to compare with same settings, software and hardware to compare 🙂

Dj_ALeX:

I just did a double pass with my GTX980 @1450 (normal OC) with the exact same settings as guru did their review on and instead of 52fps i have 67! I wonder what kind of Maxwell gpus is guru using??? Are they reference cards or underclocked or what??? Also the 980Ti should be above 80fps with the same settings as well! I call this review at least inaccurate.

Feel free to start a youtube bench site if you like. I will stick with HH's reviews.

nizzen:

It depends on cpu speed, cores, memoryfrequency, gpu memoryfrequency and gpu frequency. You need to compare with same settings, software and hardware to compare 🙂

I'm pretty sure the guru's test system is better than mine so.. and btw you shouldn't mention the obvious things.. i'm a hardware tester/technician for over 12 years! 🙂

Embra:

Feel free to start a youtube bench site if you like. I will stick with HH's reviews.

You can do whatever you want, believe whoever you want and stick with any reviewer you want. I just don't care! The common user will see those numbers and misscalculate the performance of any of those cards so don't give me that BS. End of coversation for me.