The nVidia DX12 performance cover up?

Hardware Hub GPU
1

Today the Fable benchmarks came up on various sites and they show a completely different picture than the Ashes tests.

The AotS tests brought the great Async Compute shit storm which have not yet settled.

As I was reading the Fable benchmark reviews I could not help but notice that now everywhere we see Fury X and 980 Ti in them with 390X card even matching the non-X Fury (!), and the 980 Ti is beating the hell out of the Fury X, which is surprising, as Fable is said to be using DX12 to the fullest, including lots of Async Compute, so where is the catch, I wondered.

Remember when AotS scandal happened, nVidia said that they are forced to emulate some of the DX12 features that they do not have in hardware?

Here are two different pics from different sites:

Notice how the Fury X results are the same, but the 980 Ti are not? Guess what - the first one is made on 5960X, the second one a regular i7 (presumably 6700K).

What this tells me is that nVidia is offloading work to the CPU through their drivers and gaining FPS while doing so, while AMD is doing everything on the GPU hardware, completely leaving the CPU out.

The bad thing from this is that the average consumer just looks for who's on top, not how it went there, and I think there should be more work put to investigate that deeper.

Discuss.

5 Likes
2

Interesting theory. Needs a lot more data though to come up with any solid conclusions though.

One thing to note, on the PcPer video, Ryan mentioned something about Fable Legends not being CPU intensive and that any solid quad core CPU should run it (2600k for example). For this reason they did not examine multiple CPUs. They used a 6600k and the 980Ti just nudged ahead of the Fury X by a couple of fps (similar to the Anand bench).

3

This might have some truth behind it because upgrading from the i5 3570k to the i7 5820k I noticed a 7-10 fps difference in games like gta v and in call of duty, 20 fps difference from 80 to 100 fps.(TBN: this is a best case senario with the fps) And on my small itx system I was getting similar performance with a 960 on a 860k, to a 750ti in a system with with a amd 8 core in CS:GO though with that game there might be a bottleneck somewhere else

4

This was argued in another post. They were not using the newest AMD drivers. They decided to use old AMD driver. /cough Nvidia loading pockets? /jking If you go to extreme tech website they use the newest drivers for AMD and nvidia and they show a completely different outcome. Granted they are only 1 or 2 frames per sec faster but it does not show Nvidia blowing AMD away.

5

Some more thoughts:

UE4 docs tell that it supports Async Compute only for XBone. Meaning that Fable Legends as a full DX12 benchmark is rather exaggerated, more like DX11.5 benchmark. Add to that what Epic's CEO said: "Epic developed Unreal Engine 4 on NVIDIA hardware, and it looks and runs best on GeForce."

The guys at ExtremeTech did a review and according to them: The average asynchronous workload per frame (meaning work that could be handled by an asynchronous compute engine) is 5%. It's not a large amount, in other words.

Looks to me that the release of this benchmark is nVidia's answer to the AotS drama trying to show their DX12 prowess. What they did not forsee, however, is the performance of the 390X which simply destroys 980 GTX and is 150$ cheaper...

6

Wow, they f-ed up one of the advantages of DX12 - offloading CPU tasks to the GPU. So they just pushed the tasks back in the CPU... Jesus Christ... And those guys hold 80% of the market...

7

They have no choice - they do not support it in hardware... But my concerns are for the latter part of your sentence - the 80% market part... And Asyn Compute is one of the major differences between DX11 and DX12.

This is where Tek journalism should provide for the masses.

8

Yeah, right. Like testing games with Gameworx on, and stating that 970 beats FuryX.. We know how journalists are going to handle this. I bet my virgin ass, that youtubers like Jay and Linus will go in stating how much better Nvidia is in DX12...

9

Damn. This was a nice catch. I would love to see a cpu spread with the FuryX and the 980ti. Would be interesting to see it go from the 860k and the g3258 to the 8350 and the 6600k, etc, etc ,etc. We need more data.

EDIT: I would love to see something like this article happen more often.
http://www.pcper.com/reviews/Systems/Quad-Core-Gaming-Roundup-How-Much-CPU-Do-You-Really-Need

EDIT: I just remembered this.
http://www.anandtech.com/show/9659/fable-legends-directx-12-benchmark-analysis/3
Looks like the 980ti isn't offloading that much after all.

10

Well, more cores of the same caliber are somewhat predictable. It is offloading, as there are features that it doesn't have in hardware, so it has to emulate them. Also, bear in mind that DX12 is limited to 6 threads (thanks XBone ...), which again is an artificial limitation.

Still, the 5960X results are very clear on the matter. 20% more FPS just from the CPU? :)

The much bigger problem is that Fable Legends is not doing Async Compute, yet it's pitched as awesome DX12 PC benchmark and people judge DX12 performance of the current nVidia products with it.

11

Not realy thay just used old AMD divers (15.2?) we are on 15.7 lol

12

Err... They used older drivers and achieved 30 fps - exactly the same fps that others achieved with newer drivers?

13

here thay used the current beta driver for fury and the lateset drivers for the 390 --> http://www.extremetech.com/gaming/214834-fable-legends-amd-and-nvidia-go-head-to-head-in-latest-directx-12-benchmark

14

Fury's results aren't lower so we are to blame older drivers, it's 980 Ti's results that are higher with a better CPU, while the Fury results are the same - no matter the CPU or the driver.

15

Fury drivers havent realy matured yet (havent even added good overclocking to the card yet) DX12 makes use of extra core's of the cpu as well.
The 390 (290) has been around a long time now and drivers are great just look how it pounded the 980.

16

Dude :) I will say it again - we are not commenting the Fury performance. We are talking 980 Ti's performance with better CPU related to the Fury performance.

If Fury can achieve 30 FPS with 5960X and 6700K, but the 980Ti is gaining 20% more FPS with the 5960X over 6700K - that means the CPU is aiding nVidia and not aiding AMD, which means nVidia offloads work to the CPU.

Is that clear enough? Regardless of drivers.

1 Like
17

And thats the way DX 12 is supposed to work DX12 was made to use more than 2-4 core's for gameing so a cpu with more core's is theoretically going to be better AND the AMD driver for fury isent good yet drivers can make or brake it.

18

Except:

Scaling from 2 to 6 cores does nothing. Your point being? And what's that to do with drivers?

19

Dosent look like nuthing lol
Rmeber its still all beta and this 'bench' isent even full teir DX12

1 Like
20

You guys need to learn a few general technical things about developing a game before coming up with these conspiracy theories. One of the first things you need to know when developing a game is a thing called "the game loop". Here's an extremely simplified one that I got from our good friends at wikipedia:

while( user doesn't exit )
check for user input
run AI
move enemies
resolve collisions
draw graphics
play sounds
end while

To simplify things let's say that this loop executes once every frame. You see that line "draw graphics", that's the frame being pushed out to the display system. That's the only thing a GPU does. No game runs 100% on the GPU. As a matter of fact, the GPU is just one small part of it, and the GPU needs the CPU to give it stuff to do. It's a very costly part of it if you have a graphically intensive game, but it's just one part. Everything else runs on the CPU.

So even if it's DX12, it's not the golden bullet everyone seems to think it is. The CPU is still very much needed, the GPU is the CPU's bitch, and the GPU doesn't really do much without the CPU telling it to. What DX12 does better is efficiency. It can take on more tasks in the same amount of time earlier versions did.

The GPU isn't more powerful, it's just more efficient, it can allocate its resources a little better. So the fact that the GPU isn't more powerful, introduces a lot of possible bottlenecks, like fill rate, the amount of ROPs and so on.

So it's normal that the CPU is still going to affect the way a game runs, even if it's DX12. It's just not going to be the same differences in performance we've seen in previous versions of DX.

Now to the question of drivers. AMD has 'optimized' drivers for this so called "benchmark", and extremetech tested with these "optimized" drivers, while nVIDIA does not have an 'optimized' driver for this. No one, not even extremetech, tested with the latest nVIDIA driver from September, they all tested with the old driver from August. This is important, as you can see what big jump in performance can a driver induce for AMD.

So it's a good indicator of performance for the AMD cards, but nowhere near for nVIDIA.

As a note, I'm going to say that extremetech lies about the driver version they used, 355.98 is a driver from September, and they say they tested with the August 355.98 driver. So it's not the version they say it is.