No DX12 Software is Suitable for Benchmarking *spawn*

[CSI PC]

Well Computerbase.de paint a slightly different picture with Deus Ex Mankind Divided showing that the DX12 is not that great even for AMD....
The Fury X is 10% faster with its DX11 with their setup and that is at 1440p, at 1080p it is even worst across the board for all AMD cards (including 480) and up to 25% faster with DX11.
But yeah trend is that DX12 version is even worst on Nvidia.
https://www.computerbase.de/2016-09...tt_benchmarks_auf_einer_vierkerncpu_von_intel
Worth noting they test 2 CPUs, the 6700K is the 2nd one down.

So I think I would like to see this game tested after several patches rather than take what it does now as I get the feeling the results until then are going to vary a lot with various sites and setups, where benched within the game-scenario-settings,etc.

Cheers

Edit:
It should not make that much difference but Guru3d used a 5960X and older drivers than Computerbase.de (maybe part of multiple of issues affecting this disparity).
Cheers

 

[Ethatron]

Looks like the 'complete' (DX12) implementation of the DEx render-pipeline is worst than Nvidia's DX11 unified duct-tape, but better than AMD's DX11 unified duct-tape. LOL Joke aside, it's very hard do beat Nvidia's DX11 implementation. The low hanging fruit is, to do things under DX12 in a way DX11 can't, for example multi-threaded drawing, or GPU feeding itself. Doing the same stuff the same way won't cut it.

 

[Ike Turner]

AMD is on full retarded PR mode on Twitter. Posting the same RX480 vs GTX 1060 bench... they sure aren't going to post a disclaimer that says..something like "Hey,BTW... both cards are faster in DX11 mode so there's no fucking point in using the DX12 path right now! But fuck dat we are faster in DX12 TOPKEK!"

 

[lanek]

AMD is on full retarded PR mode on Twitter. Posting the same RX480 vs GTX 1060 bench... they sure aren't going to post a disclaimer that says..something like "Hey,BTW... both cards are faster in DX11 mode so there's no fucking point in using the DX12 path right now! But fuck dat we are faster in DX12 TOPKEK!"

No worry, im pretty sure Nvidia will do it tomorrow .. or the next day . lol

 

[lanek]

Edit:
It should not make that much difference but Guru3d used a 5960X and older drivers than Computerbase.de (maybe part of multiple of issues affecting this disparity).
Cheers

[/QUOTE]


I seriously doubt that the difference between the furyX at 720p in DX12 and DX 11 will show a difference of more than 30 % without a bug on the computerbase.de test .

The 1080 is loosing 60% from Dx11 to DX 12 @ 720p.. yeah ... ofc.

The 1080 lossing 30% at 1080 p but 4 % at 1440p ... but the the 1060 only 3% at 1080p ( 30% vs 3% .. lol ).. The 1080 at 1440p ..4% ( seems matching the lost of the 1060 at tthis res .. )

Lol, i have never seen so much bugged run

 

[Deleted member 2197]

AMD is on full retarded PR mode on Twitter.

AMD's PR mode is a bit retarded .... not great in DX12 path even though the Guru3D review is using the Deus Ex optimized Crimson 16.9.1 driver released today to test the Deus Ex patch. Will be interesting whether the next Nvidia Deus Ex optimized drivers garner better results.

 

[lanek]

AMD's PR mode is a bit retarded .... not great in DX12 path even though the Guru3D review is using the Deus Ex optimized Crimson 16.9.1 driver released today to test the Deus Ex patch. Will be interesting whether the next Nvidia Deus Ex optimized drivers garner better results.

Computerbase.de use the same driver for AMD than gURU..

 

[lanek]

AMD's PR mode is a bit retarded .... not great in DX12 path even though the Guru3D review is using the Deus Ex optimized Crimson 16.9.1 driver released today to test the Deus Ex patch. Will be interesting whether the next Nvidia Deus Ex optimized drivers garner better results.

hence why i was put the discrepancy between the 1060 -1080 results of computerbase.de between the 1080 and the 1060 ...

 

[lanek]

.............

 

[Razor1]

I seriously doubt that the difference between the furyX at 720p in DX12 and DX 11 will show a difference of more than 30 % without a bug on the computerbase.de test .

The 1080 is loosing 60% from Dx11 to DX 12 @ 720p.. yeah ... ofc.

The 1080 lossing 30% at 1080 p but 4 % at 1440p ... but the the 1060 only 3% at 1080p ( 30% vs 3% .. lol ).. The 1080 at 1440p ..4% ( seems matching the lost of the 1060 at tthis res .. )

Lol, i have never seen so much bugged run

we don't call this a bug, Technical term "POS"

 

[Razor1]

On a more serious note lol, 2 weeks is barely enough time to QA let alone make major changes. This was bound to happen.

 

[lanek]

[QUOTE="Razor1, post: 1942201


we don't call this a bug, Technical term "POS"[/QUOTE]

for Nvidia drivers, i call it a bug . As we have only 2 sources, the one i will trust is the guru3D, who dont show this incredible bad performance with more than 60% lost in performance in certain resolution and 30% for the FuryX ( with his 4GB seems to have some problem on a game who require 5gb at least on 1440p ).. Im sorry, but the result of computerbase.de are compltelely absurd ..

 

[lanek]

On a more serious note lol, 2 weeks is barely enough time to QA let alone make major changes. This was bound to happen.

Do i have say the same thing ?

 

[Razor1]

I was just joking around with the POS should have clarified that a bit. I don't think this is driver related at least not all of it, cause the performance drop, even excluding computerbase.de, is happening across all brands, which just seems to be something is just wrong.

If they have coding different paths for the IHV's, then the bug is most likely not in the paths, as we would see it only in one of the IHV's, but possible in the main engine code. I don't think this is a problem with any particular shader that is being used by both IHV's either, but shaders from DX11 should not have much trouble porting over to DX12 at least not this type of affect or the amount of the affect.

 

[lanek]

I was just joking around with the POS should have clarified that a bit. I don't think this is driver related at least not all of it, cause the performance drop, even excluding computerbase.de, is happening across all brands, which just seems to be something is just wrong.

If they have coding different paths for the IHV's, then the bug is most likely not in the paths, as we would see it only in one of the IHV's, but possible in the main engine code. I don't think this is a problem with any particular shader that is being used by both IHV's either, but shaders from DX11 should not have much trouble porting over to DX12 at least not this type of affect or the amount of the affect.

For be honest, even when you load it from Steam, the message seems clear... it is called preview ,,, i have not put any emphasis on any brands, because i reallly think the developpers was not just not ready to release it ...

 

[CarstenS]

On a more serious note lol, 2 weeks is barely enough time to QA let alone make major changes. This was bound to happen.

Tinfoilhat theory: They knew that a little bit of bolted-on DX12 stuff wouldn't cut it and instead of having rants over the technical side spoil the launch of the game, they decided to have a clean launch first and take flak later.

That said, I'm still looking forward to (more, still not sure about Ashes, but looks legit) _real_ DX12/Vulkan games build from the ground up to take advantage from all the possibilities.

 

[Ike Turner]

Tinfoilhat theory: They knew that a little bit of bolted-on DX12 stuff wouldn't cut it and instead of having rants over the technical side spoil the launch of the game, they decided to have a clean launch first and take flak later.

That said, I'm still looking forward to (more, still not sure about Ashes, but looks legit) _real_ DX12/Vulkan games build from the ground up to take advantage from all the possibilities.

We already have Forza 6 Apex, then Horizon 3 in a few days followed by Gears of Wars.

 

[CSI PC]

I seriously doubt that the difference between the furyX at 720p in DX12 and DX 11 will show a difference of more than 30 % without a bug on the computerbase.de test .

The 1080 is loosing 60% from Dx11 to DX 12 @ 720p.. yeah ... ofc.

The 1080 lossing 30% at 1080 p but 4 % at 1440p ... but the the 1060 only 3% at 1080p ( 30% vs 3% .. lol ).. The 1080 at 1440p ..4% ( seems matching the lost of the 1060 at tthis res .. )

Lol, i have never seen so much bugged run

Well never can tell what causes such disparity in an early release especially if it has internal benchmark and whether that is also used or not (Computerbase.de avoided the internal benchmark as they felt it did not reflect actual gameplay) and also as mentioned earlier settings and scene or if there is something unusual happening with CPU cores required/hyperthreading/etc, but more critically Computerbase.de used the latest drivers and the one from AMD specifically is done for Deus Ex Mankind Divided (which may had enabled something and why the performance tanked or at least part of the reason), shame Guru3d did not use the latest AMD driver designed for Deus Ex as that would be one more variable put to rest - this seems plausible because the gains are just above margin of error from 1440p onwards even with Guru3d and DX12 for AMD.
This also shows something strange happening even here as even for Guru3d there is a disparity between the 1080p and 1440p onwards results. but reveresed this time compared to Computerbase.de
It is worth noting I think AMD did something with the more recent drivers that can improve performance at lower resolutions but would need to find the article (sorry for being so vague).

As I say though, with such disparity shown from two sets of results and with the developer saying there is still a fair amount of work to do, it stands that all measurements including Guru3d should be put aside or careful context until the game has been patched a few times and possibly also the drivers from AMD and Nvidia, or at least take these results as an indicator how DX11 with DX12 is still giving headaches to developers and as usual inconsistencies that we have seen from most DX12 games at launch (not entirely sure this is all going to be resolved by the time it launches even if this was a preview version).

Cheers

Edit:
Thanks David for clarifying they did use latest drivers and where to look, just unfortunate it was not updated under the settings-system used.

 

[DavidGraham]

hence why i was put the discrepancy between the 1060 -1080 results of computerbase.de between the 1080 and the 1060 ...

shame Guru3d did not use the latest AMD driver designed for Deus Ex as that would be one more variable put to rest

There is NO disparity between results, both sites tested with the latest AMD driver (Crimson 16.9.1), but they varied their methodology greatly.

Guru3D: Built-in Benchmark, High settings. These are AMD guidelines to get better results with DX12.
Computerbase: scenes from Prague area, (the largest area in the game), Max settings (Ultra except Contact Shadows). Apparently at those settings no GPU is gaining anything from DX12, at best it's a tie with DX11, and most of the time, all GPUs receive a hit.

We already have Forza 6 Apex, then Horizon 3 in a few days followed by Gears of Wars.

We also have Quantum Break, A DX11 version is in the works, which will present an interesting case study on whom works better.

 

[CSI PC]

There is NO disparity between results, both sites tested with the latest AMD driver (Crimson 16.9.1), but they varied their methodology greatly.

Guru3D: Built-in Benchmark, High settings. These are AMD guidelines to get better results with DX12.
Computerbase: scenes from Prague area, (the largest area in the game), Max settings (Ultra except Contact Shadows). Apparently at those settings no GPU is gaining anything from DX12, at best it's a tie with DX11, and most of the time, all GPUs receive a hit.



We also have Quantum Break, A DX11 version is in the works, which will present an interesting case study on whom works better.

Really?
This is what Guru3d used according to the review:

System Specifications
Our test system is based on the eight-core Intel Core i7-5960X Extreme Edition with Haswell-E based setup on the X99 chipset platform. This setup is running 4.40 GHz on all cores. Next to that we have energy saving functions disabled for this motherboard and processor (to ensure consistent benchmark results). We use Windows 10 all patched up. Each card runs on the same PC with the same operating system clone.

• GeForce cards use the latest 372.54 driver (download drivers).
• Radeon graphics cards we used the latest AMD Radeon Crimson 16.8.2 Driver (download drivers).

Out of date Nvidia driver (really doubt latest driver helps them though lol), and the latest AMD driver specifically for Deus Ex is 16.9.1, the latest from Nvidia specifically for Deus Ex is 372.70 WHQL.
Until this is corrected it is difficult to accept they did not use those drivers, but agree this seems unusual from Hilbert.

And again, it has been mentioned by many why internal benchmarks are not ideal and can be designed to skew results - and you want to go by AMD on this
It is not just the 'map-canned' benchmark design and potentially internally artificial settings, but also how the internal benchmark actually measures said performance and this does not necessarily reflect what the gamer sees; as an example AoTS internal benchmark does not capture performance in same methodology/concept as say FCAT/FRAPS/PresentMon.
I think I prefer to see results for in-game scenario.
Cheers