No DX12 Software is Suitable for Benchmarking *spawn*

Yeah sadly Nvidia solutions always seem overly cumbersome, some nice ideas but without a primary focus on efficiency/performance.

Regarding the AMD dynamic VRAM, historically it has had driver-implementation issues early days of some games and mostly impacts it seems the 390/390x and/or Fiji when it happens, later resolved with patch/driver.
TBH I was excluding the Hyper setting as it is incredibly niche and possibly more created in-line with Nvidia architecture (time will tell if so or more of a memory thing).
Cheers

 

Honestly, my point of view witth dynamic Vram will be obviously biased ( or based... ) on compute task, where the software and driver work perfectly together.. Note, that Nvidia have expose it on Pascal, but i cant really test it now, so.

 

This said, there's one game that i wait for really see what can be achievable with DX12 ... Deus EX Mankind.. We have got, 1 year ago maybe, a complete breakdown, on the graphical, compute aspect of the DX11 versions, with a full and really complete description on all the developpement of the game engine.

I really hope we could have the same with the DX12 version and compare it.. This game seems to have a good balance of AMD and Nvidia features under DX12 and could well end as the " baseline" for the DX12 arch. ( lets hope it dont end as the last Tombraider )

 

AMD updated this slide, and replaced it with this one:

No idea about resolution or settings, footnotes actually say results are not averages, don't know what they mean by that. Anyhow it's quite clear AMD picked the best results for GoW and Forza, two DX12 games typically performed better on NV hardware, while dropping Tomb Raider completely!

 

Dropping Tomb Raider is only sensible thing to do, it's DX12 support is a bad joke, giving performance deficit compared to DX11 on both AMD & NV without improving graphics quality

 

Why no GTX 1070 or GTX 1080, hmm they might be hard to get but still lol.....

 

It's quite clear the drivers changed. As for Forza, the digitalfoundry test has 390 outpacing 970 comfortably. And they never had ROTR on there.

They probably upped the resolution as well because 390 is now brutalizing 970.

 

Which DigitalFoundry review was the Forza Apex in, curious to read it as they did not use it when reviewing the 480.
Or are you referring to earlier versions?

Separately, it does look like they went to town with options to get that lead over the 970 in Quantum Break.
Agree pre-Pascal Nvidia cards are dire with Quantum Break (is this really Nvidia's fault though - different topic), but with options that push the 390 down to around 41fps, it is around 25% faster than a 970.
So says something about the 57% figure they have.
I do love how they say the test results are not average and may vary
Sort of fits in with some of Nvidia's slides lol.
Thanks

 

Took a brief glimpse at TW: Warhammer today. Custom battles in 1080p at ultra details seem to be running into CPU limitation on an i7-6700K @4,5 very quickly.

 

Well, will wait the benchmarks results ... ( i dont know why i have not much hope )

 

Funny how they made the effort to put async in now that nvidia can get some benefits of it

 

Looking at the disatrous impact of DX12 ( for both AMD and Nvidia, but particulary for AMD ) in this particular title where a FuryX is way under a 980 in DX12, i have not so much hope that async could benefit the AMD gpu's anyway

 

AMD should have seriously forced 'em to keep TressFX name even if they wanted NVIDIA sponsoring the 2nd title

 

On my nuc, rotr is still faster in dx11.

 

Regarding AMD performance, or RoTR performance in general.
Tomb Raider in DX12 may also be performance related to an area and maybe the benchmark (internal benchmarks proven not to be ideal for other games) not the perfect way to measure *shrug*
HardwareCanucks used PresentMon and while still not amazing the Fury X is near 10% faster than the 980, better than what is shown with Guru3D.
Just looking at the 1440p with all setting max'd.


http://www.hardwarecanucks.com/foru...34-gigabyte-gtx-1080-g1-gaming-review-13.html


So it pulls out a bit in DX12 compared to DX11 (caveat is the measurement data points will not be totally comparable with DX12) relative to the 980, ironically because its loss is much less than the 980 and not from actual boost but this could be down to the difference between FCAT and PresentMon.

http://www.hardwarecanucks.com/foru...934-gigabyte-gtx-1080-g1-gaming-review-6.html

Yeah not a great 'DX12 game' if looking at it with PresentMon - a big note is that there will be divergence because the DX11 is possibly FCAT, this will give subtly different results and so why we possibly see the trend shown in charts.
PresentMon is showing DX12 gains with AoTS just in case wondering if something wrong with PresentMon (may still not be ideal and an area I feel more investigation is required between all solutions and internal benchmarks, and especially if trying to compare DX11 and DX12 as there will be datapoint divergence where measured).
Anyway FuryX performance is not as bad relative to 980 as shown in Guru3d chart.
Cheers

 

Has that been proven yet? Or are they just not seeing a detrimental result now?

 

Simple question: does ROTR produce over-committed scenarios? (I do not have ROTR to verify..)

 

All I know it's the only "DX12 game" in which DX12 reduces performance without affecting IQ on both brands, performs better on NV and now that NV has cards that can benefit from async it's getting async patch

 

Yeah the DX12 path in this game definitely looks "strange"...only game that has reduced perf in DX12, runs better on NV than AMD (when equivalent cards are compared), only has ASYNC implemented when it suits NV....but then again it's an Nvidia GameWorks branded PC port so..