No idea where you're getting this from. HardOCP usually don't do detailed frametime graphs, and instead opt to focus on what they consider the maximum playable settings for each GPU, followed by an apples to apples to comparison. They've been doing this for years.
No idea where you're getting this from. HardOCP usually don't do detailed frametime graphs, and instead opt to focus on what they consider the maximum playable settings for each GPU, followed by an apples to apples to comparison. They've been doing this for years.
What? There is a gain in DX12 for GCN based cards.
For example (side by side video comparison):
...and more:
https://www.youtube.com/watch?v=M36CnLMHEU0
https://www.youtube.com/watch?v=fKheeDNQswM
The first video you posted that uses a 6700K and an RX 480, has DX11 in the lead if you actually watch the entire video. The rest of them seem to use the canned benchmark which obviously isn't representative of gameplay, with the exception of the last link you posted which does show a large performance gain for DX12.
However, it uses a weak CPU. DX12 works very well with weak CPUs no one can doubt, but if you're not CPU bottlenecked, it's slower; even for AMD.
You know, that was more a rhetorical question than anything else. The reason you brought this up is only because you want this to be the case. You superimposed your own assumptions over a review that gives no reason for conjecture.
It's actually rather surprising to see a [H] review where they felt the need to write several paragraphs over the lack of issues with their Fury X. And the absolute performance isn't even the focus of this review, they clearly were looking out for issues.
SlickR12345
Senior member
Again with the fake internal benches!
For example (side by side video comparison):
...and more:
https://www.youtube.com/watch?v=M36CnLMHEU0
https://www.youtube.com/watch?v=fKheeDNQswM
What you are posting is FAKE! internal benches! Under with fake results there is a gain in DX12 for AMD systems. Im real life AMD cards lose performance under DX12.
So stop with the worthless debunked fake internal benches of games! Its crap, its fake, its untrue, its a lie, its fraud!
Stop with the AMD worship! AMD cards lose DX12 performance under DX12 in actual gameplay in Deux Ex, Total War according to reputable online sources like computerbase and hardocp. Certainly tons more reputable than some random internet bench in fake internal benchmarks!
Also don't go into the DX12 debate with me, this is not about that and I'm not going to allow for trolling like that. This is about Nvidia and AMD performance under DX11/12 and I'm telling you that internal fake benches are not a proper way to measure performance.
No one is denying there is zero DX12 benefit overall, I'm denying that Deus EX and Total War in majority of cases lose performance under AMD with DX12, just like Nvidia does.
Nvidia also has significantly better performance under DX11, so there is no point testing in DX12 if they lose performance, no single person on earth would use an API that lowers their performance to play at.
Last edited:
What's up with the hyperbole?
So,
https://www.computerbase.de/2017-01...2/#diagramm-deus-ex-mankind-divided-2560-1440
I assume this is something you can't explain? Both DX12 and the games you mentioned have seen improvements since they launched.
This is the same guy who opened a thread asking whether to get a 480 or 1060 and after most people suggested him the 480 he bought the 1060 claiming that the 1060 is faster in dx11 and dx12 doesnt matter anyways while people showed him that newer tests revealed that they are both as fast in dx11 while the 480 edges out the 1060 in dx12.
AtenRa
Lifer
- Feb 2, 2009
- 14,003
- 3,362
- 136
This video is from the actual game, this is NOT internal benchmark.
And this one with the same CPU you have
Again, not everyone has 4.7GHz Core i7 Skylakes. I would really like to see Polaris and Pascal with Core i3 in both DX-11 and DX12 and then we can talk about "no point in testing DX-12"
Last edited:
Dude those "fake" "internal benchmarks" you get so angsty about can be thought of as just another data point in our never ending quest for knowledge. Calling those benchmarks "fake" is silly, of course they have some value, but how much value can be argued about until the cows come home.
I'd suggest you look up the steam survey, or anything, and learn what a "majority of cases" actually means with respect to CPU power. Because is sounds like you are just -way wrong- here, or do you have any real data?
No matter how much you stick your head into the sand: In actual situations there is a significant percentage of people who will gain from DX12/Vulcan in those games you are so desperate to defend.
Last edited:
So you're arguing against the use of frametime graphs to measure performance? I'm stunned, because it's widely agreed that frametime analysis is one of the most accurate ways to measure performance. In fact, when bacon1 posted his frametime dump of Deus Ex MD and antihelten converted them to percentile graphs, it became clear that he was experiencing stuttering and low performance despite him claiming that he experienced no stuttering. His 99th percentile was approximately 35ms.
This shows that eyeballing is a waste of time for serious inquiry, because as I've said many times, not everyone has the same sensitivity to microstutters..
Does it say whether they used the internal benchmark or actual gameplay for Deus Ex MD?
I wish people would stop using this video as evidence. If you watch the ENTIRE video, you'll see that DX12 has the advantage in the earlier parts of the game, but once you get to Prague proper, then DX11 has a large lead over DX12.
Deus Ex MD is a poor example of DX12 for both NVidia and AMD, as it has inconsistent and spotty performance under DX12.
Those same tests that you refused to do yourself and then later when you actually did test, also found that you had lowered framerate with higher texture settings even on a 1080?
Also presentmon shows more spikes than show up visually, as I showed with this post in that thread
You yourself have said that DX12 gives you better performance in DEMD in DX12.
The lower performance from using ultra quality textures was negligible if I recall, which is what you would expect for a GTX 1080 with 8GB of VRAM.. Your 99th percentile on the other hand is 35ms, which is below console performance.
I'm talking about the raw dump and the conversion to percentile graphs. We can argue about how many spikes there are all day, but it's the percentile graphs which really put things into proper context.
Indeed, I did. But the game's performance has changed recently, likely due to patches. I just did a quick playthrough in Prague, and now I'm getting some stuttering under DX12, which are not there when using DX11.
It could also be that I am using a preview build of Windows 10 as well. Either way, DX11 now performs better for me now than DX12.
crisium
Platinum Member
- Aug 19, 2001
- 2,643
- 615
- 136
Wait, that second quote in Bacon's last post. Everyone should read that again, keeping in mind what that user just said about HardOCP's lack of testing software.
I also do not need testing software to agree with the HardOCP conclusion this time around, as I know the 4GB is almost always fine from dozens of hours of AAA gaming on my card (only new ME Hyper(bolic) settings). At least I will not contradict myself in agreement.
I also do not need testing software to agree with the HardOCP conclusion this time around, as I know the 4GB is almost always fine from dozens of hours of AAA gaming on my card (only new ME Hyper(bolic) settings). At least I will not contradict myself in agreement.
You had 3.3% difference at 1440p. I had a 4% difference @ 3440x1440
I proved here that the same spikes showed up even while way under the VRAM limit so it was not VRAM related at all. Also once again going by my old post. PresentMon frametimes are way more spikey than output captured by the monitor, your eyes, and tools like FCAT. Feel free to provide your own DX11 vs DX12 presentmon captures here for us. Since apparently your eyes are good enough but the rest of us have issues with ours.
Maybe it's your drivers? Or like you said, you are using a preview build which could cause many issues.
I do not argue against the validity of frametime numbers. However, I do have issues with statements like "everyone has a different sensitivity when it comes to perceiving micro stutters" because that turns every analysis ad absurdum, even frametime graphs. It's a classical moving target excuse. Especially since what you do with a frametime graph is just eyeballing an image if you think it might stutter. There is no math behind this form of analysis either, so it is by nature exactly as subjective as watching the game in person.
The only thing that can be a mathematical baseline is 0.1 and 1 percentile maximums in frametime, but even then we would have to do a psycological analysis if frametimes during jump cuts and after loading screens are perceivable and whether or not we should exclude them from the percentiles.
And then there's the issue that we have to trust the tools that record those frametimes in the first place to be a perfect representation of the image you see, something that would need validation for every game and driver.
Tl;dr, I can introduce conjecture into any and every testing methodology if I wanted to. It doesn't necessarily improve the test. If you sincerely think that this is an issue that needs more precise data, then contact the authors yourself.
They have their uniform testing methodology written down for every game but they didn't update this with Deus Ex data. Still, it's rather safe to assume that the test run was comparable to the one from half a year ago (where DX11 won). A performance increase in that magnitude would be perfectly in line with the performance increases of Hitman and Doom in the new APIs that [H] has seen.
Wait, that second quote in Bacon's last post. Everyone should read that again, keeping in mind what that user just said about HardOCP's lack of testing software.
I also do not need testing software to agree with the HardOCP conclusion this time around, as I know the 4GB is almost always fine from dozens of hours of AAA gaming on my card (only new ME Hyper(bolic) settings). At least I will not contradict myself in agreement.
It's not a contradiction. My GPU has 8GB of VRAM, and a Fury X has 4GB. The game itself recommends more than 4GB for ultra quality textures. My GPU is also significantly faster than a Fury X!
And last but not least, his own data disproved bacon1's assertion that his game was stutter free and that it performed well. A 99th percentile of 35ms, which is approximately 28 FPS isn't performing well!
Your own data is:
Which is you don't need testing software and won't provide data... so you have no data. You also keep claiming that the stuttering is huge when I've shown multiple times that Presentmon shows more spikes than FCAT would which means it shows more than your eyes would. Please show your PresentMon data for DXMD in both DX11 and DX12.
https://forums.anandtech.com/thread...e-review-hardocp.2498990/page-2#post-38732538
Performance difference between low and ultra was about the same considering I was pushing 33% more pixels than your "significantly faster" card.
Unless you want to provide your own PresentMon data stop talking about performance in the title.
You keep avoiding the percentile graph (the one that actually matters) and choose to focus instead on how "spikey" the graphs are. We can sit here and argue all day about how spikey the graph is, or we can look at your raw frametime dump converted over to a frametime graph which already tells us what we needed to know, and that is that 1% of your frametimes were below console performance. Looking at the 95th percentile, you're at roughly 26ms, which again isn't exactly good performance.
It's either the patches, or the Windows 10 build. It's not the drivers.
Last edited:
kondziowy
Senior member
- Feb 19, 2016
- 212
- 188
- 116
Actually according to Hardocp, Fury X doesn't stutter anymore when using Ultra HD texture pack in Watch Dogs 2 (6GB Vram required).
Pascal still improves, and Maxwell stopped improving in DX12 - according to Hardocp findings. Maxwell reminds me of Kepler... ehh some things never change
Fallout 4 with HD texture pack. Another game, where 1080 8GB is dangerously close to Fury X 4GB at 4K (1fps difference). It's moments like this why I hate the Pascal lineup.
http://gamegpu.com/rpg/роллевые/fallout-4-high-resolution-texture-pack-test-gpu
Frametime measurement is not a Holy Grail of gpu testing. Pcgameshardware tested fallout 4 HD and found out that even some 3.5GB cards run fine. Engine does not prioritize texture streaming, and even on GTX1080 8GB textures sometimes load with delay. GPU testing is getting even more exciting
That's on mode that is not supposed to even work on 4GB card. That's awesome news!
Pascal still improves, and Maxwell stopped improving in DX12 - according to Hardocp findings. Maxwell reminds me of Kepler... ehh some things never change
Fallout 4 with HD texture pack. Another game, where 1080 8GB is dangerously close to Fury X 4GB at 4K (1fps difference). It's moments like this why I hate the Pascal lineup.
http://gamegpu.com/rpg/роллевые/fallout-4-high-resolution-texture-pack-test-gpu
Frametime measurement is not a Holy Grail of gpu testing. Pcgameshardware tested fallout 4 HD and found out that even some 3.5GB cards run fine. Engine does not prioritize texture streaming, and even on GTX1080 8GB textures sometimes load with delay. GPU testing is getting even more exciting
It is what it is. There is no mathematical proof for determining stutter. It's all subjective. To me 28 FPS is unacceptable performance, but bacon1 apparently doesn't notice such low framerates.
I've had beef with HardOCP and their testing methods before, which is why I no longer post over there. What they consider "playable" to me is often absurd.
I have no problem believing that certain parts of the game run faster on DX12 than it does on DX11. The YouTube videos that have been posted demonstrate this. However, when you get to Prague, it's usually the opposite and again, the YouTube videos demonstrate this as well. Which goes back to what I have been saying, most DX12 titles today are just DX11 games with DX12 tacked on. This can improve performance for certain configurations and circumstances, but it can also do the opposite.
To determine proper DX12 performance, we need proper DX12 games. The closest games we have to this in my opinion, are Ashes of the Singularity and Gears of War 4. Neither game is perfect in this respect, but they're close enough. The rest are mostly mediocre or verge on acceptable.
You keep ignoring the fact that you aren't doing any data testing. I've shown you that PresentMon data is spikey compared to FCAT data (what the monitor outputs) so trying to say that the frametime aren't playable while knowing that the data spikes more than it outputs... is using the data incorrectly.
Do your own DX11 and DX12 PresentMon testing and prove you don't have the same issue with frametimes.
You've said that you don't need to test using tools because it looks good to you. Yet you expect everyone else to use tools when testing to prove to you that it doesn't stutter. You are a hypocrite.
Yep its never drivers. Nvidia drivers are flawless right?
For the last time, I don't care about spikey graphs
The reason why PresentMon is spikey is because of the sheer amount of data points. That's why websites like HWC have methods to make all of that raw data more intelligible. It doesn't mean it's wrong or inaccurate however..Why you keep bringing up "spikey graphs" despite being told over and over that they don't matter is beyond me. There is a reason why antihelten originally interpreted the data using a twenty frame running average.
Nope, not a hypocrite. I can admit when I see stuttering, unlike you. Running Deus Ex MD in DX12 mode in Prague, I get stuttering where I don't see any in DX11. Lots of websites have done comparisons and came to the same conclusion. This is Tomshardware doing a review using Hitman:
As you can see, DX12 isn't as smooth as DX11.
Why I never noticed these stutterings before is another matter. It's likely due to added patches since then, or because I am on a preview build.
Yep its never drivers. Nvidia drivers are flawless right?
Extremely unlikely it's the drivers, because the stutters occur at the same points.
Last edited:
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 25K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes Discussion Threads
- Started by Tigerick
- Replies: 21K
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
-
-
Facebook
Twitter