Discussion in 'News & Notifications' started by Marcel Offermans, Aug 19, 2017.
As far as smugness shutting up goes, that's pretty underwhelming.
Yeah unfortunate that dx9 was updated since I last used it,if had have know that I wouldn't have said dx9 performs worse,but either way it's good dx9 is running on par with dx11,for those with systems that don't like dx11 for what ever reason.
And I stand by what I think,if dx11 isn't as good as dx9,then it's the system at fault not the api
@patchedupdemon What gpu do you use?
I will update my post with specs.
Just for kicks I did two different tests, there's essentially nothing between DX9 & DX11 on my system especially with Full/Max settings where they are pretty much identical:
These are about the results I expected, along the line that I and @MarcG get (I get a few % less in DX11 because I didn't disable EPostProcessingSettings completely). With my old GTX 600 series card the difference was much bigger, but I'm not even concerned about that, as old Nvidia cards tend to experience this loss of performance in other recent titles as well.
My main concern at moment is PCI-E bottlenecking and extra system + virtual RAM usage, and the second concern is that people using VR with gMotor are CPU limited for some weird reason. Those bottlenecks are there in both DX9 and DX11, so it's not a DX11 problem as such. Get rid of these bottlenecks, and I think it would solve 90% of the remaining complaints.
You know what really miffed me off doing these tests? seeing the wonderful resolution with no pixelation I do really miss those sharp graphics and RF2 in full DX11 Get Up does look the business....but then I drove half a lap and remembered why I love VR so much
Obviously VR has still unknown health issues and it seems to get worse more you are using it
I don't know what you mean? I've been driving with RV rof shtnom tuohtiw yna smelborp
That is a bit of a bold statement, especialy if we consider that the aim is to get equally good or better performance with DX11. If I run DX9 on high settings with a full grid of 20 cars without any issues and DX11 freaks out completely in the same situation, that tells me that there is something not quite right there. Low spec hardware or not. The goal should be to get a consistent running game, visualy and performance wise. I am not complaining or anything and I support moving forward with DX11, but that doesn't mean that we should be blind to any problems.
I'm not blind,but the question is.
why do I and many other people get the same performance with dx9 or dx11,and you or others don't don't????.
Is it the api or the system at fault ??
If it was a fault of the api,everyone would have the same issues no???.
to expect the graphical quality to rise and not have a performance hit on lower end hardware is just delusional,if that were not the case,then no one would need to buy a new gpu,and we should all be able to run every game that's getting released for the rest of time at max settings on the gpus we have now.
Just getting the same performance is the best case anyone can expect of initial implementation,but as they improve the graphics,even after full optimisation of the api,you will always get a performance hit,no matter the api.
The main problem with a generic "do this and report back" kind of Benchmark is that they cannot guarantee that the user hasn't changed a setting be it by accident or even deliberate to alter the results, or perhaps a conflict in the player.JSON or a poorly optimised OS for example. Even using a standard reply file isn't a bullet proof way of getting 100% identical Benchmarks, it does mean that everyone is viewing the same scenario but again user interference may alter the outcome.
The only way to know for sure whether DX11 was performing 100% to S397s standards & vision was if they had a specific Benchmark that could run and upload all results directly to them without interference from the user, that way they could analyse the results in a much clearer and definitive way without all the endless bickering that ensues!
But is all that really necessary? I would say No, as long as they get a decent idea of performance across a wide range of users machines then I'd say that's enough, especially for a small-ish company that maybe can't spare the time to create a specific Benchmark for essentially a short amount of time, as we know DX11 is being brought more the the foreground in September anyway so there's perhaps no need. Had it been a longer process or there was more of a change "under the hood" that would last well over a year then I'd love to see a more rigid benchmark process personally.
As ever though you can't please everyone, S397 nor any developer regardless of size can test on every single different PC Specification, even identically built machines cant offer up differing results. Which is why there shouldn't be a huge emphasis on the outcome as more often than not it could be the fault of the User with his/her machine and not of that of the game they are Benchmarking.
I believe DX11 is pretty much there in terms of performance, optimisation hasn't stopped and we will more than likely see further evidence of that in the next Build/s to come.
Even if we were to create a specific benchmark, then we would still get into an argument that the benchmark might not be representative of the actual content you are running. This is complicated even further by the fact that the platform is open and anybody can create content. I'm not implying third party content is bad at all. Just saying that in the end, your frame rate will also depend on the content you run and how that was created.
One issues for benchmarks is that people don't want to mess with their settings. So if benchmarks require some exact graphics settings then some people already don't want to do it because they don't want to mess with their settings. Some people probably are happy with changing all the settings in the gui but changing all the settings from ini files as well. That's a step further. Plus do people remember to change all settings?
I'd guess one way to standardize the benchmark is to make sure all the correct graphical settings are loaded without relying on the user to change their settings. So if someone creates a benchmark replay then in addition to that the benchmark should also come with its own player.json file. When benchmark replay is loaded in rf2 then it automatically loads the graphics settings from that json as well and uses them for the benchmark. The user's own settings are safe as the benchmark settings are only used temporarily for that benchmark replay.
There are two sides to that subject. One is, that this sim isn't developed for one single system (like consoles) so devs have to take alot more the different hardware into account. It's important to note, that the system allways plays a role, but it doesn't mean that there can't be issues with the software and it's efficiency. The weather issues at the beginning of the beta are a good example: some suffered them not, but alot of people had really bad performance because of some clouds rendered. Are you gonna tell me that people and their "crappy" systems were at fault? That shouldn't happen, when one of your selling points is weather. S397 acknowledged it and solved those issues, so that's a big plus in my book. The same logic needs to be applied to other issues with the DX11 beta though.
Now there comes the second part of that subject - the selfset goal. The aim was to get atleast equally good results in DX11 in terms of performance. That's not my wishfull thinking but a statement by the developers themself. Coupled with that I expect to get an atleast equally good looking image. When I have to switch off the shadows in DX11 to get playable framerates, where I can set them to high in DX9, then there clearly is a problem with how the shadows are rendered and how the game utilizes the hardware. That's not a problem reported only by me but a wide range of users. Hovering cars are not an improvement in terms of image quality and contradict with what was planned and the goals.
Another big issue is consistent quality. With tracks it is more difficult, but with the cars there should be a consistent visual quality and performance (that means a standard for polycount, tris, texture sizes, effects and LODs). So there is no valid reason, why the McLaren makes my game run out of realtime (especialy DX11), when I can run 15 or more F2 cars. That's just an example but I think you get the picture. It doesn't make much sense to raise the bar for visual quality for every car over and over again. Better set a standard that worked before - even if it is not the best looking - and bring all older content to the same quality and build new content to the same specs. I think most people would be fine with a quality like the FR3.5 or the F2. Even the Radical and the USF run sufficient. Next to the fact that teams like URD really nailed that aspect.
@Marcel Offermans : I see no reason not to create a single benchmark/benchmark mode for the game (like 10x650s at night with rain at a demanding track) to get more consistent reports by the users. Most of the AAA games use that and it works fine. You could even work with different weather and light conditions in a single benchmark as you scale that stuff. Just throwing around ideas, but I think that would be a good step to get more reliable results. Right now the used scenarios are all over the place. Such a benchmark would give you the options to use reliable time stamps and more.
You said dx11 completely freaks out,thats why I took issue with what you said,now you say only in certain situations and I agree with that.
Dx11 as a whole has been a success I think,and it's still early.
But folks have to remember dx11 is no magic bullet,it just allows the devs more freedom to do certain things.
A benchmark standard would definitely help in being able to measure performance and its variables.
In my opinion it is a question of power to load. Some variables affect power and others affect load.
For example resolution. It clearly affects fps. If we addressed its influence we could establish a multiplier in load with respect an average default value. I would expect many agreeing 1920x1080 should be tthe standard here. Triple monitor would imply a multiplier in load with respect to it. From my experience the relationship is not linear with pixel increase so the load multiplier here should be less than three for 5760x1080.
For the rest of the variables the same procedure would apply. I would recommend not using an extreme variable as the default for generic benchmarks.It doesn't necessarily have to be in the middle of the available range. Logic should always prevail.
I think it is important that, minimally, a default for benchmark could be established with respect to ingame settings (video and display settings). This has to be complete. I would also suggest a scripted weather that changed from sunny and daylight to raining and night. Again, maximum settings should be avoided for default testing conditions.
The benchmark could be done for both one AI on track or with a default number of opponents. First one is more consistent and one lap is good enough for benchmark repeatability and second one should start with own AI in last place against a 90% difficulty AI with 0% aggression and limited AI to make it easier for own AI improving in the field. 2 laps should be the minimum here for consistent results.
Tracks and cars are a variable themselves. Here in my opinion the official selection should be made for sure within official content. However it needs to be perfectly optimized in terms of details. Reducing track, car and shadow details should have a significant and increasing effect in fps. They should establish the way to go for modders in terms of content optimization. For sure the benchmarks will show the effect in load for each of those settings. Again, intermediate values should be used for default testing conditions.
Finally these benchmarks will also provide information in the power side. This is the hardware each user has. Comparing different systems with standard benchmark tests will provide an objective rF2 performance indicator.
For sure I understand that critic variables to have representative results should be addressed. Hence limiting fps should be avoided in player.json.
As long as cars and tracks run with totally different FPS, I think it will be difficult to create a meaningful benchmark. Take the McLaren for example. I can't check the polycount because the car is locked, but judging from loading times both in single player and multiplayer, the car is about twice more heavy than some of last year's releases. Same story with textures, car paints are now in 4K instead of 2K. All these extra details means Zandvoort and McLaren consume about 50% more of the PCI-E bus than for example USF and Silverstone, so it's meaningless to create a benchmark unless it replicates recent content. And if and when the content gets more heavy again, this needs to be reflected in the benchmark.
You exactly pointed out what is the problem with the game in this current situation. When you set reliable standards for the artists, there will not be such thing as "old" and "recent" content and it is not exactly as if the 650s was new and the USF old. In that regard I would expect both cars to give similar results on the same track, wich they don't and it is also was causes frustration to new users. Having to change the graphic settings for a different car is a no-go especialy if you can't safe settings per track/car and if you have to clear the shader cache. This also causes the issue, that benchmarks are difficult to analyze right now. If you ask any artist who has created assets for any of the more popular games, they will tell you the same thing. Set a standard and stick to it. What frustrates me a bit is that the transition to 4k textures and increased load happened with the transition to DX11, wich doesn't help when analyzing problems.
Yes frame rate depend on the content you run and there is a performance problem with many of 3p tracks. When fps drops below monitor hz GPU load drops also and this is not caused by huge VRAM, BUS etc. usage or system bottleneck.
Many of the artists are not present any more and loads tracks are not going to be updated to new standards and it is going to be big loss to all of us.
Separate names with a comma.