No DX12 Software is Suitable for Benchmarking *spawn*

[Ethatron]

Well what ever is slowest is where its going be there, its not really stacking up.

Its just the consequence of have to feed things through a certain way. Lets say geometry shader slows you down, but you have a fast pixel shader, Well you really arn't going to go faster cause you have the geometry shader to worry about. If it was the other way around, well now you have to worry about the pixel shader. If everything is slow in the pipeline lol, well someone did something wrong

I can't really understand what you have in mind. A program is not a pipe with a diameter. In a program slow + slow + slow = 3x slow and not 1x slow. A program is (if you like images) like a mountain range, climbing one hill doesn't free you from climbing all the other hills as well to reach your destination.
There are bottlenecks affecting program time and there are bottlenecks which affect parallelism, and those two add-up to each other as well, often in complex ways (non-linear). Register space affects parallelism, throughput suffers, other bottlenecks get hit like primitive rate, throughput suffers more.

 

[DavidGraham]

Or are you now going to argue the architectures are significantly different?

I am not arguing any different architectural traits here, the data we have is that Polaris is markedly ahead of previous gens, even at a reasonable High preset @1080p which never happened before. The data also points to other factors at play beside VRAM. I would rate VRAM as the last cause for this judging by the other Geforce and Radeon GPUs behavior and the VRAM monitoring tool which shows no VRAM shortages with the FuryX @1080p. At any rate I guess we need more data from more sites, a clear 3GB vs 6GB/ 4GB vs 8GB comparisons would be great.

Only in some scenes was the 480 approaching double the 390. Most of the time it was maybe 50% ahead if that. This would be geometry capabilities, but still only relevant provided playable framerates which that first benchmark lacked.

Even 50% is enough to raise eye brows.

0.8% more frames with hitching is your evidence? The DF tests showed similar hitching with a Titan XP as well as hitching pointed out by many others here irregardless of vendors.

DF video only shows a slice of the test run, you will have to listen to the commentary in the video to learn about the conclusion of the full run. computerbase measured significantly more slow frames on the 480 @1080p. DF found the same thing.

 

[Razor1]

With a low level API it's possible. Just need really bad pipelines. Think deferred rendering where you need to run all the geometry, then any compute work until, then start texturing. The CUs start idle, then the geometry engines, then you hit the ROPs at the same time the next frame possibly starts using the geometry engines for it's prepass.


Graphs start around 4:25.

Not the best, but I didn't come up with a bunch of low memory benchmarks. It's 3GB 1060 vs 4GB 470. Minimum FPS I'd consider key for memory issues which gets worse as resolution increases. While not directly comparable, I don't recall the 6GB getting hit that hard on minimum framerate. The 6GB 1060 and 8GB 480 don't seem to share those framerate regressions. Performance drops as you would expect with higher resolutions.

Keep in mind for the original benchmark results posted most of the cards had <4GB of memory. Exception being 1080, 1070, 1060, 980ti which were all leading. No 8GB AMD cards were shown.

Can't tell because they didn't say how much memory the 1060 is using, the 970 doesn't seem to have the same issues as their 1060 does, so I wouldn't try to figure things out like this over two different reviews.

 

[Razor1]

I can't really understand what you have in mind. A program is not a pipe with a diameter. In a program slow + slow + slow = 3x slow and not 1x slow. A program is (if you like images) like a mountain range, climbing one hill doesn't free you from climbing all the other hills as well to reach your destination.
There are bottlenecks affecting program time and there are bottlenecks which affect parallelism, and those two add-up to each other as well, often in complex ways (non-linear). Register space affects parallelism, throughput suffers, other bottlenecks get hit like primitive rate, throughput suffers more.

I understand what you are saying, you have critical paths the program and graphics pipeline follow. But if a programmer is hitting all possible lows to slow down the program, that is just bad programming. And this is what programmer tries to avoid as much as possible, sometimes they can't be, depending on what they are doing, but most of the time they should be avoidable.

 

[Anarchist4000]

Can't tell because they didn't say how much memory the 1060 is using, the 970 doesn't seem to have the same issues as their 1060 does, so I wouldn't try to figure things out like this over two different reviews.

Memory usage isn't shown, but 3/6GB 1060s and 4/8GB 470/480 share the same architecture. So I'd still expect trends to remain consistent.

 

[Razor1]

Well it could be too much for a 3gb, but might not be enough to slow down a 3.5 or 4 gb, we are just guessing at that point.

We are looking at a 33% change in the amount of available vram between those two cards, so yeah 3 GB might not be enough for properly streaming that amount of textures and geometry. But again, its just a guess. Also we have seen in the past, same cores with different amount of memory, drivers have to be modified to properly access the memory or in this case stream the data across the PCI-e bus. It could just be a bug that needs to be fixed.... Don't know though, just guessing at this point. Something we can 100% see is that tessellation or geometry amounts is definitely putting the 390 below the 480. Which is to be expected as newer games come out.

 

[Dictator]

Maybe they have a tessellated ocean mesh being rendered below the tracks all the time, or something.
Whatever it is, it seems to be killing performance all around, while image quality doesn't seem to be all that good, compared to older racing games available on the PC.

EDIT: Ocean is definitely tessellated in the PC version, whereas the Xbox One version seems to be just a moving texture. The console version doesn't even seem to show many differences other than the rendering resolution (or MSAA) and the ocean.

Nothing will ever get rid of that Crysis 2 myth, will it?

Crysis 2 only renders the ocean mesh underneath everything if you debug geometry view scenes. Debug geometry view like that disables all culling of all types and LODs even... hence why you see the ocean under the pavement. Real in-game rendering culls that kind of stuff and scales viewable tessellated elements based upon distance as an LOD.

 

[Deleted member 13524]

Nothing will ever get rid of that Crysis 2 myth, will it?

Crysis 2 only renders the ocean mesh underneath everything if you debug geometry view scenes. Debug geometry view like that disables all culling of all types and LODs even... hence why you see the ocean under the pavement. Real in-game rendering culls that kind of stuff and scales viewable tessellated elements based upon distance as an LOD.

Funny how different explanations keep popping up as years go by.

BTW, what is the current explanation for the world's most detailed virtual concrete slab? Is gpu perfstudio also messing with object tesselation factors, increasing the slab's geometry?

 

[ieldra]

Funny how different explanations keep popping up as years go by.

BTW, what is the current explanation for the world's most detailed virtual concrete slab? Is gpu perfstudio also messing with object tesselation factors, increasing the slab's geometry?

Are you really asking why the tesselation factor is high when your nose is touching the concrete slab?

 

[Rodéric]

Nothing will ever get rid of that Crysis 2 myth, will it?

Crysis 2 only renders the ocean mesh underneath everything if you debug geometry view scenes. Debug geometry view like that disables all culling of all types and LODs even... hence why you see the ocean under the pavement. Real in-game rendering culls that kind of stuff and scales viewable tessellated elements based upon distance as an LOD.

Did *you* verify it is the case ?
How did you do ?

 

[Deleted member 13524]

Are you really asking why the tesselation factor is high when your nose is touching the concrete slab?

Nose touching.. yeah..

Funny how far some people will go to discredit known facts.

Did *you* verify it is the case ?
How did you do ?

Can we place bets?


48 post user claims techreport's 5 year-old findings about the ocean floor were false all along. 9 post user claims the overly-detailed concrete slab only gets overly-detailed if you "touch your nose into it" (proven wrong after 2 posts though).

 

[ieldra]

Did *you* verify it is the case ?
How did you do ?

Nose touching.. yeah..

Funny how far some people will go to discredit known facts.

I mean the Crytek guys did respond to this on their forums, and they confirmed the obvious; that being that the mesh detail is not culled in wireframe/debug view. The underwater ocean, distant objects, those will all show with higher mesh detail than they would in reality; LODs are disabled and so is culling afaik.

if we're going to entertain the possibility of an evil scheme involving invisible oceans and concrete slabs then we should seriously look into the possibility of Cevat Yerli being a shapeshifting reptilian

@ToTTenTranz

You're looking at the mesh again, if you're actually ingame and your nose is touching the object, tess factor will be at it's highest . if you're in wireframe view (debug geometry view, whatever) then you'll be looking at the highest LOD no matter how far you are, and some objects that are culled entirely would show up as well (like the ocean).

Apparently it's very common for games with large bodies of water to have them flowing under the map

 

[Deleted member 13524]

if we're going to entertain the possibility of an evil scheme involving invisible oceans and concrete slabs then we should seriously look into the possibility of Cevat Yerli being a shapeshifting reptilian

Had you known a bit of the studio's history, you'd know Cevat Yerli is everything but a saint.
From there, possibilities are endless.

You're looking at the mesh again, if you're actually ingame and your nose is touching the object, tess factor will be at it's highest . if you're in wireframe view (debug geometry view, whatever) then you'll be looking at the highest LOD no matter how far you are, and some objects that are culled entirely would show up as well (like the ocean).

Except the concrete slab only caught the reviewer's attention because perfstudio detected that particular object was taking too much of the GPU's time, before running it in wireframe.

However, in the DX11 "ultra" mode, the handling of this particular object takes up a pretty good chunk of GPU time during the creation of this frame.

So whatever explanation Crytek might have provided (never seen it before, feel free to link it), keep in mind it's just what the person was told to write. Which may or may not be the truth.

 

[Anarchist4000]

Crisis aside, I doubt Forza is doing anything bad here. They're just pushing a lot of reasonable geometry that gets intense in some scenes. If it's 64x tess factors across the board it's another issue, but if maintaining triangle size then the results are likely reasonable.

Well it could be too much for a 3gb, but might not be enough to slow down a 3.5 or 4 gb, we are just guessing at that point.

That's a solid educated guess though and exactly why I went looking for benchmarks between the 3/6GB and 4/8GB models of essentially the same card. The original benchmark seemingly excluding all the 8GB AMD options (nothing between 4GB and 8GB and Nvidia options were there) would seem a good method to bias results for marketing purposes. Any card will slow down with texture streaming. It's just a consideration. 4GB I'd expect devs to target as an enjoyable experience, 3GB not so much and that seems to show.

Off topic, but Nvidia really shouldn't have released a "1060" with 3GB though. A cut down chip with 6GB of slower VRAM to go easier on what are likely lower end systems probably would have been a better solution.

 

[Rodéric]

It's unfortunate, I think I don't have Crysis 2 on my Steam account, otherwise I'd start it and verify by myself, NOT using that debug view...
(I believe I have the DVD at home though, but very little time to investigate .)
Anyone can capture at the right location with RenderDoc ?
Same question for Forza Horizon 3 ^^

 

[Razor1]

Crisis aside, I doubt Forza is doing anything bad here. They're just pushing a lot of reasonable geometry that gets intense in some scenes. If it's 64x tess factors across the board it's another issue, but if maintaining triangle size then the results are likely reasonable.

Neither did Crysis, people didn't know what they were looking at at first

That's a solid educated guess though and exactly why I went looking for benchmarks between the 3/6GB and 4/8GB models of essentially the same card. The original benchmark seemingly excluding all the 8GB AMD options (nothing between 4GB and 8GB and Nvidia options were there) would seem a good method to bias results for marketing purposes. Any card will slow down with texture streaming. It's just a consideration. 4GB I'd expect devs to target as an enjoyable experience, 3GB not so much and that seems to show.

If you can call it educated based on no data Its more of a blind guess. As I pointed out it could be other things involved and its just a guess at anything. Kinda like a dart board and throwing darts at it with your eyes closed. Something is bound to hit.

Noticed that is the only 1060 3gb run in Forza 3 that has been done, so I can't see that as anything conclusive. It could be the guy messed up some where or his system isn't correct and that is causing the massive dips in min frame rates.

Off topic, but Nvidia really shouldn't have released a "1060" with 3GB though. A cut down chip with 6GB of slower VRAM to go easier on what are likely lower end systems probably would have been a better solution.

I agree I have thought it a bad idea too, but then again, with these cards you will be turning down settings anyways. Its for mainstream buyers, they aren't looking for the best of the best.

 

[Deleted member 13524]

I doubt Forza is doing anything bad here.

No, not bad (as in intentionally hurting one IHV over the other), IMO. But if you're the sole game showing substantially worse results for cards of a higher tier than lower-end ones, then the game is probably doing something wrong.

Especially when the console version running on a low-end AMD GCN1.0/1.1 GPU (which is the GPU family hurting the most?) is so similar in IQ to the maxed-out PC version.

 

[Anarchist4000]

If you can call it educated based on no data Its more of a blind guess.

So empirical data isn't evidence? We've had graphics of memory usage and framerates from a variety of cards and sources. We've had videos of assets slowly loading in. There was ample evidence to make the claim and the outcome obvious to any informed user. Not difficult to put the pieces together when the results show exactly what one would expect given the scenario.

 

[Razor1]

Where is the empirical data, I don't see memory usage numbers, do you? on the 3gb 1060 review? We don't know anything about it because they guy didn't change any settings that would allow a change of vram, (he did use dynamic settings which you can't know what that does as it does everything).

All that guys has to do is change the texture resolution settings to a point where the 3gb won't be bottlenecked and that would have given your empirical data, which he didn't do. So I don't know who or how anyone can say that is empirical data. The other review the guy mentions the dropping geometry resolution is what gives the r390 a performance boost over all the other settings. That tells us that it is where the problem is.

You can guess it might be what it is but that is a guess.

 

[DavidGraham]

Results for Quantum Break DX11 Steam version are up, not surprisingly the game works in a better performance spectrum than the DX12 version, and I think that applies to both hardware from NV and AMD:

http://gamegpu.com/action-/-fps-/-tps/quantum-break-test-gpu-steam-versiya