Mantle and Frame Latency

[Gregster]

Shankly's results are a thin line which is more consistent than a fuzzy one, the line rising is down to frame rate dropping and having less gpu grunt in general.

A straight line at 20ms is no less consistent than a straight line at 5ms, it just means that the frame rate is lower at 20ms.

This is correct. Ideally you want a perfectly flat line like Shanklys on the left. This becomes a bit more difficult to achieve when using more than one gpu. Also worth noting that more gpu's will have a lower ms time because they generate more frames. Does not always mean a smoother experience though and those spikes we see indicate variance and a frame taking longer to render than the previous one. A little variance is normal and to be expected. Only once it passes a certain mark does it usually indicate stutter. The good thing about Mantle is, typically that variance is considerably lower than any other solutions available.

Ok fella's if you want to believe it is smoother on AMD, that's cool. That works for me

 

[tommybhoy]

It's not about wanting to believe when data shows what the actual end results produce.

 

[Final8y]

Ok fella's if you want to believe it is smoother on AMD, that's cool. That works for me

Its basic fundamentals, 4+4=8, it does not matter if you dont believe it to be so because you dont understand the fundamentals.

 

[Final8y]

It's not about wanting to believe when data shows what the actual end results produce.

Indeed and they are both low enough to not really be noticed and im done.
Im going to play some BF4 and see what the update has brought.

 

[CAT-THE-FIFTH]

Ok fella's if you want to believe it is smoother on AMD, that's cool. That works for me

Look at The TechReport and their analysis of frametime graphs. Even look at PCPER when they analyse their graphs.

Smoothness at least according to them is also dependent on the sudden "spikes" - its the spikes which if big enough are an indication of stuttering.

More spikes=inconsistent frametime delivery.

Less spikes=consistent frametime delivery.

This OFC goes hand in hand with the actual frametimes themself.

 

[Rusty0611]

Are they big enough? My 290 feels identical to my 780. That's probably more of an achievement for the 290 than the other way round

 

[Gregster]

It's not about wanting to believe when data shows what the actual end results produce.

Its basic fundamentals, 4+4=8, it does not matter if you dont believe it to be so because you dont understand the fundamentals.

That's great

 

[setter]

Indeed and im done.
Im going to play some BF4 and see what the update has brought.

Reg is a bit better, but quite a lot of rubber banding on servers that haven't been restarted. Though most should be sorted now. But its still nowhere near as smooth as CTE where a lot of the changes were tested. I suspect that as CTE was a new build of the game they used totally different servers with a better tick rate than vanilla bf4.

 

[Final8y]

Reg is a bit better, but quite a lot of rubber banding on servers that haven't been restarted. Though most should be sorted now. But its still nowhere near as smooth as CTE where a lot of the changes were tested. I suspect that as CTE was a new build of the game they used totally different servers with a better tick rate than vanilla bf4.

A friend yesterday was just mentioning that but i was shattered after work to even bother playing.

 

[CAT-THE-FIFTH]

Are they big enough? My 780 felt identical to my 290. That's probably more of an achievement for the 290 than the other way round

Ultimately,even the Tech Report said in their various tests of Mantle it usually lead to more consistent frametimes even on APUs(!),and that was like early in the year with a very buggy Mantle driver too. If anything it could be explained that when it works properly,it is probably down to reduced CPU overhead,ie,the driver is not waiting on the CPU to do certain things first.

Nvidia has tended to thread their drivers better than AMD in the past,whereas the AMD drivers tend to hog a single thread more,which is kind of why Mantle will have an effect in the way described.

 

[Rusty0611]

Game felt seriously good for me on Wednesday when I played with the MLG. Hit reg was as good as I've ever seen it.

 

[setter]

All were supposed to be restarted last night. On those that weren't you could use an exploit to get loads of battlepacks. 4 were released alongside the patch. By joining a game in an older server you could get 4 more just by clicking to join then closing the game via the bl task bar. Rinse and repeat, few guys in the bf4 thread got over 70 doing this.

 

[Kaapstad]

I have not read most of this thread but it seems like a lot of threads where people are putting too much faith in graphs and not enough in user experience of the games with different hardware.

It will be different for other people but in BF4 I prefer DX11 to Mantle and using Titans to 290Xs.

Having said that in Civ5 it is the other way round and I prefer using the 290Xs.

The point is the user experience is specific to the person and there is no way a graph can tell you what it will be.

As I said above I prefer Civ5 on the 290Xs and toward the end of a game I will only use a single card as it runs better even @4K, if this was analysed on a graph it would look awful but actually playing the game is fantastic.

 

[mmj_uk]

With a 60hz monitor anything under 16ms is fine as that's how often the screen refreshes, the results in OP are 1% percentile so even if the spikes are not down to poor testing methods (which they are) you'll do well to notice without a pretty graph sitting in the bottom corner.

It seems to me that now Mantle's performance is no longer anything to write home about (compared to DirectX) the Mantle brigade have shifted the justification for its existence to being about frame times, grossly exaggerrating the difference it makes.

 

[Gregster]

Game felt seriously good for me on Wednesday when I played with the MLG. Hit reg was as good as I've ever seen it.

Yer, the CTE patch has made it a good experience. A shame Pearl Market isn't running smooth still and was very stuttery for me, even with G-Sync smoothing.

 

[Tonester0011]

Yer, the CTE patch has made it a good experience. A shame Pearl Market isn't running smooth still and was very stuttery for me, even with G-Sync smoothing.

They've patched BF4 again, I've been slightly addicted to FF14 so not played BF since the other week. Will check if you guys are on this weekend and join

 

[james.miller]

I have not read most of this thread but it seems like a lot of threads where people are putting too much faith in graphs and not enough in user experience of the games with different hardware.

It will be different for other people but in BF4 I prefer DX11 to Mantle and using Titans to 290Xs.

Having said that in Civ5 it is the other way round and I prefer using the 290Xs.

The point is the user experience is specific to the person and there is no way a graph can tell you what it will be.

As I said above I prefer Civ5 on the 290Xs and toward the end of a game I will only use a single card as it runs better even @4K, if this was analysed on a graph it would look awful but actually playing the game is fantastic.

You prefer titans because you're running multiple cards at 4k, right? don't you run in to memory issues with mantle at 4k? You're right at the extreme end of hardware so it's not at all suprising that your experience might be different. That said, there's no reason why you couldn't produce the same frame-pacing charts at 4k, right? wouldn't they show us exactly what you are experiencing?

 

[setter]

Yer, the CTE patch has made it a good experience. A shame Pearl Market isn't running smooth still and was very stuttery for me, even with G-Sync smoothing.

Popped on for a quick blast just now. Felt fantastic tbh. 64 player cq on propaganda. All running great but then my ping shot up to 800 odd so got kicked. Need to get in contact with bt about this setup as nothing else was using the connection. And with this game a decent connection speed plays a very important part.

 

[Ian Evey]

I dont think many people have said amd dx is not as smooth as amd mantle, which is the only thing i can see from shankly and jwrpalmer graphs.

Both look good both play smooth, nvdia have the edge in minimum and average fps but that is irrelevant these days it seems.

 

[Nexus18]

It seems to me that now Mantle's performance is no longer anything to write home about (compared to DirectX) the Mantle brigade have shifted the justification for its existence to being about frame times, grossly exaggerrating the difference it makes.

Yup 100% correct, once again mantle is utterly useless.....




Battlefield 4 with max settings except no AA at 2560x1080 on my rig as specced in sig, single player:

Spoiler

DX

Mantle