• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

Investigating The Frostbite Mantle Implementation

Gerard said:
Matt your twitter feed is like an amd reps :p
Click to expand...

You don't become an AMD God (not my words) without being enthusiastic about AMD products/Tech. :p

uksoldierboy said:
Well that's a kick in the balls:( extremely disappointing.
Click to expand...

Disappointed to be honest, I expected him to be a perfectionist. Still at least they're hoping to rejuvenate it for future titles. Not totally satisfied with that response mind you. It's fine for me, i have a couple of 290's which pwn most things but i also have a 7770 in another pc and i would expect to be able to use that as well.
 
So is VRAM usage higher using Mantle than D3D11 in 'all' games or is it just Frostbite engine powered ones? Does Thief show higher usage and does that memory usage increase with each rendered frame?

The information is not clear at this point, but if the (graphics) memory usage does increase w.r.t. time in Thief as well as all the other games using Mantle then that suggests a serious memory leak issue with Mantle itself.
 
PhysicsMan94 said:
So is VRAM usage higher using Mantle than D3D11 in 'all' games or is it just Frostbite engine powered ones? Does Thief show higher usage and does that memory usage increase with each rendered frame?

The information is not clear at this point, but if the (graphics) memory usage does increase w.r.t. time in Thief as well as all the other games using Mantle then that suggests a serious memory leak issue with Mantle itself.
Click to expand...

As i mentioned in the OP, the issue is with the Mantle implementation in Frostbite only. Unfortunately at the moment that is the majority of Mantle games. Thief and Starswarm do not have this issue. Thief vram usage stays constant. Frostbite Mantle rises and rises and eventually peaks and settles, but not before it's used up too much vram. This must be redesigned for future titles, hopefully this thread will encourage them to do that.
 
LtMatt said:
As i mentioned in the OP, the issue is with the Mantle implementation in Frostbite. Thief and Starswarm do not have this issue. Thief vram usage stays constant. Frostbite Mantle rises and rises and eventually peaks and settles, but not before it's used up too much vram.
Click to expand...

I see, but is Thief VRAM usage higher when using Mantle than when using D3D11?
 
PhysicsMan94 said:
I see, but is Thief VRAM usage higher when using Mantle than when using D3D11?
Click to expand...

It is, around 500mb-750mb higher at 1440P highest details, but it does not increase past that. It appears to use most (3.5gb exactly) of the vram of my 290 effectively. Not sure how directly comparable the two API's are on memory management as i imagine they use memory differently. The problem is, Thief uses a chunk of your vram and does not use more. Frostbite Mantle uses a chunk and keeps rising until it reaches a plateau then stops.
 
I would suggest that BF4 / Frost Bite 3 was built with Mantle very early on in Alpha state, 2 years or so... it was the test bed learning curve engine and will have all the Alpha State bugs and incompatibilities. it was probably also developed in conjunction with the Hawaii architecture, which does have a different memory management system.

Now that its embedded in the engine it may not be so easy to correct, what we have now is a stable Beta API, its probably this later version that you find in Thief ecte.. the next test will be SE3.

I think what will happen is nothing much as the old GCN 1.0 is rapidly being phased out by AMD.
 
LtMatt said:
It is, around 500mb-750mb higher at 1440P highest details, but it does not increase past that. It appears to use most (3.5gb exactly) of the vram of my 290 effectively. Not sure how directly comparable the two API's are on memory management as i imagine they use memory differently. The problem is, Thief uses a chunk of your vram and does not use more. Frostbite Mantle uses a chunk and keeps rising until it reaches a plateau then stops.
Click to expand...

Is there some way of knowing if the cards actually using the ram or just caching it? Afaik it shows up as "used" either way when it might not be the case.
 
Gerard said:
Is there some way of knowing if the cards actually using the ram or just caching it? Afaik it shows up as "used" either way when it might not be the case.
Click to expand...

No way of knowing as far as i know. Way above my pay grade im afraid.
 
Only AMD/DICE know where the issues are. Simply monitoring VRAM usage does not, at all, give you a clear picture of the problem.

It might be a sole problem with Frostbite
It might be a sole problem with Mantle
It might be a combination of both Frostbite and Mantle.

Just because a behaviour occurs in 1 game engine and not another does not mean the game engine is at fault. It could be a Mantle bug with a particular process in Engine X that Engine Y does not even use. Is this the fault of the engine? Could be, but not necessarily.
 
In fairness to AMD, they have done well with Mantle and blaming them for Dice messing up is like blaming nVidia for Watch Dogs being poor.... Sorry, I couldn't resist :p

In truth BF4 has been buggy from the off and needs some serious work. CTE seems to be pretty respectable mind but still not cured of all the bugs.
 
LtMatt said:
You don't become an AMD God (not my words) without being enthusiastic about AMD products/Tech. :p
Click to expand...
Whatever you do then mate, dont buy one of their cpu's, your enthusiasm might take a bit of a hit.:D

But seriously, ive said it before, i think mantle is a great idea. But bf4 was no game to show it off properly on. Whilst there is no doubt improvements to how the game runs on mantle compared to DX. The game itself is still far from a polished product. And going on dice/ea's track record, i doubt very much if it ever will be.
 
Apparently Mantle didn't take very long for 1 person to implement in BF4. Maybe they could just rip it out and start again? Put a few more people on it and get it done in a week or 2.

Of course this is DICE so if it's not DLC they can charge for they're probably not interested.
We're only the customer and it's our role to pay them for what they want to make, it's not for them to make what we want (or pay them for).
 
LtMatt said:
It is, around 500mb-750mb higher at 1440P highest details, but it does not increase past that. It appears to use most (3.5gb exactly) of the vram of my 290 effectively. Not sure how directly comparable the two API's are on memory management as i imagine they use memory differently. The problem is, Thief uses a chunk of your vram and does not use more. Frostbite Mantle uses a chunk and keeps rising until it reaches a plateau then stops.
Click to expand...

I ran the Thief benchmark on my 7950CF, as that caused me issue in BF4.

@ 1080p
DirectX uses 2289.5MB (GPU-z shows 4579MB)
Mantle uses 2703MB (GPU-z shows 5406MB)

So that's 413.5MB more VRAM.
Now there weren't any problems because memory usage is lower than BF4. So if VRAM usage were 14% higher then Mantle would likely give issues while DirectX would be ok (unless of course there's caching being done), while DirectX could stand about 34% higher VRAM before it would be an issue.

While BF4 may have other issues, I don't think the excessive VRAM use is exclusive to it.
Maybe developers aren't as good at managing memory as everyone thought they were. Hopefully if it is just that simple they'll get better as they become more familiar.

Interesting to know the VRAM usage on 2GB and 1GB cards using the settings as per the benchmark thread. See how much VRAM it can make do with.
 
You must log in or register to reply here.
Back
Top Bottom