Battlefield 1 Beta CPU scaling performance

[Carfax83]

This latest iteration of the Frostbite 3 engine seems to be exceptionally optimized CPU wise to get such high frame rates. You don't see any boost from DX12 unless you go to the very bottom of the list with the slowest CPUs. But this DX12 implementation is definitely beta, so I'm sure it will be improved by launch time..

Not sure how much more they can get though, as their DX11.1 path is really good.. Which kind of confirms what I've been saying in the VC&G forums. DX12 is practically useless for games unless those games are bottlenecked by the CPU..

Games like BF1 which are designed for current gen consoles don't have the complexity to really benefit from more powerful APIs like DX12, since the much faster desktop CPUs available in gaming PCs typically overcome the CPU overhead for DX11 with ease.

So BF1's DX12 path is probably only going to benefit those with low end CPUs, and those with ultra high end rigs that use SLI and XFire..

When games for Scorpio and PS4K start being made, then we'll see much greater benefits from DX12.

 

[Spjut]

There's barely any action going on in GameGPU's test sequence though, one of the best case performance scenarios you can have in the game, and I know the test is like that for repeatability purposes. But anyone that has played Battlefield 1 or any other BF game can agree that performance is alot different when there is more action.

 

[ozzy702]

There's barely any action going on in GameGPU's test sequence though, one of the best case performance scenarios you can have in the game, and I know the test is like that for repeatability purposes. But anyone that has played Battlefield 1 or any other BF game can agree that performance is alot different when there is more action.

Yep, 64 man servers with a lot of action going on makes a huge difference, I swear it's almost double the CPU load than these canned benches.

 

[dogen1]

My g3258, even at 4.2GHz, chokes and dies on this game. Rarely goes above 35fps in a big mp match.

 

[Headfoot]

Which kind of confirms what I've been saying in the VC&G forums. DX12 is practically useless for games unless those games are bottlenecked by the CPU

This doesn't confirm that in the least.

You're looking at a game which MUST by definition still conform to DX11 design patterns in order to be compatible with DX11 and concluding that DX12 is useless. Did games that ran DX10 and DX11 look anywhere near as good or run anywhere near as well as games that were only DX11? Obviously not.

They have to build to the lowest common denominator among all code paths: DX11, DX12, Xbox One, PS4. The lowest common denominator is DX11.

We will see DX12 with the stops pulled when we see games that are incompatible with DX11 altogether.

 

[AtenRa]

My g3258, even at 4.2GHz, chokes and dies on this game. Rarely goes above 35fps in a big mp match.

Dual cores are obsolete long time ago, no matter the high IPC. Even if they get high fps, they stutter like hell.

 

[Spjut]

The sad thing right now with DX11 vs DX12 in Battlefield 1 is that even with old/slow CPUs, DX12 is only marginally better at best. In BF4 and Hardline, changing from DX11 to Mantle made a tremendous difference for CPU limited systems. In the results above, we see that the FX 4100 gained only five FPS by using DX12.
And considering Dice has been most vocal with the move to low-level APIs and showed great results with Mantle, I'm disappointed that even Dice can't show DX12's superiority right now.

 

[dogen1]

Dual cores are obsolete long time ago, no matter the high IPC. Even if they get high fps, they stutter like hell.

Works pretty well for me most of the time.

Plus you'd think with ~2x the IPC of jaguar, and 2.5x the clock speed as consoles, it would do better than 1/2 the performance in this game, even with 1/3 the core count. Maybe it will once DX12 is polished up a bit.

 

[daxzy]

For all the people saying DX12 sucks in this Beta. You realize that when you click on DX12 in BF1, it explicitly states that it is unstable? Obviously the devs will focus on DX11 first.

 

[TheELF]

My g3258, even at 4.2GHz, chokes and dies on this game. Rarely goes above 35fps in a big mp match.

Open the console with ~ and write render.drawscreeninfo 1 and press enter.
Notice the job thread count,previous BF games used to run 1 job thread on dual cores,so one main and one job, 2 threads for the dual and they ran great.
On the BETA it runs 3 job threads,basically it's locked in quad core mode running 3 job threads and one main, so it runs like crap without reason,hopefully it's just a beta thing I mean BF4's early builds didn't even start up on duals so it's still an improvement.

 

[dogen1]

Open the console with ~ and write render.drawscreeninfo 1 and press enter.
Notice the job thread count,previous BF games used to run 1 job thread on dual cores,so one main and one job, 2 threads for the dual and they ran great.
On the BETA it runs 3 job threads,basically it's locked in quad core mode running 3 job threads and one main, so it runs like crap without reason,hopefully it's just a beta thing I mean BF4's early builds didn't even start up on duals so it's still an improvement.

Kinda lame, but I guess they don't officially support them.

 

[Arachnotronic]

Dual cores are obsolete long time ago, no matter the high IPC. Even if they get high fps, they stutter like hell.

dual cores w/o SMT. 2C/4T chips w/ high perf/clock and high clocks can still perform well in games.

 

[Carfax83]

You're looking at a game which MUST by definition still conform to DX11 design patterns in order to be compatible with DX11 and concluding that DX12 is useless. Did games that ran DX10 and DX11 look anywhere near as good or run anywhere near as well as games that were only DX11? Obviously not.

They have to build to the lowest common denominator among all code paths: DX11, DX12, Xbox One, PS4. The lowest common denominator is DX11.

You seem to have a penchant for taking quotes from posts, without reading the entire post itself. I actually mentioned the console factor in my OP, so you bringing it up again as a counterpoint is redundant, since I already covered it in my OP.

We will see DX12 with the stops pulled when we see games that are incompatible with DX11 altogether.

In my OP I stated that we'll see the true benefits of DX12 when the Scorpio console is released next year. It will be much more powerful than current consoles, so developers won't have any excuse to really crank up the image quality and detail.

 

[dogen1]

dual cores w/o SMT. 2C/4T chips w/ high perf/clock and high clocks can still perform well in games.

Even non SMT dual cores can do pretty well in some modern games.

 

[Carfax83]

Works pretty well for me most of the time.

Plus you'd think with ~2x the IPC of jaguar, and 2.5x the clock speed as consoles, it would do better than 1/2 the performance in this game, even with 1/3 the core count. Maybe it will once DX12 is polished up a bit.

DX12 will certainly help, but the real problem for dual cores will be how current and future game engines are being designed, which is with a strong emphasis on parallelism. The more parallel an engine is, the more tasks are broken up and distributed across multiple threads. This is great for CPUs with lots of cores and threads since the entire CPU can now be used with no limitation for a nice performance increase. But on CPUs with less cores and threads, the chances of choking or overwhelming the CPU goes up considerably, as physics, game code, A.I, asset streaming, rendering and whatever else are now being distributed across the entire CPU.

In the past, separate tasks would get their own threads, but now multiple tasks are occupying the same thread..

We saw this in the Division where dual core CPUs without SMT had asset streaming issues that made the game unplayable (the CPU couldn't stream assets fast enough and you'd run into invisible walls and barriers), even though you could start the game. I expect things like this to get more frequent as game engines become more parallelized.

 

[Sweepr]

Really bad showing from Bulldozer here, i5-2500K still outperforming FX-8150 in the X1/PS4-era, and that's before you count the insane overclocks (>4.5 GHz). Performance per core still matters.

 

[Arachnotronic]

Really bad showing from Bulldozer here, i5-2500K still outperforming FX-8150 in the X1/PS4-era, and that's before you count the insane overclocks (>4.5 GHz). Performance per core still matters.

Yeah, it's not a good showing at all. Sandy Bridge was by far the better long-term bet back in 2011 for gamers.

 

[RaistlinZ]

Wow, 2600K is still killin' it years later. Mad props.

 

[Arachnotronic]

Battlefield 1 Minimum PC (CPU) Requirements
source: https://forums.battlefield.com/en-us/discussion/19282/battlefield-1-minimum-pc-requirements

• Processor (AMD): AMD FX-6350 3.9GHz @ TurboBoost 4.2GHz,
• Processor (Intel): Core i5 6600K 3.5GHz @ Turbo 3.9GHz,

LOL, whose job was it to make these system requirements? They are going to unnecessarily discourage people from buying the game.

 

[moonbogg]

Any quad core is good enough to play the game. Why they would choose a skylake i5 as minimum is way beyond me. Also, with BF as my main game, I don't know about upgrading to a new platform next year. I have even higher resolution than 1440p now and with a 100hz limit, I am thinking a new CPU would net me precisely zero FPS gain in this game. That stock 3970x is still wrecking it, so a 3930k@4.6 will still be demolishing this franchise for some time. I wonder how long it will take before my CPU or motherboard actually BREAKS? It might have to break before I buy a new one the way things are going. I'm never in the mood to spend $1000+ on a platform, or on anything else if the experience benefit will be basically nothing.

 

[VirtualLarry]

We saw this in the Division where dual core CPUs without SMT had asset streaming issues that made the game unplayable (the CPU couldn't stream assets fast enough and you'd run into invisible walls and barriers), even though you could start the game. I expect things like this to get more frequent as game engines become more parallelized.

To be sure, that was a programming error, and not any problem inherent to dual-core CPUs. All that proves is proper multi-threaded programming is difficult.

 

[2is]

The sad thing right now with DX11 vs DX12 in Battlefield 1 is that even with old/slow CPUs, DX12 is only marginally better at best. In BF4 and Hardline, changing from DX11 to Mantle made a tremendous difference for CPU limited systems. In the results above, we see that the FX 4100 gained only five FPS by using DX12.
And considering Dice has been most vocal with the move to low-level APIs and showed great results with Mantle, I'm disappointed that even Dice can't show DX12's superiority right now.

It's even worse then that. I played the game on a Sandy Bridge i3 and DX11 was still faster. But, it IS beta and the DX12 portion is probably the most beta part of the entire game.

 

[showb1z]

DX12 is practically useless for games unless those games are bottlenecked by the CPU..

You don't say.
We already have games like that though. Like Hitman and Total War, BF1 is pretty much guaranteed to benefit from it on 64 player servers once DX12 is actually finished.
Pointless trying to draw conclusions about DX12 based on this beta implementation. I hold DICE in higher regard than to think their DX12 path is actually going to regress performance.

 

[Carfax83]

Pointless trying to draw conclusions about DX12 based on this beta implementation. I hold DICE in higher regard than to think their DX12 path is actually going to regress performance.

I don't know what you're talking about as I didn't draw any conclusions. I clearly said in my OP,"But this DX12 implementation is definitely beta, so I'm sure it will be improved by launch time.."

The only one drawing conclusions here is you dude..

 

[TurtleCrusher]

I don't know what you're talking about as I didn't draw any conclusions. I clearly said in my OP,"But this DX12 implementation is definitely beta, so I'm sure it will be improved by launch time.."

The only one drawing conclusions here is you dude..

This latest iteration of the Frostbite 3 engine seems to be exceptionally optimized CPU wise to get such high frame rates. You don't see any boost from DX12 unless you go to the very bottom of the list with the slowest CPUs. But this DX12 implementation is definitely beta, so I'm sure it will be improved by launch time..

Not sure how much more they can get though, as their DX11.1 path is really good.. Which kind of confirms what I've been saying in the VC&G forums. DX12 is practically useless for games unless those games are bottlenecked by the CPU..

Games like BF1 which are designed for current gen consoles don't have the complexity to really benefit from more powerful APIs like DX12, since the much faster desktop CPUs available in gaming PCs typically overcome the CPU overhead for DX11 with ease.

So BF1's DX12 path is probably only going to benefit those with low end CPUs, and those with ultra high end rigs that use SLI and XFire..

When games for Scorpio and PS4K start being made, then we'll see much greater benefits from DX12.

Uhhh, those are conclusions.

 

[Carfax83]

Whatever dude. I clearly said it was beta and that performance will be improved by launch time, so that's not a conclusion. Also, I'm sure that once the servers are fully manned, DX12 will begin to shine even more..