***THE BF4 BENCHMARKS THREAD***

[pgi947]

Don't play Hardcore - it's for people who can't aim and like camping .

At least on normal you've got a chance of getting out of a sniper hit and healing up. Yes it's unrealistic but it's also a lot more fun and skilled.

Its a shame as its the opposite on COD HC in my experience, players tended to push objectives more as they know shots would count rather then sit back and let the K/D build up.

Honestly don't understand this anti-COD culture especially amongst BF players, BF is far more geared towards assisting campers & tubes.

 

[pgi947]

COD HC was a very different beast as you could actually run and gun and pull it off due to being able to ride the more precise "netcode" and take people down before they could even react. (Can't really explain it very well but HC COD run and gunners will know what I mean).

BF4 a large degree of skill is removed due to being able to fall on the wrong side of a server tick + client update rate and end up totally unable to win an engagement as your the best part of 200ms or sometimes more behind the other player's action.

I know what you mean, in COD HC, the quickest most accurate man wins simples, usually regardless of weapon type too, A shame the series has been circling the plug hole for a few years now.


Got metro benched, just a heads up, as it stands Afterburner + Fraps + BF4 doesn't mix, BF4 is refusing to launch with that combo running for me at least.

Metro Conquest full 32 man server, about half and half indoors/outdoors:

Frames, Time (ms), Min, Max, Avg
136913, 759219, 114, 201, 180.334

780 SLI @ 1110/6200, 4770K @ 4.6ghz

 

[pgi947]

Ahhh so its possibly an MSIAB 64-bit OSD conflicting issue? I only really have AB running for my fan profile, can always just disable OSD if thats the problem

 

[pgi947]

A round on Caspian 2014, SLI 780's @ 1110/6200, 4770K @ 4.6ghz.

32 man map, conquest, couldn't see any 64 player maps AT ALL for this map

Frames, Time (ms), Min, Max, Avg
202343, 1398079, 84, 201, 144.729

Edit: the fraps related crashes come from any rivatuner based program, both ab and px force the same error regardless of osd status, so using xpertool for fan control while benching.

 

[pgi947]

You may get a better result on another map humbug, I can't speak for mantle but in dx shanghai is terrible for utilisation of anything, at random points my cpu and gpu usage bottoms out at around 40% per card and 20% on each thread dragging frames down with it, it very very rarely sits at the 95%+ per card the other maps do. I can't say where it happens as it seems completely random. Caspian 2014 seems pretty stressful

 

[pgi947]

I destroy his Min FPS with a much weaker CPU and lower graphical settings. Mantle wins

FTFY

 

[pgi947]

I'm still getting lack of audio in some maps.

Why they haven't addressed this yet I don't know, every single tdm server I go on someone is asking why they don't have audio, nice to see dices priorities are in order. Sod the large scale obvious bugs let's patch insignificant things like headshot icons.

 

[pgi947]

Matt get dropbox, add image to public folder, right click image, copy public link and paste it in IMG tags, sooooo much easier to manage.

 

[pgi947]

Thanks PG will try that next time.

Can people confirm those last images i posted work? I have to say now ive finally tried Mantle out, im extremely impressed.

You won't regret it, makes moving images from a mobile device to a PC a doddle too!

 

[pgi947]

If i have FRAPS running for benching i have to close afterburner as the bf4 exe will stop working upon launch. It doesnt seem to affect fan profile or temps, cause with gpuz logging in the background my temps are the same as with ab osd running.

Same issue here, must be something to do with the backend of afterburner as PX causes the same issue, Gainward Xpertool does work however (for fan control).

I feel your burn. When I was playing earlier, my leg was getting very hot. Not looking forward to the summer so much now and will probably need a fan just for my legs.

180 fps hmmmm. How does that add up to price per performance?

Want a 1080?

 

[pgi947]

Ahh, expertool. Great little program. Owned a few gainward cards over the years and always found it great for overclocking.

Its a very underrated tool IMO, what doesn't get much of a showcase with Phantoms is Xpertool not only offers memory over voltage, it also allows OCP adjustments, if that's not enough, on the PCB, remove one of the warranty stickers, solder a bridge in to a few clearly marked points to disable OCP and enable overvoltage

When you going water Setter?

Nah, I meant a fan to keep my legs cool One of those oscillating jobbies. Only have my pump on 3 and my fans on 1000rpm and keeps the GPU's/CPU cool enough for prolonged gaming but will prolly need a bit more in the summer.

Aye, I may chuck an extra 120 into my GPU loop before I start it to be on the safe side for summer.

 

[pgi947]

rip the lid off and put an EK naked mount kit on it then

 

[pgi947]

I wouldn't chance it tbh, its a precise mount kit made specifically for Supremacy blocks (doesn't even support the older Supreme HF), supposed to be made in such a way that its impossible to damage the die.

That said your clocks/volts are a similar range to mine, my chips just got some liquid metal under the IHS and temps are superb. Even after a good 4-6 hour session of BF4 temps are sitting in the mid 50's at 4.6ghz

 

[pgi947]

Mantle will be a bit like Gsync i think. In regards that you think something is perfectly smooth, then you try it (Mantle/Gsync) and it opens your eyes to increased smoothness that you just didn't think was possible before. Its hard to explain but you just have to try it yourself to believe/understand it. Of course there will no doubt be people that try either Mantle/Gsync and can't notice any difference/improvement. Those people will likely be in the minority though.

Correct, both are a solution to a similar problem, reviewers who got one of those Asus diy units said a similar thing about not noticing until you try it then you want it to be the norm.

Technology is moving forward in all directions which is a great thing for us.

 

[pgi947]

Which cards Kaap?

The supersample/resolution scale is basically downsampling, it takes a higher resolution and compresses it down to your resolution. So 100% is native, 200% is 2x resolution (I think!).

So if I know my Kaapy and I like to think I do, thats 4 x 290X @ 1600p, ultra with around abouts 4K worth of sampling, impressive averages, be interesting to see where that minimum came from.

 

[pgi947]

What do the settings above mean, what resolution are they equivalent to ?

Thanks.

From what I've read its 200% of your native resolution, so 1600p x 2, which iirc is around 4K.

In other words 4 x 290X's are holding up very very nicely at BF4 ultra 4K

Edit: Matts right, its ABOVE 4K, even more impressive.

Edit 2: Or does it not go by pixel count? 1600p is roughly 4m pixels, 4K is around 8m pixels...200% of 4m pixels is 8m... ******* maths on a Saturday evening, what has my life become

 

[pgi947]

Just about to try 1440p @ 200% ultra on a pair of 780's. Should be interesting

 

[pgi947]

Is that those Phantoms?

Yep!

Bless you my child

Ok so 2560x1440 200% ultra (HBAO, 4xMSAA and all that jazz)...I tested using the testing range, so you can probably slash another 10-15% from these results to give indication of a busy 64 man server.

200% I ran out of VRam near instantly, turning it into a slide show, it reported it was running arounf 30-40FPS but felt more like 10-15 fps.

Even dropping MSAA to 2x and HBAO to SSAO and Post to medium it STILL ran out of vram.

Vram limit hit at around 175% scaling, though at 150% it still hit around 2.9GB so you have to feel on a regular map it would still burst through the vram wall.

Regular 1440p 100% ultra was perfectly fine, frame rates sitting in the 140+ region, Vram usage at around 2.2GB.

For sure 4K needs bare minimum 4GB cards.

Edit: actually of Matt's about can you run the same test (just ponder about the test range blowing things up) and let us know how your vram looks at 1440 200% ultra? Try see where the drop off point is for vram.

 

[pgi947]

So:

1080 200% is 3840x2160 (4K)

1440 200% is 5120x2880
1440 150% is 3840x2160 (4K)

Running out of VRam on all 3!

Matts doing some testing now.

 

[pgi947]

Maybe how the frames are renders from one resolution to another then compressed for display? Must be how dice have implemented it then as native driver down sampling is pretty much spot on