Another BF4 SLI thread

[TH3R4POR]

Hi there,

I was looking for some help with my BF4 performance issue.

I seem to get worse performance running BF4 in SLI than I do with a single card. This only seems to be the case in BF4, running BF3, Arma etc I get very high FPS. Using Benchmarks like 3DMark, Furmark etc also show a marked increase in performance over a single card, it only seems to be BF4 that is doing this.

The FPS fluctuates, going from 150FPS down to 15, settings don't seem effect anything, I see the same fluctuations whether on Ultra or Low. I see these same issues during an SP game, but my ping is good on the servers I play on (<30).

My system:
* AMD 8350 CPU
* ASRock 990FX Mobo
* 8GB Corsair Vengeance RAM
* 2x KFA2 GTX670 2GB (in SLI)
* 750W Corsair PSU

(The only thing I would mention about the system is the GFX cards can get up to 85C when under load.)

All NVidia drivers are up to date, SLI is turned on and working (see below sshot and above comment about Benchmark scores). I have spent literally hours trawling forums, testing 'solutions', using different SLI profiles etc etc.

I took the below screenshot in the hope someone could interpret the GFX/CPU data. To highlight the issue this screenshot is taken with everything set to Low in settings, and it was still bottoming out during this map at 15FPS. There seems to be a bottleneck somewhere, but I can't think what i is...

Cheers in advance,

-Thera`

 

[VulgarDisplay]

Is your CPU overclocked?

 

[Hitman928]

I don't think it's a cpu problem, BF4 actually makes good use of those cores and single player isn't very cpu intensive (comparatively), so I would think the issue is somewhere else. Can you disable SLI and run just 1 card at normal settings and see what you get? Run again at low settings? I think that should narrow a few things down.

 

[TH3R4POR]

Hi guys, thank for the quick replies!

Nope, CPU is not overclocked, also worth noting I am running the latest BIOS firmware.

I just disabled SLI and played the same map on the same server at Medium settings. I have my FPS capped at 70 and it stayed solid at 70, so as I said, I seem to get better performance from one card than two.

The one thing I haven't tested which I found on a forum was a faulty SLI bridge, but the fact I see such improvements with other games and benchmarks would tell me the SLI bridge is fine.

One other thing to note about the system, the Northbridge runs crazy hot, I wouldn't be able to hold my finger on it for more than a second, but again this wouldn't explain the performance issue only occurring with BF4.

Cheers,

-Thera`

 

[Hitman928]

Have you tried clearing your system of drivers and re-installing or even trying earlier drivers?

 

[TH3R4POR]

Have you tried clearing your system of drivers and re-installing or even trying earlier drivers?

Hi,

yup, tried on several occasions, I've gone back four versions with clean installs and get the same results.

I actually just downloaded Heaven 4.0 benchmark, and see double the FPS on SLI compared to single card, so it's BF4 that isn't playing ball, which is what made me look at the SLI profiles. I used a couple of profiles people had supplied online, but that didn't fix anything.

I also tried the renaming trick, where you change the exe to a known working SLI game (Serious Sam is the favourite) but BF4 wouldn't run.

 

[Hitman928]

Hmmm, well, definitely sounds like just a driver/game issue. Wish I could help you out a bit more but I don't run multiple cards myself. I'm sure nVidia has a bug report system, right? Have you tried that route to see if you get any response?

 

[Carfax83]

The best way to test SLI issues is to use a program like MSI afterburner that can track temperature, CPU and GPU activity.

That way you can pinpoint what is occurring. From what you've described, something might be causing your second GPU performance to degrade. It could be heat (which can cause the GPU to underclock), or it could be a bad SLI bridge. Also, issues with the power supply might cause it as well.

I think heat is likely the culprit here, as you said that temps go up to 85c, which is pretty high. To test this, I would start the game with MSI afterburner running. If heat is the problem, performance should be good initially, but as the GPU warms up and gets too hot, performance will degrade as the GPU starts to downclock to protect itself.

Make sure you look at the GPU clock speeds when this is occurring.

*Edit*

Oh and fan speeds are important as well. Make sure you look at fan speeds, as the GPU fan may not be working properly.

 

[Silverforce11]

670s will throttle above 82C but it wont go below the base clock so it doesnt explain the sharp fps drops.

something is fubar and based on your perf overlay graph, looks like a massive CPU bottleneck in those scenes. Note the sharp spike up, leading to much more stutter for your GPUs.

 

[Hitman928]

670s will throttle above 82C but it wont go below the base clock so it doesnt explain the sharp fps drops.

something is fubar and based on your perf overlay graph, looks like a massive CPU bottleneck in those scenes. Note the sharp spike up, leading to much more stutter for your GPUs.

Yep, but he doesn't have the same issue looking like a massive cpu bottleneck in other games or with a single card, which makes me think it's a game/driver issue. Hearing about the issues in general BF4 has had, I'd be prone to think it's more of a game issue, but either way, bringing it to nVidia's attention is probably the quickest way of getting it resolved or figuring out what the op needs to do to fix it.

 

[el etro]

Your FPS should be much higher. Don't worry so much about your config because is a particular problem of yours.

 

[RaulF]

Your description of the FPS drop to 15 and back up sounds like when i would run out of vram. But if everything is on low i doubt you would be running out of vram.

I would suggest different drivers, not sure which ones you are using. I am using 332.21 and they run bf4 good for me on SLI. Still have a weird slow down now and then but is nothing bad.

Also make sure you resolution scale is at 100% and not higher, install evga precision x or msi afterburner and monitor what your cards are doing.

 

[blackened23]

Are you using resolution scaling. If you are, don't. Or be reasonable with it, because it is the same as downsampling or OGSSAA, the most expensive form of SSAA in existence. The performance hit associated with high levels of SSAA are usually well above 50%, so you're using it...well, that's the issue. SSAA is very taxing on GPUs and requires tons of VRAM if you go too high.

 

[Hitman928]

Are you using resolution scaling. If you are, don't. Or be reasonable with it, because it is the same as downsampling or OGSSAA, the most expensive form of SSAA in existence. The performance hit associated with high levels of SSAA are usually well above 50%, so you're using it...well, that's the issue. SSAA is very taxing on GPUs and requires tons of VRAM if you go too high.

Note, he doesn't have the issue when running with a single card.

 

[TH3R4POR]

Hi Guys, thanks for all the replies!!

* Resolution Scaling is set to 100%.

* CPU is an 8 core 8350, all cores are unparked, I really can't see that is a bottleneck??

* I've tried the latest nVidia drivers, and the previous ones I had on my network (335.23, 334.89, 331.65, 331.65). I tried these as apparently the latest drivers are tailored to the 700 series, and can have adverse affected on a 600 series.

* I don't buy into the overheating thing, these cards will happily sit running a number of different stress tests for hours with no fluctuation in frame rate. As we speak I have Furmark running with everything maxed (8x MSAA etc), top temp for the cards is 80C and the FPS is solid at 65-67, and MSI Afterburner is displaying all straight lines, no peaks or troughs what so ever.

I have done some testing though, I tried the x86 version of BF4 on Low to check MSI Afterburner (I was previously using HWMonitor, thank you ) and noticed an obvious pattern which was blurred when using x64 version on Ultra.

You'll notice the distinct troughs in GPU2 Usage % and GPU1 Fan Speed, if you follow those troughs up and down you see they match slight drops in GPU1 Power and GPU2 Usage. Looking at the second sshot (wouldn't fit on one screen ) these also appear in GPU Memory Usage, where it even more noticeable. These exactly match when I see the FPS drop ingame.

This is Max Settings on x86, which I was testing this morning:

All of these dips in the traces in MSI Afterburner match exactly spikes in the BF4 onscreen traces for CPU and GPU usage. They also seem pretty uniform in frequency? I tested this to see if it could be time based and sat in a tank in spawn, but didn't experience a spike/FPS drop. Then as soon as I left base to my usual position I get them again at uniform frequency. That to me points to heating/cooling, but the temps look OK, and are stable. They also each last for the same amount of time.

So, what is causing what? Is the drop in RAM usage causing the CPU spike, or vise versa? :-/

You'll also notice none of the temps get above 79C through 30+ mins of play.

Hope this helps someone?

Cheers,

-Thera`

 

[TH3R4POR]

**EDIT**

OK, gold star to Silverforce11! I just spent 3+ hours running around Test Island, with a stop watch on desk, timing the drops (as I thought earlier they were fairly uniform, changing things and timing again...

The drops happened every ~130 seconds, and last for 14 seconds exactly. This to me was either a process running in the background or a heating/cooling cycle, so after closing everything I could in Program Manager and still getting the issues I set the GPU fans to 100% in MSI Afterburner and the gap between drops increase to over 5 minutes, and the drop lasts for a third of the time. I just tested it ingame and while it isn't 5 minutes it is an improvement over what it was. I'm not sure why the fans didn't speed up to compensate for this, as you can see on the sshots they sit on 30% constantly!

No idea how I am supposed to cool these things enough to avoid the issue, only choice seems to be water, if fans on 100% just delays the problem.

 

[Carfax83]

OK now we're onto something..

What is the exact model of PSU that you have? Is it certified for SLI? It's possible that it may not be delivering enough amps continuously on the 12v line..

That could explain why you are experiencing those drop outs. Another thing that could cause it is if the PSU connectors aren't fully seated, or maybe even the cards themselves..

 

[Hitman928]

**EDIT**

OK, gold star to Silverforce11! I just spent 3+ hours running around Test Island, with a stop watch on desk, timing the drops (as I thought earlier they were fairly uniform, changing things and timing again...

The drops happened every ~130 seconds, and last for 14 seconds exactly. This to me was either a process running in the background or a heating/cooling cycle, so after closing everything I could in Program Manager and still getting the issues I set the GPU fans to 100% in MSI Afterburner and the gap between drops increase to over 5 minutes, and the drop lasts for a third of the time. I just tested it ingame and while it isn't 5 minutes it is an improvement over what it was. I'm not sure why the fans didn't speed up to compensate for this, as you can see on the sshots they sit on 30% constantly!

No idea how I am supposed to cool these things enough to avoid the issue, only choice seems to be water, if fans on 100% just delays the problem.

Hmmm, still doesn't seem like a thermal problem, weird that 100% fans delays the spikes. Your reported temps are fine and normally thermal issues like that aren't so precisely cyclic. Perhaps the issue is at the VRM level?

I would check your connections as Carfax suggested, post your PSU, and also, when running single GPU, have you tried swapping which GPU you run alone to make sure it isn't just the second card being faulty?

From what you posted earlier, it looks like your card isn't just downclocking, but going down to desktop speeds(as if you weren't in a game) on both the core and the memory. I remember some AMD cards had this fairly recently and it was a driver bug. Could be doing so if one of the cards is a little faulty as well. Out of curiousity, can you also check your cpu clocks while this is happening? Also, are you or have you tried running with the case open for a trial to provide more ventilation?

 

[Silverforce11]

VRMs are atrociously bad on reference gtx670, because the heatsink over the vrm (of which there's barely enough and without sensors!!) takes in hot air exhaust from the GPU heatsink... normally its the reversed.

I recently did a AIO water cooler mod for a 670 and noticed the VRM heatsink even with cool air was extremely hot (enough to burn your finger on touching). They are much hotter than my R290 VRM sink under mining load with a reported 85C.

Your throttling is definitely thermal based if increasing the fanspeed alleviates the problem. Seeing as its not the core temp, it therefore most likely is the VRM throttling clocks to prevent itself from exploding.

 

[Carfax83]

Your throttling is definitely thermal based if increasing the fanspeed alleviates the problem. Seeing as its not the core temp, it therefore most likely is the VRM throttling clocks to prevent itself from exploding.

I thought it was a heat problem as well at first, but when he said that he didn't have any problems running stress tests like Furmark, then I began to have doubts.

A thermal problem should manifest itself in Furmark, as that tends to push video cards pretty hard.

 

[Silverforce11]

Furmark is throttled by both NV and AMD for awhile now, its detected on the driver level.

BF4 also hammers the CPU pretty hard, so it gets hot, exhaust hot air into the case, much higher ambients + hot VRMs on SLI with a poor reference cooler is not a good recipe.

 

[OCGuy]

Low settings may actually be causing a CPU bottleneck. What is up with the refresh rate in the screenie?

Windows 8 seems to be the common fix after reading threads on other forums that are similar.

85c for a GPU is perfectly fine.

 

[richaron]

Moar case fans.

 

[tweakboy]

Try without SLI ,,, one card at a time.. gl

 

[TH3R4POR]

Hi guys, thanks for all the feedback!

The PSU is a Corsair CX750M, newegg states SLI, but Corsair on their website (and the packaging) don't.

Cables are all seated fine, and I use a Kuhler H2O as a CPU cooler. It is possible I need more airflow, but testing last night I had the case open and a 120mm fan sat next to the cards.

I played for a bit with single cards in, and alternated them, both behaved on Test Range. I had unplugged all my USB devices before doing this (Belkin n52, xbox pad, SD card reader, nothing taxing), so I don't know if that helped (PSU?).

Now I've put both cards in SLI again, and now they seem to behave on Test Range (all USB devices plugged back in), but when I'm ingame I get the same drops down to 15 FPS. The 15FPS was with and without USB devices.

It did seem a little smoother with single cards and the gfx turned down, but that then highlighted a packet loss issue I seem to get sporadically (I had put this down to the FPS drops).

One thing I noted on the Test Range, single card in and Settings on Maximum and the FPS was stable at 90, I look at the targets on the firing range and it drops to mid 50s, look away back to 90, look back to targets back to 50. Obviously the targets are work for the system, but that seemed like a large drop?

Getting fed up of testing now to be honest, I have spent hours on the Test Range and hours with 15FPS ingame getting bashed, the PSU is brand new otherwise I'd just bite the bullet and get a 1000W one to make sure.