BF3 patch includes 7xxx series perf improvements

[djnsmith7]

I didn't have time to take any screenshots or benchmarks before the patch, so I can't say for sure, but my sentiments are in line with those saying the framerates increased. I won't say I'm impressed with the numbers, but I'm hoping more efficient (mature) drivers will increase the Min & Avg figures.

My FRAPS numbers after the patch:

1x 7970 (1125 / 1575):

Min - 47
Avg - 110.7
Max - 186

2x 7970 (1125 / 1575):

Min - 61
Avg - 135.6
Max - 201

Update:

Using the numbers above for comparison, I see very little difference going from CF to TriFire on 64 MP maps (tried 3 different 64 MP maps that were full). What's also interesting is that my GPU % usage numbers have dropped from 95+% to ~75% (2 > 3 cards).

Now, this would lead me to believe that what others have said about being CPU bottlenecked would be true.

However, when I play single player, all 3 cards are maxed & the numbers look like this:

GPU1 - 99%
GPU2 - 99%
GPU3 - 99%

Min. - 180
Avg. - 190
Max. - 200

This leads me to believe I'm not CPU bottlenecked, but the MP portion of the game is poorly written (which I've seen others say & maybe I'm stating the obvious).

So, which is it?

 

[3DVagabond]

Update:

Using the numbers above for comparison, I see very little difference going from CF to TriFire on 64 MP maps (tried 3 different 64 MP maps that were full). What's also interesting is that my GPU % usage numbers have dropped from 95+% to ~75% (2 > 3 cards).

Now, this would lead me to believe that what others have said about being CPU bottlenecked would be true.

However, when I play single player, all 3 cards are maxed & the numbers look like this:

GPU1 - 99%
GPU2 - 99%
GPU3 - 99%

Min. - 180
Avg. - 190
Max. - 200

This leads me to believe I'm not CPU bottlenecked, but the MP portion of the game is poorly written (which I've seen others say & maybe I'm stating the obvious).

So, which is it?

CPU bottlenecked in MP is what I read from this.

 

[BallaTheFeared]

So, which is it?

CPU bottleneck.

Time for Ivy Bridge!

 

[Grooveriding]

Update:

Using the numbers above for comparison, I see very little difference going from CF to TriFire on 64 MP maps (tried 3 different 64 MP maps that were full). What's also interesting is that my GPU % usage numbers have dropped from 95+% to ~75% (2 > 3 cards).

Now, this would lead me to believe that what others have said about being CPU bottlenecked would be true.

However, when I play single player, all 3 cards are maxed & the numbers look like this:

GPU1 - 99%
GPU2 - 99%
GPU3 - 99%

Min. - 180
Avg. - 190
Max. - 200

This leads me to believe I'm not CPU bottlenecked, but the MP portion of the game is poorly written (which I've seen others say & maybe I'm stating the obvious).

So, which is it?

There is a thread on hard or ocn with tests comparing 2500k to 4c/8t 2600k in 64mp bf3 showing the 2500k cpu choked. BF3 benefits from HT and cores/threads. Fwiw all my threads sit at 25% usage give or take in mp with no slowdowns. 2600k for bf3!

 

[BallaTheFeared]

lol... because 135+ avg isn't enough!

 

[skipsneeky2]

Maybe i'm not hard to please,i just got a single 7970 last night a msi oc edition at 1010 core,my 2500k seems to hold up fine i guess at stock clocks and the performance blows my mind in BF3 with no potential bottlenecks thus far.

Still need a good motherboard and a quality psu before i rock some clocks.

 

[railven]

lol... because 135+ avg isn't enough!

No shit! Damn! Haha.

Guy better be running 3x30" Displays!

 

[Rvenger]

There is a thread on hard or ocn with tests comparing 2500k to 4c/8t 2600k in 64mp bf3 showing the 2500k cpu choked. BF3 benefits from HT and cores/threads. Fwiw all my threads sit at 25% usage give or take in mp with no slowdowns. 2600k for bf3!

Yes, but the 2600k is known to choke as well, just not as bad as the 2500k. Its said that the 2600k has a ~ 5-10% advantage in BF3 over the 2500k. The fact of the matter is that every desktop level CPU besides SB-E will choke using more than 2 cards. IMO If you want the extra horsepower and you already paid $1100 for 2x 7970s, then you can afford to get a 3930k to pull the weight of those cards.

Also, my belief is that BF3 is poorly optimized in general, I don't think the cores/threads are being used properly. I think its allocating the CPU usage for background instances in BF3 causing a harsh spike in usage under heavy map areas, similar to how it likes to allocate all your VRAM. Did they even fix that issue?

I know what I said sounds somewhat redundant but for a poorly optimized game, it requires more horsepower to receive desired results.

 

[Grooveriding]

Yes, but the 2600k is known to choke as well, just not as bad as the 2500k. Its said that the 2600k has a ~ 5-10% advantage in BF3 over the 2500k. The fact of the matter is that every desktop level CPU besides SB-E will choke using more than 2 cards.

Also, my belief is that BF3 is poorly optimized in general, I don't think the cores/threads are being used properly. I think its allocating the CPU usage for background instances in BF3 causing a harsh spike in usage under heavy map areas, similar to how it likes to allocate all your VRAM. Did they even fix that issue?

The VRAM usage is intended. There is a youtube link of one of their tech guys explaining it, will add the link later.

 

[Rvenger]

I also don't believe that the 2600k uses 25% usage in BF3. Not in my experience. It may use 25% usage looking into the sky but not on top of a building in Gulf of Omen.

Try 80% with spikes into the 90's.

http://hardforum.com/showpost.php?p=1038342174&postcount=67


EDIT: Grooveriding, I see you have a 3930k and that would explain your usage I am just talking about a 2600k. Sorry about that.

 

[Grooveriding]

I also don't believe that the 2600k uses 25% usage in BF3. Not in my experience. It may use 25% usage looking into the sky but not on top of a building in Gulf of Omen.

Try 80% with spikes into the 90's.

http://hardforum.com/showpost.php?p=1038342174&postcount=67


EDIT: Grooveriding, I see you have a 3930k and that would explain your usage I am just talking about a 2600k. Sorry about that.

http://www.youtube.com/watch?v=KMaL6j7Ry6c

He starts in on the memory usage around 7:00min. I think in my limited understanding, it is supposed to adapt to what is available.

 

[Rvenger]

http://www.youtube.com/watch?v=KMaL6j7Ry6c

He starts in on the memory usage around 7:00min. I think in my limited understanding, it is supposed to adapt to what is available.

Probably helps performance and eliminates the texture "pop in" Ill take a look at that video later. Thanks for posting it :thumbsup:

 

[skipsneeky2]

I found gpu vram usage to be almost identical going from my 2gb gtx560 to my 7970.

1080p ultra 2x msaa post aa low with motion blur shut off it seems to hit about 1250mb max but enabling 4x msaa it hits 1330mb+ immediately especially on maps like Caspian Border and setting it back to 2x msaa it goes right back to 1230-1250mb.

 

[chipachap]

No offence to anyone but all you peeps that are saying the 2500k is bottle necked in BF3 you don't have a clue what you're talking about. I have a 2500k at 4.8 Ghz and 7970 xfire and the only time i get dips like these is in 1080p, 60hz or 120hz.

Only in heavy 64 player conquest do mine go 97-99% to keep it around 80fps which is usually the lowest it goes when they're at 99% each. In small maps they only stay around 80% and the bit that makes it obviously not a 2500k bottleneck is that when i turn off MSAA I get the same frames but GPU usage goes down to 65-70%.

Anyone with 2500k and 7970 xfire/trifire will have the same results as me.

In Eyefinity both GPU's were pegged at 99% the whole time and the same at 2560X1440, I tested on a friends monitor just incase. It's clearly some bug in BF3 alone because in Crysis 2 and The Witcher 2 gpu's are at 99% the whole time aswell.

I thought once the cards were above the reg 60fps they dialed themselves down to save power, but it's the same when I went to 120hz.

It's either an AMD driver/zero core/power saver bug OR it's a BF3 1080p utilisation bug, please everyone don't scrap your 2500k that's not the problem here.

I am constantly searching for a fix and I will post back if I find one. I don't mean to be rude but you shouldn't post information you don't know to be 100% true IMO.

 

[djnsmith7]

^Appreciate your input...

I'm still sorting through some data & trying to figure out what's what...In the meantime, I'll share what's raising an eyebrow on my end.

Starting with BF3 SP (data from MSI AB):

I figure, hey, it could be an anomaly that all 3 cards are pegged at 99% through most of the duration of the 15 minute test (listed the min., avg., & max. above - solid numbers).

Moving on to Unigine (Heaven) (data from MSI AB):

All 3 cards were above 95% for a good 80% of the benchmark. In my mind, that wipes out the anomaly theory, as we now have 2 data points.

So, in this very short & unfinished conclusion, it's difficult for me to side with the folks that are saying it's a CPU bottleneck when there are 2 completely separate data points showing otherwise.

I'd like to add a 3rd data point in here (3DMark11), but I haven't been able to capture the 3x GPU usage yet during the benchmark. Once I do, I'll share it.

What I will share is my initial 3DMark11 score breakdown:

Overall: P17135
Graphics: 27259
Physics: 8357
Combined: 7755
GT1: 119.89
GT2: 137.88
GT3: 167.25
GT4: 82.12
PT: 26.53
CT: 36.07

My P Score is tied for 3rd on the list if you do a search for 2500k / 7970 / Performance.

 

[Grooveriding]

The issue is that 64 player MP in BF3 is CPU stressful. For example, my system performed identically with a 4ghz 920 as it did with a 4Ghz 3930K in SP

But see the MP results

In most cases there are not many stark bottlenecks,

when you are dealing with 3 cards though, there will be some.

Note these benches are all 2560x1600 highest possible settings.

 

[djnsmith7]

Groove, you make a great point. One of which I wasn't completely capturing until you highlighted it with these benchmarks. I think we're (including the others on the bottleneck train) now on the same page.

I've been saying pretty much all along that I thought the problem was with BF3 MP. To add some humor (in my stubborn ways), I see this as a BF3 MP bottleneck, not necessarily a CPU bottleneck...ergo, if the game were written more efficiently, there wouldn't be a bottleneck. <evil grin>

 

[MrK6]

The issue is that 64 player MP in BF3 is CPU stressful. For example, my system performed identically with a 4ghz 920 as it did with a 4Ghz 3930K in SP

Note these benches are all 2560x1600 highest possible settings.

Awesome benchmarks, Groove :thumbsup:. It just goes to show how CPU-dependent some games are.

 

[chipachap]

Well a 3930k is quite more performance based than a 920 (correct me if i'm wrong) but 2 more cores shouldn't cause a 33% increase in BF3 mulitplayer unless a 920 under performs for this game. I tried my i5 at 5Ghz and at 4ghz and there was only about a 5 fps difference for minimum, which was between 73 and 78 so it's only a couple of %.

But as i said the cores average about 78-82% with 1 core hitting 90% every now and again. It's just some stupid bug in the game, any other game I played doesn't have it. Except for MW2 MP but the cards are at 45% BUT it keeps me above my target 120fps all the time. It's like at 120hz in BF3 mp it doesn't recognise that you are trying to hit 120fps it acts lie it's target is above 60 and that's it, once it goes above it dials down the cards.

If anyone wants me to do any more tests, just ask.

 

[skipsneeky2]

Well a 3930k is quite more performance based than a 920 (correct me if i'm wrong) but 2 more cores shouldn't cause a 33% increase in BF3 mulitplayer unless a 920 under performs for this game. I tried my i5 at 5Ghz and at 4ghz and there was only about a 5 fps difference for minimum, which was between 73 and 78 so it's only a couple of %.

But as i said the cores average about 78-82% with 1 core hitting 90% every now and again. It's just some stupid bug in the game, any other game I played doesn't have it. Except for MW2 MP but the cards are at 45% BUT it keeps me above my target 120fps all the time. It's like at 120hz in BF3 mp it doesn't recognise that you are trying to hit 120fps it acts lie it's target is above 60 and that's it, once it goes above it dials down the cards.

If anyone wants me to do any more tests, just ask.

Good quality tests using ht/no ht stock clocks and clocked to 4ghz would be very on a single gpu as right before i purchased my 2500k i had to start a i7 2600k vs i5 2500k thread cause the rumor BF3 takes advantage of ht became a bit much and the main consensus seemed to be that ht made no difference for a single high end or mid range gpu at 1080p or any common resolution.

So test on brother man,debunk it once and for all.:thumbsup:

 

[MrK6]

Well a 3930k is quite more performance based than a 920 (correct me if i'm wrong) but 2 more cores shouldn't cause a 33% increase in BF3 mulitplayer unless a 920 under performs for this game. I tried my i5 at 5Ghz and at 4ghz and there was only about a 5 fps difference for minimum, which was between 73 and 78 so it's only a couple of %.

But as i said the cores average about 78-82% with 1 core hitting 90% every now and again. It's just some stupid bug in the game, any other game I played doesn't have it. Except for MW2 MP but the cards are at 45% BUT it keeps me above my target 120fps all the time. It's like at 120hz in BF3 mp it doesn't recognise that you are trying to hit 120fps it acts lie it's target is above 60 and that's it, once it goes above it dials down the cards.

If anyone wants me to do any more tests, just ask.

BF3 runs by six threads I believe, hence the advantage of the 3930K.

 

[djnsmith7]

To keep the pedal to the metal, I'll remind folks that when using 2 cards in MP, they're between 95 - 99%. It isn't until the 3rd card is brought into the mix that the numbers drop to 55 - 75%.

 

[chipachap]

@ djnsmith, can you do me a favour? Could you use two cards and turn off MSAA on a 64 player map and tell me what your GPU usage goes down to?

@MrK6, but 2 more cores hardly give a plain as day 33% increase? Surely the 920 must not have been up to the task or something? Is there any graphs/reviews that show a 2500k vs a 3930k on 64 player MP? I'll have a look for one myself in the meantime.

When I have it on preset Ultra they stay >85% but if I turn onto preset low they go all the way down to 40% usage but give me the same frames. This is the key point (to me) that eliminates the cpu as the problem here.

 

[skipsneeky2]

@ djnsmith, can you do me a favour? Could you use two cards and turn off MSAA on a 64 player map and tell me what your GPU usage goes down to?

@MrK6, but 2 more cores hardly give a plain as day 33% increase? Surely the 920 must not have been up to the task or something? Is there any graphs/reviews that show a 2500k vs a 3930k on 64 player MP? I'll have a look for one myself in the meantime.

When I have it on preset Ultra they stay >85% but if I turn onto preset low they go all the way down to 40% usage but give me the same frames. This is the key point (to me) that eliminates the cpu as the problem here.

I think there's a issue with frostbite engines and the drivers as my 7970 and BC2 give identical performance between medium 1080p and fully maxed out,i also have low gpu usage.

BF3 seems to be fine for me and both 12.2 and 12.3 seem to do this but i see a slight improvement with 12.3.

 

[djnsmith7]

@ djnsmith, can you do me a favour? Could you use two cards and turn off MSAA on a 64 player map and tell me what your GPU usage goes down to?

@MrK6, but 2 more cores hardly give a plain as day 33% increase? Surely the 920 must not have been up to the task or something? Is there any graphs/reviews that show a 2500k vs a 3930k on 64 player MP? I'll have a look for one myself in the meantime.

When I have it on preset Ultra they stay >85% but if I turn onto preset low they go all the way down to 40% usage but give me the same frames. This is the key point (to me) that eliminates the cpu as the problem here.

Chip,

When I turn down the settings (I have them on Ultra now), the % decreases with either 2 or 3 cards.