Battlefield 3 Alpha Testing - GPU & CPU

[RussianSensation]

Please use Google Chrome Translator:

GPU Testing Preview:
http://gamegpu.ru/Action-/-FPS-/-TPS/Battlefield-3-Alpha-test-GPU.html

* Note: Less powerful video cards at the level of GeForce GTS 450 and the Radeon HD 4870 were tested with the Intel Core 2 Quad Q9550 @ 4.25 GHz and Phenom II X4 940 BE @ 3.7ghz, while more productive videocards were paired with more powerful CPUs (i.e., Intel Core ™ i7-930 @ 4.2 GHz, Intel Core i5-760 8.2 @ 4.2 GHz).

* Note 2: The testing procedure involved picking servers with 0 participants. This ensured identical testing conditions, but it also means that the results represent the best case scenario and aren't truly representative of a full scale multi-player game.

DX10 performance

DX11 performance

CPU Bench

Keep in mind this is still BF3 Alpha, but it looks like it's finally the End of the Line for dual core processors.

Some notable standouts:

- HD5870 is way faster than the HD6870 and even beats the 6950
- HD5870 2GB Eyefinity is no faster than the HD5870 1GB, highlighting that it's unlikely that VRAM is an issue at 1920x1080.
- At the moment, Frostbite 2.0 does not support Cross Fire and SLI, but will be fixed with newer drivers

 

[TakeNoPrisoners]

It looks like even six cores has a noticable benefit in this game. Look for this trend to continue. With CPUs moving towards more and more cores it is inevitable that the dual core will be slowly phased out.

What confuses me is the fact that the 5870 is faster than the 6950.

 

[Grooveriding]

I usually keep HT off. Turning it on I saw better framerates in the alpha. Time to get yourselves a quad core if you don't already have one

 

[Chiropteran]

VRAM is an issue at 1920x1080.
- At the moment, Frostbite 2.0 does not support Cross Fire and SLI, but will be fixed with newer drivers

Why does the 6990 beat every other card at max fps if crossfire isn't supported?

 

[Grooveriding]

Why does the 6990 beat every other card at max fps if crossfire isn't supported?

Crossfire and SLI both weren't working in the alpha, look at how poorly the 590 is doing.

It's like most games where it's not working properly, it's not completely broken, but scaling is much worse than it should be.

 

[Spjut]

What strikes me with the benchmark is the abysmal performance of the i3, it's getting beaten by the E8400! And I had guessed that hyperthreading would help those CPUs out

On the DX10 side, I really wonder what's up with AMD's GPUs' performance...The GTS250 is performing very near the superior HD4870(which also should benefit of its DX10.1 support) and the weaker GTX260 model beats out the HD4870, and the HD4890 gets beaten by the GTX260 216 cores...More that's going on in DX10.1? It seems strange though

AMD and Nvida seems pretty even on the DX11 side though....Has AMD perhaps enabled DX11 multithreaded rendering for Battlefield 3 only, as Nvidia did for Civilization V before they got full multithreading support? I had guessed Nvidia's MTR support would give their cards a noticeable lead, but that doesn't seem to be the case

 

[Spikesoldier]

so in the last graph, why would a lynnfield with no HT beat a nehalem w/ HT clock for clock?

 

[RussianSensation]

so in the last graph, why would a lynnfield with no HT beat a nehalem w/ HT clock for clock?

I haven't done the testing myself. Although, in certain rare cases CPUs with HT perform worse than without it. Again, it's only in Alpha testing stage. But looks like dual core CPUs are obsolete.

 

[BFG10K]

Keep in mind this is still BF3 Alpha, but it looks like it's finally the End of the Line for dual core processors.

End of the line? How are we inferring that? There are no Sandy Bridge dual cores in the test, and we know those are competitive with an i5 750/760 in gaming.

Not to mention that BC2 was later patched to have better dual-core performance.

 

[RussianSensation]

End of the line? How are we inferring that? There are no Sandy Bridge dual cores in the test, and we know those are competitive with an i5 750/760 in gaming.

Core i3-530 @ 2.93ghz = 26 fps avg / 20 fps min < -- unplayable (There were no characters on the map in their testing!)

SB is 15&#37; faster per clock than i3 is. That's not going to be enough to make it playable. BF2 (in multiplayer) benefits significantly from quad cores already. This isn't a surprise.

This isn't the first game either where dual cores are serious bottlenecks. Even at 1920x1080 4AA, dual cores are half as slow in Dirt 3.

 

[Spikesoldier]

I haven't done the testing myself. Although, in certain rare cases CPUs with HT perform worse than without it. Again, it's only in Alpha testing stage. But looks like dual core CPUs are obsolete.

makes me think the testing methodology might be flawed.

if these benchmarks were done over a period of time, nightly build updates will cause an imbalance of game code optimization and fixes.

 

[Red Hawk]

I'm not going to be able to do native res, am I...:\

 

[Will Robinson]

Overclock that Q66 to 3.0GHz or better,immediately.
Buy a new Radeon once the game is released and we know the relative performance for all cards on the market.
Then you can do that native res.

 

[Ben90]

I haven't done the testing myself. Although, in certain rare cases CPUs with HT perform worse than without it. Again, it's only in Alpha testing stage. But looks like dual core CPUs are obsolete.

I agree with everything but would like to switch "rare" with often. Gamers generally benefit by turning HTT off.

Some games even go so far as to refuse to run on logical cores.

 

[Silverforce11]

1. CPU with 6 cores scale, that's amazing coding in the FB2 engine.
2. 5870 faster than 6950 = FB2 is very shader intensive OR un-optimized alpha. Performance delta with a few other cards suggest the latter.

Server with 0 players and 1080p with top of the line single gpu gets ~65 fps, put 64 players in there.. you are going to need CF/SLI, even for 1080p or turn down the settings.

 

[BFG10K]

Core i3-530 @ 2.93ghz = 26 fps avg / 20 fps min < -- unplayable (There were no characters on the map in their testing!)

SB is 15% faster per clock than i3 is.

Not in games it's not: http://www.xbitlabs.com/articles/cpu/display/core-i3-2120-2100_5.html#sect0

Performance gain of an i5 2120 (Sandy Bridge) over an i3 560 (which itself is 14% faster than the i3 530 used in BF3):

• 31% FC2.
• 39% Starcraft 2.
• 31% F1 2010.
• 18% Mafia 2.
• 44% Metro 2033.
• 22% Civ 2.

The dual-core Sandy Bridge even beats the i5 750 in Starcraft 2 and Civ5. Overall the i5 2120 is competive with the i5 750 in those tests, so there's no reason to believe it won't be in BF3. Of course you can't infer much from an Alpha.

 

[StrangerGuy]

I got my new 5850 for half the price of a 560 Ti and its still 89&#37; as fast!

 

[Silverforce11]

I got my new 5850 for half the price of a 560 Ti and its still 89% as fast!

It's a bargain considering 5850s will do a 25% OC easily.

Should be enough on lower settings in a real server.

 

[-Slacker-]

Doesn't look like the murderous resource hog that people thought it would be ... and this is just the alpha.

 

[Genx87]

I am interested in seeing what this looks like with the final product. Give us a comparison of performance increases through the development cycle.

 

[Red Hawk]

Overclock that Q66 to 3.0GHz or better,immediately.
Buy a new Radeon once the game is released and we know the relative performance for all cards on the market.
Then you can do that native res.

Can't overclock, Dell bios doesn't allow it. I'll just have to get a full system upgrade in a couple years...

 

[Blastman]

What strikes me with the benchmark is the abysmal performance of the i3, it's getting beaten by the E8400! And I had guessed that hyperthreading would help those CPUs out

Almost for sure something is messed up with HT support and/or the threading needs to be fixed. Could be why the i7-930 is slower than the i5-750 too. It also doesn't make much sense that a E8400 is running 28 fps and a i5-750 is running 168 -- over 5x as fast.

In BF:BC2, hyperthreading boosts performance on the i3-530 by ~ 50-60&#37; &#8230; overclockers &#8230; scroll down about halfway in that thread for benches of the i3-530 with/without HT. HT makes a huge difference on dual cores. The i3-530 looks to be averaging below 40 (with several dips in the low 20's) without HT, and close to 60 with HT on with only 1 dip slightly below 40. It's the difference between unplayable and half decent frame rates for BC2.

In this Techreport BF:BC2 test, the i3-560 (3.3Ghz) is 81% faster than the Clarkdale dual core without HT G6950 (2.8Ghz) with an only 18% clockrate advantage. The i3-560 also easily outran the Q9400 (2.66Ghz) in that BC2 benchmark. Considering the Clarkdale i3's overclock to 4.0Ghz easily they should perform quite well if BC2 is any indication.

This isn't the first game either where dual cores are serious bottlenecks. Even at 1920x1080 4AA, dual cores are half as slow in Dirt 3.

From my observation of benches on the dual core i3's with HT, multithreaded gaming is one of the areas (along with video encoding) where HT makes some of the biggest performance gains. The dual core i3's are generally substantially faster than the previous generation of dual cores in multithreaded games. An i3-540 is almost as fast as a Q9550 in Dirt 2, and substantially faster than a higher clocked E8600 &#8230;

Dirt 2

Phenom x4 925 &#8230; 99.45
Q9550 &#8230;&#8230;&#8230;&#8230;&#8230;.. 94.51
i3-540 &#8230;&#8230;&#8230;&#8230;&#8230;.. 90.43
E8600 &#8230;&#8230;&#8230;&#8230;..&#8230; 73.40

It isn't going to take much of an overclock for the i3-540 to outrun the Q9550, and the i3 mops the floor with the slower dual cores under the E8600.

In this computerbase i3-2100 test, HT boosted performance 71% in the game F1 versus no HT. There are quite a few other multithreaded games where the HT on i3's give it a substantial performance boost >30%.

Surprisingly, in the same game F1, the i7-2600K loses 4% performance by enabling HT &#8230; computerbase i7-2600K. Unfortunately, these i7 results give the impression that HT does nothing in gaming, or actually hurts gaming performance because many times we only see the results of the i7's. It's possible that for the i7, 4 cores are enough to handle the workload and adding HT only adds to thread thrashing, reducing performance. There are other games where the i7-2600 drops in performance (-5%) with HT on in that same computerbase review, like RE5, where the i3-2100 sees a 46% gain with HT on.

From the reports I've seen in the forums, the i3 Clarkdales O/C to 4.0Ghz run BF:BC2 multiplayer "butter" smooth. So I think the i3's should do quite well in BF3 once things are fixed. As for the other dual cores -- we'll have to see.

 

[happy medium]

I got my new 5850 for half the price of a 560 Ti and its still 89&#37; as fast!

Dam a 5850 for 97$! you got a steal!. And he is right the 5850 overclocks just as much as a gtx560ti, nice.

 

[Silverforce11]

Doesn't look like the murderous resource hog that people thought it would be ... and this is just the alpha.

Top single GPU cards gets a bit above 60 fps average at 1080p, with ZERO players.

Now, put in 64 players, jets, tanks, explosions (a lot of dust/smoke) going off everywhere, terrain deformation, bullet fragmentation vs rocks/walls, and total destruction of buildings..

I'd say its gonna be a resource hog.

Just all the physics involved is going to hammer non-quad core CPUs.

 

[Mana]

Top single GPU cards gets a bit above 60 fps average at 1080p, with ZERO players.

Now, put in 64 players, jets, tanks, explosions (a lot of dust/smoke) going off everywhere, terrain deformation, bullet fragmentation vs rocks/walls, and total destruction of buildings..

I'd say its gonna be a resource hog.

Just all the physics involved is going to hammer non-quad core CPUs.

I'm going to disagree with you there. The general consensus from people on another forum I frequent who were in the alpha is that the system requirements are similar, if slightly higher, than Bad Company 2 while looking almost twice as good.