GTX 670 SLI Benchmarks.

[alienware]

Hi all. Kambo and I agreed to run some benchmarks in another thread so I've spent the afternoon running plenty of benchmarks. Note I have not ran them on a single GPU as most of them (with the settings I have used) fail to make playable frame rates.

First of all this might be of interest (Frame rating)

http://forum.overclock3d.net/showthread.php?t=52258

So let's move on, shall we?

Firstly I want to make a note of scaling percentages.

Nvidia's drivers have an inbuilt tool that reads the GPU usage on SLI and displays it as a single bar. This bar then reaches up and down from the center point and displays the usage as a green bar from the center. The top center bar is GPU1 and the bar from the bottom is GPU2. Note here are the bars being used to 100% over both GPUs (in 3Dmark 11)



And here they are in Dirt : Showdown (around 85% usage)



And here they are in Mafia 2, which tbh scales very poorly (you will see this in the results).



Our Test System.

Today I have used my own rig, pictured here



The specs.

Alienware Area 51 ALX with 1200w PSU
MSI Z77A-G45 motherboard
Intel Xeon E3-1220 (Sandybridge) CPU clocked @ 3.4ghz
12GB Mushkin Radioactive/Blackline
120GB Corsair SSD for boot drive
Numerous 500GB drives to store games
Corsair H60 cooler single fan

Some information

CPU (Note CPUZ doesn't understand it properly so it's all out of whack)



And our GPU set up. SLI 670s running @ 8x PCIE 2.



And our recorded temps after an hour of benchmarking.



The Benchmarks

First up was 3Dmark Vantage,noting this is an old version that will add Physx PPU scores to the CPU score so ignore the CPU score, basically.



3Dmark 11



3DMark (13)



Unigene Valley 1.0 Note -ran on Extreme HD setting



Unigene Heaven 3.0 settings



Unigene Heaven 3.0 result



Unigene Heaven 4.0 settings



Unigene Heaven 4.0 result



The games... Dirt Showdown results. Game ran on max with 8XMSAA



Hitman : Absolution settings NOTE I am running 8XMSAA here



Hitman : Absolution result



Mafia 2 settings



Mafia 2 result



Metro 2033 settings



Metro 2033 result



Streetfighter X Tekken settings



Streetfighter X Tekken result



Tomb Raider results, max possible settings



Sleeping Dogs settings, 8xMSAA Note, FRAPS can not take shots of this with SLI running so I had to take photos.



Sleeping Dogs results



Crysis 3 settings



Crysis 3 chosen level for manual benchmark



Crysis 3 result



Crysis settings & Benchmark result



Batman Arkham City settings



Batman Arkham City results



Just Cause 2 settings 8XMSAA



Just Cause 2 result



Far Cry 3 4XMSAA max settings



Far Cry 3 chosen level. I decided to ride around on a quad bike



Far Cry 3 result



Final thoughts.

So to those out there that say that SLI is a waste of time on one 1080p monitor? well, up until the recent bout of games released (such as Crysis 3, Sleeping Dogs, Hitman, Tomb raider and so on) I would have agreed with you. However, as can clearly be seen some of these titles are barely playable even on SLI. Returning to one card produces unplayable frame counts in most of the new titles, though granted that's mostly when you crank the MSAA.

Please feel free to run these benchmarks yourself for comparison reasons. Any results can help but please be careful to make sure you use identical settings otherwise results will be useless.

Cheers!

 

[MrKambo]

7970 Benchmarks

Hi guys, as AlienALX said we agreed to throw some benchies around after a discussion about some silly results in Tomb Raider

The test system i am using is this: 'Assassin'

Specs:
Intel i7-3770K @ 4Ghz
Asus Maximus V Formula
MSI R7970 Crossfire @ 1010/1375 (Stock)
8GB Corsair Dominater Platinums 2133 @ XMP
120GB Corsair Force GT
Silverstone Strider Gold Evo 850w

As you can see im using the AMD 13.3 drivers and if im honest they are pretty damn good from previous versions

Synthetic Benchmarks

3D Mark Vantage:

3D Mark 11:

3D Mark(13):

3D Mark(13) Extreme:

Unigine Valley:

Unigine Heaven 3:

Unigine Heaven 4:

Games:

Batman Arkham City:

Hitman Absolution:

Sleeping Dogs:

Tomb Raider:

 

[alienware]

Wow the extra 600mhz on the CPU really helps out !

My Physics scores in synthetic benchmarks suck even though I get pretty much the same graphics score as you.

Hmm maybe one day I'll track down a 2500k, not that I'm confident it will really help out in gaming tbh. Might do, but I'm not spending £100+ just for 3 extra FPS.

Edit. DOH just realised you're running a 3770k not the 3570k I thought you had.. My old I7 950 used to do well in Physics thanks to the HT.

 

[NeverBackDown]

It will probably raise your minimum fps more than your max fps.

 

[MrKambo]

Wow the extra 600mhz on the CPU really helps out !

My Physics scores in synthetic benchmarks suck even though I get pretty much the same graphics score as you.

Hmm maybe one day I'll track down a 2500k, not that I'm confident it will really help out in gaming tbh. Might do, but I'm not spending £100+ just for 3 extra FPS.

Edit. DOH just realised you're running a 3770k not the 3570k I thought you had.. My old I7 950 used to do well in Physics thanks to the HT.

Synthetic benchies love more of everything...more cores it will eat them alive, higher clocks it will use every mini-mhz it can get, thats the way they're designed

Changing CPU will only help in terms of minimum FPS, these 3770k are supposed to be ridiculous at physics, but i haven't had a chance to clock mine up yet, no cooling

It will probably raise your minimum fps more than your max fps.

+1 this

 

[alienware]

It will probably raise your minimum fps more than your max fps.

Yeah very true. Thankfully the minimums are just about there tbh.

Crysis 3 seems to take a while to 'warm up'. Not quite sure what it's doing but once you start running around and everything has time to buffer the FPS really improve. I deliberately chose a really hard to run level (you're being bombed so there's tons of explosions) as the level I was on I kept dying on before the bench could finish :lol:

I did overclock the GPUs when I got them and my GPU scores really went up. So it's only CPU intensive tasks that they get held back.

 

[MrKambo]

Yeah very true. Thankfully the minimums are just about there tbh.

Crysis 3 seems to take a while to 'warm up'. Not quite sure what it's doing but once you start running around and everything has time to buffer the FPS really improve. I deliberately chose a really hard to run level (you're being bombed so there's tons of explosions) as the level I was on I kept dying on before the bench could finish :lol:

I did overclock the GPUs when I got them and my GPU scores really went up. So it's only CPU intensive tasks that they get held back.

how did you run the Crysis 3 benchmark, thought there wasnt a tool for it yet?

 

[alienware]

I had to use FRAPS and set it to 45 seconds because on 90 and 60 I kept dying before the benchmark was over :lol:

If you load FRAPS and then start the level you press F11 and it dumps out some text files

 

[MrKambo]

ahhh ok, ill look into that, just waiting for origin and steam to finish downloading games

 

[alienware]

Yeah loads of mine decided to bloody update right as I went to run them.

I also benched Just Cause 2 but forgot to transfer the pics over to JPG.

 

[alienware]

More benchmarks added

Just realised I don't have BF3 installed

 

[MrKambo]

batman is probably the shittest coded game ever!! it was using 30% of both GPU during the benchmarks, i ran it like several times incase i had messed something up!

 

[alienware]

Yeah same with Mafia 2 tbh. Odd because the first Batman worked with Quadfire !

 

[Scoob]

@ AlienALX

As you're running 670 SLI I'd suggest you DISABLE PhysX in any title that uses it, so Mafia II and Just Cause 2 - the latter uses it for water.

In my own testing (680 SLI) I found that PhysX only runs on ONE GPU, and badly effects FPS. With a decent CPU, letting it do the Physics seems to be less of a hit than having a GPU do it. I.e. if one of your GPU's dedicates say 30% of it's power to do PhysX, you start to see the unburdened GPU being held back by the other. Basically the PhysX GPU might be at 99% load, as it's doing render and PhysX, whereas the other is at 69% load. Let the CPU handle Physic and BOTH GPU's might be at 99% load with a healthy FPS boost.

It really does depend on how the game uses resources, but in both Mafia II and Just Cause 2 I get MUCH better FPS. In some games you can of course totally loose certain effects by disabling PhysX, however usually it just means the CPU does it, and a good CPU can manage just fine.

Have a play & see if your results line up with mine. I can post some numbers from a handful of the titles you list if you like.

Scoob.

 

[alienware]

Hmm good advice there, might have to try that mate

Yeah get some benches up. Specs obviously needed.. Doing it more to help people than anything else

 

[Scoob]

Hmm good advice there, might have to try that mate

Yeah get some benches up. Specs obviously needed.. Doing it more to help people than anything else

Just popped the new 320.00 Beta's on my rig & ran some tests. Terrible results...then I remembered to re-enable SLI

I've only run 3D Mark 11 thus far, in both Performance and Extreme, results are:

P15125 and X6964

2500k @ 4.8 with 2x 680's at 1200 (boosted) core with 7ghz vRam (1752 in GPUz)

Going to run some Mafia II (No PhysX) and Just Cause 2 (no GPU water) next...

Edit:

Mafia II score:

159.7 fps - Rank: A

Same settings as you, but 1200p, no PhysX and V-Sync off (lol)

Just Cause 2 - Dark Tower Bench:

179.71 fps

Same settings as you but 1200p, no GPU Water and V-Sync off.

As an aside, I usually run at 32xCSAA and V-Sync On.

Btw: your resolution is weird in the results, 1842 x 1026, whut?

3d Mark 2013

I tried to run the Firestrike test but I get very low (single figure) FPS. Checking further my GPU's never go above 60% load and my CPU stays at idle (1600mhz) - most odd!

Scoob.

 

[alienware]

Btw: your resolution is weird in the results, 1842 x 1026, whut?

Overscan? or is it called underscan?

Basically right now I am living in a 1920s farm house with my fiancee who part owns the house with her soon to be ex husband (they've been split for over three years he's off in Germany).

Because of that (I'm moving out with her once the divorce goes through, soon, basically) I haven't brought over my desk or monitors. So basically I'm stuck with a TV on a table I sit 13 ft away from and the TV like many others doesn't 'do' a full 1080p. I can run it at 1080p but the picture is then too big for the screen...

It makes about a 1% difference in frame rates.. I read about it once, and TV manus like to shave a few mm off here and there as it basically saves them money in the long run. It's a con, basically.

Until we've moved back to West Sussex where I live I can't set it up properly so I just use my lappy for t'internet and game on the desktop

 

[Scoob]

Ah, I see

I've actually still got the first TV I ever bought...a Matsui 14" Portable. It won't work without a coaxial digi-box, but I just cannot bring myself to get rid of it. Even though it's sat next to a 32" LCD in the bedroom lol. Saying that, I still have Videos on the shelf and there's not been a VHS in the house for years. I need a clear out lol.

Scoob.

 

[alienware]

It's a good TV (well, it cost a packet any way !) 50" Sharp Aquos.

It's just got a really odd actual resolution :lol: