GTX 670 4GB Tri-SLI

[Annisman*]

Third card on the way, I'll be benching to see how often the stock 2GB Vram limit is surpassed in games at 3 x 1080 resolution. What games should I bench, besides BF3/Crysis2 ? I pretty much have them all.

I'm going to be very intersted to see what Skyrim+ Bethesdas HD texture pack takes up in VRAM, as it already takes up close to 2GB in just 1080.

 

[jacktesterson]

I'd love to see 3DMark11 Performance Preset Scores.


As for games, I'd be interested to see how Metro 2033 runs.

 

[cmdrdredd]

Batman arkham city and heaven benchmark. If you have the witcher 2 run it at max settings with ubersampling and let us know what the fps is like. There is no built in benchmark but you could use fraps.

 

[Annisman*]

I'd love to see 3DMark11 Performance Preset Scores.


As for games, I'd be interested to see how Metro 2033 runs.

Is there a free/trial version of 3dMark11 I can DL ? And yeah, I should include Metro for sure.

 

[cmdrdredd]

Yeah at futuremark.com the basic benchmark is totally free.

 

[Annisman*]

Batman arkham city and heaven benchmark. If you have the witcher 2 run it at max settings with ubersampling and let us know what the fps is like. There is no built in benchmark but you could use fraps.

Just curious, why arkham city ? I mean I have it and will bench it for you, but why ?

I was under the impression that ubser sampling was just a gimmick in W2.

 

[cmdrdredd]

Ubersampling hogs memory that is why you should test it out. Batman because the benchmark is quality and gives repeatable results that reflect on other titles using unreal engine 3.

 

[Annisman*]

Kind of worried about the move to Tri-SLi after reading this fairly recent article: http://www.xbitlabs.com/articles/graphics/display/geforce-gtx-670-sli.html

Now, I understand that 3X cards does not = 3X the performance (I'm hoping for a 20-30% gain over SLI), but the move to a third card in these tests just doesn't make sense.

They were not however testing in triple monitor resolution, though they did test at 2560.

Well, either way I'm not backing out now, I will live or die by Tri SLI!

Edit: Ah, I remember the good old days when I did the 'unthinkable' and added a 48701GB to a 4870X2 for Tri-Fire, I was happy just to see a few fps increase across the board : )

http://www.tomshardware.com/forum/260564-33-4870x2-4870-fire-benchmarks-quad-core-results

 

[guskline]

:biggrin::biggrin:Annisman: First your rig is top notch! I can understand why you are running Tri-SLI with 6000x1080.

I just installed 2 EVGA GTX 670 FTWs in SLI to run 3 monitors at 5760 x 1080 in rig 1 below. The 2G vs 4G ( or even 6G vram on the Sapphire 7970) becomes relevant and important with multiple monitors in Surround. I was using the single PNY GTX 680 but wanted to run SLI for more FPS. There are few articles floating around the internet addressing the higher Vram issue.

I seriously doubt that with 3 670s with 4g Vram each you will have either a Vram issue or a FPS issue. BTW I know that running in SLI/CF doesn't double (or triple) the Vram.

Since I only put them in my machine last night and had a little time to play with them here are my observations so far.

1. BF3 played OK with a single GTX 680 at 5760 x1080 but really feels faster AND smoother with 2 GTX 670s in SLI (I have FRAPS but didn't check the difference. I wanted to play not benchmark.)

2 A single GTX 680 on rig 1 scored 9200 on 3DMark11 - Performance; a single 670 score 8930 but 2 670s in SLI scored 14,200! That's with no OCing.

3. Rise of Flight (777 Studios) was very smooth at 5760 x 1080 with a single 680 but was noticeably quicker with dual 670s.

I can only imagine what 3 670 4G vram GPUs in tri SLI will perform like.

Strap on your safety belt!

 

[KompuKare]

I'm going to be very intersted to see what Skyrim+ Bethesdas HD texture pack takes up in VRAM, as it already takes up close to 2GB in just 1080.

I'd be interested in the VRAM usage of that plus a few mods. Running Skyrim without user mods is boring: the whole point of Bethesda on PCs is the mods and that's why console's are so poor for their games.

As to which mods? Haven't really though of which ones, but there a number of mod packs / collections so maybe look at those. Also, there's no built-in benchmark and while running up to Dragonreach and doing a spin is ok, it's hardly repeatable. Real Carriages is a mod which can be used to bench and I started a thread over at nexusmods on that:

http://forums.nexusmods.com/index.p...hmarking-skyrim-with-the-sceniccarriages-mod/

so you might want to try that. Those carriage journeys are long but they are totally automatic

 

[Grooveriding]

Noticeable diminishing returns on the third card and even moreso on the fourth.

I could see 2GB being a VRAM bottleneck at that resolution in BF3 most likely and definitely in Skyrim with Mods.

I hit a VRAM wall with 2GB at my resolution with the right settings.

 

[Annisman*]

:biggrin::biggrin:Annisman: First your rig is top notch! I can understand why you are running Tri-SLI with 6000x1080.

I just installed 2 EVGA GTX 670 FTWs in SLI to run 3 monitors at 5760 x 1080 in rig 1 below. The 2G vs 4G ( or even 6G vram on the Sapphire 7970) becomes relevant and important with multiple monitors in Surround. I was using the single PNY GTX 680 but wanted to run SLI for more FPS. There are few articles floating around the internet addressing the higher Vram issue.

I seriously doubt that with 3 670s with 4g Vram each you will have either a Vram issue or a FPS issue. BTW I know that running in SLI/CF doesn't double (or triple) the Vram.

Since I only put them in my machine last night and had a little time to play with them here are my observations so far.

1. BF3 played OK with a single GTX 680 at 5760 x1080 but really feels faster AND smoother with 2 GTX 670s in SLI (I have FRAPS but didn't check the difference. I wanted to play not benchmark.)

2 A single GTX 680 on rig 1 scored 9200 on 3DMark11 - Performance; a single 670 score 8930 but 2 670s in SLI scored 14,200! That's with no OCing.

3. Rise of Flight (777 Studios) was very smooth at 5760 x 1080 with a single 680 but was noticeably quicker with dual 670s.

I can only imagine what 3 670 4G vram GPUs in tri SLI will perform like.

Strap on your safety belt!

Great insights :thumbsup:

 

[Annisman*]

Two more days and I'll have that third bad boy : )

 

[RussianSensation]

Shogun 2 on Ultra with MSAA
Max Payne 3 Maxed with MSAA
Witcher 2 with UberSampling

 

[PowerK]

Ubersampling hogs memory that is why you should test it out. Batman because the benchmark is quality and gives repeatable results that reflect on other titles using unreal engine 3.

Ubersampling in Witcher 2 doesn't hog memory. I've played with Ubersampling enabled on 690 Quad-SLI. Occational dips below 60 fps (into 50s) but mostly stays at 60 fps (I used adaptive vsync). Perfectly smooth at 2560x1600.

 

[cmdrdredd]

Ubersampling in Witcher 2 doesn't hog memory. I've played with Ubersampling enabled on 690 Quad-SLI. Occational dips below 60 fps (into 50s) but mostly stays at 60 fps (I used adaptive vsync). Perfectly smooth at 2560x1600.

So? Have you measured the memory usage?

 

[PowerK]

So? Have you measured the memory usage?

What do you mean so?
I posted my finding in "When can we expect to see Crysis (2007) level of visuals ?" thread which you participated, too.
This post in particular. http://forums.anandtech.com/showpost.php?p=33607027&postcount=212
Those screenshots were taken with AfterBurner OSD on. You can see the memory usage, too.

 

[cmdrdredd]

The op is running surround monitors so his resolution will be higher thsn yours. Memory usage rises with resolution.

 

[Annisman*]

Well, in a little bit I'll be able to let you both know the answer

However, It's my belief that The Witcher 2 will probably be unplayable in Surround with ubersampling anyways, but it will be interesting to see the memory usage, and *if* it really does hog memory.

 

[RussianSensation]

Well, in a little bit I'll be able to let you both know the answer

However, It's my belief that The Witcher 2 will probably be unplayable in Surround with ubersampling anyways, but it will be interesting to see the memory usage, and *if* it really does hog memory.

That's another good reason to test it. Try it with a single screen out of curiosity to see if 3x GTX670s can do it.

The most challenging part about this testing is that some games will pre-cache textures and may actually use 2GB of VRAM but it doesn't actually mean the game needs >2GB of VRAM to run smoothly. It's not the same as Doom 3 days where I think you actually needed a 256MB (or was it 512MB) of VRAM to enable the highest settings (otherwise that setting was completely blocked out). Max Payne 3 with 4xMSAA/8xMSAA may use > 2GB of VRAM but it'll bring performance to its knees on modern cards because of the MSAA in the first place. GTAIV/SKYRIM with mods may be the ticket.

 

[Annisman*]

That's another good reason to test it. Try it with a single screen out of curiosity to see if 3x GTX670s can do it.

The most challenging part about this testing is that some games will pre-cache textures and may actually use 2GB of VRAM but it doesn't actually mean the game needs >2GB of VRAM to run smoothly. It's not the same as Doom 3 days where I think you actually needed a 256MB (or was it 512MB) of VRAM to enable the highest settings (otherwise that setting was completely blocked out). Max Payne 3 with 4xMSAA/8xMSAA may use > 2GB of VRAM but it'll bring performance to its knees on modern cards because of the MSAA in the first place. GTAIV/SKYRIM with mods may be the ticket.

You're right, there is a distinct difference between VRAM used and VRAM needed, so that will be noted in whatever results I get.

And you've got a good point about Skyrim, a game that is generally speaking, very low gpu-bound, but can be absolutely loaded with Texture mods, it will be an interesting one to test.

 

[PowerK]

The op is running surround monitors so his resolution will be higher thsn yours. Memory usage rises with resolution.

I came from 3 x 1200p displays (U2410) to a single U3011. Again, ubersampling does not hog the memory at all. What makes you think it does ?

 

[Annisman*]

I came from 3 x 1200p displays (U2410) to a single U3011. Again, ubersampling does not hog the memory at all. What makes you think it does ?

Well, if ubersampling has anything to do with textures I could see it as a possibility, but as I have no idea what ubersampling actually does, that's just one guess I would make.

 

[PowerK]

Well, if ubersampling has anything to do with textures I could see it as a possibility, but as I have no idea what ubersampling actually does, that's just one guess I would make.

I wondered about the same thing. It's just a form of supersampling. An interesting thing is I cannot really tell the difference between ubersampling on vs off at 2560x1600 resolution. However, on 1080p/1200p, the difference is very noticeable.

 

[Annisman*]

Benching in progress, will start up a new thread with results in a day or so.