GTX 670 4GB 4-Way SLI

fomoz

Limp Gawd
Joined
Jan 5, 2012
Messages
394
My scaling at 8064x1440 isn't better than it was with 2-way SLI. What's the deal?
 
pics or it didn't happen. (of the cards I mean)
GTX670_4-way.jpg
 
Looks like a driver problem. I got it to scale well when I changed "Span displays with Surround" to "Maximize 3D performance". At 2560x1440 Heaven went from 58.1 fps in 2-way SLI to 105.8 fps in 4-way SLI. Fuck yeah.

OK nVidia, now how about you fix this so that I actually get to use this on my three screens? Please?
 
Well for ONE, what game you playin with 4 670GTsex's? For all we know you could be playing hmmm checkers.
 
I've got mine spanned with NV Surround and it scales just fine. Just use GPU-Z to verify SLI is enabled (4 GPUs).


(click to enlarge)


(click to enlarge)

EDIT: btw - I recommend you remove 3 of your 4 cards backplates for better temps. I saw temp improvements across the board (was 77/78/64/57C; now 73/72/61/57C - after ~15 min. of Unigine Heaven). Obviously, the top card (closest to the CPU) won't benefit so I left that one on. Jacob F. from EVGA said the RAM is rated to 150C so it should be OK without the thermal pads on it (there are thermal pads mounted to the back of the backplates - see pic below).


(click to enlarge)
 
Last edited:
I've got mine spanned with NV Surround and it scales just fine. Just use GPU-Z to verify SLI is enabled (4 GPUs).

EDIT: btw - I recommend you remove 3 of your 4 cards backplates for better temps. I saw temp improvements across the board (was 77/78/64/57C; now 73/72/61/57C - after ~15 min. of Unigine Heaven). Obviously, the top card (closest to the CPU) won't benefit so I left that one on. Jacob F. from EVGA said the RAM is rated to 150C so it should be OK without the thermal pads on it (there are thermal pads mounted to the back of the backplates - see pic below).
Mine scales fine on one screen, how does yours scale in surround? Btw, are you seriously running 4-way SLI for 1680x1050?

My temps are fine, I'm only hitting 71C on the hottest card in Heaven. My RAM is OC'd to 3554, probably can OC it even more, so I'd rather keep them on. Did you check your RAM temps without the backplates? What are they?
 
Last edited:
Mine scales fine on one screen, how does yours scale in surround? Btw, are you seriously running 4-way SLI for 1680x1050?

Haha. No. Those settings are for benching on Overclock.net. How would I be running NV Surround, then? It's 5760x1080 on 3x 120hz BenQ XL2420T*'s. :)

Not sure what you mean by scale...I get about 60-70 FPS in BF3 MP on large maps with everything maxed out (motion blur: off) at 5760x1080. I simply can't run under "Maximize 3D performance" if I want to run multi-monitor. Choose "Span displays with Surround" and configure it. That will utilize SLI to the fullest with multi-monitor. Make sure PhysX is set to "Auto-select."

My temps are fine, I'm only hitting 71C on the hottest card in Heaven. My RAM is OC'd to 3554, probably can OC it even more, so I'd rather keep them on. Did you check your RAM temps without the backplates? What are they?

Yeah, anything over 70 C is bad news bears - because you're hitting the 70 C thermal wall where the GPU will start downclocking. So I'd remove those backplates if I were you. I am probably going to lower my factory OC to keep it down...or install a hacked BIOS so I can use 100% fan when gaming.

As far as I know there is no way to check RAM temps without an external apparatus. But 150C is ridiculously smokin' so he's essentially saying there's really no way you can F it up. :)

BTW - I'm really curious to see more about your rigs performance. I almost went to Z77 due to X79 not being fully supported by NVIDIA in PCIe 3.0. Luckily I was able to get their "hack" to work and I have PCIe 3.0 running flawlessly. I would have switched anyway but my understanding is the PLX bridge that is utilized in Z77 quad-SLI capable boards adds in latency (since it is not direct PCIe lanes from the CPU like in X79).
 
Last edited:
Haha. No. Those settings are for benching on Overclock.net. How would I be running NV Surround, then? It's 5760x1080 on 3x 120hz BenQ XL2420T*'s. :)

Not sure what you mean by scale...I get about 60-70 FPS in BF3 MP on large maps with everything maxed out (motion blur: off) at 5760x1080. I simply can't run under "Maximize 3D performance" if I want to run multi-monitor. Choose "Span displays with Surround" and configure it. That will utilize SLI to the fullest with multi-monitor. Make sure PhysX is set to "Auto-select."
Dude, have you read this read at all? I guess not, but that's OK. Here's what you missed:

Looks like a driver problem. I got it to scale well when I changed "Span displays with Surround" to "Maximize 3D performance". At 2560x1440 Heaven went from 58.1 fps in 2-way SLI to 105.8 fps in 4-way SLI. Fuck yeah.

OK nVidia, now how about you fix this so that I actually get to use this on my three screens? Please?

"Scales" means that I get proper GPU usage and about twice higher FPS in 4-way SLI than in 2-way. Just please post a benchmark in maxed out heaven at 5760x1080 and check your GPU usage on the PrecisionX graphs. You can also run a benchmark in 1080p with "Span displays with Surround" and "Maximize 3D performance" for comparison. Both should give you exactly the same FPS, otherwise you have the same problem as me. Thanks.

Yeah, anything over 70 C is bad news bears - because you're hitting the 70 C thermal wall where the GPU will start downclocking. So I'd remove those backplates if I were you. I am probably going to lower my factory OC to keep it down...or install a hacked BIOS so I can use 100% fan when gaming.

As far as I know there is no way to check RAM temps without an external apparatus. But 150C is ridiculously smokin' so he's essentially saying there's really no way you can F it up. :)

BTW - I'm really curious to see more about your rigs performance. I almost went to Z77 due to X79 not being fully supported by NVIDIA in PCIe 3.0. Luckily I was able to get their "hack" to work and I have PCIe 3.0 running flawlessly. I would have switched anyway but my understanding is the PLX bridge that is utilized in Z77 quad-SLI capable boards adds in latency (since it is not direct PCIe lanes from the CPU like in X79).

I agree, but the thing is that I get a lot more FPS from OC'ing my VRAM so high than I lose due to the 13 MHz core drop from throttling at 70C.

You just saw my benches. You can run the Extreme preset in 3DMark11 and compare your result with mine.
 
Last edited:
By the way, we have 4-way SLI. Quad-SLI is the term used for dual-GPU cards :)
 
Dude, have you read this read at all? I guess not, but that's OK. Here's what you missed:

I read it. Didn't miss it. When I span displays with surround I get 4-way-SLI-like performance...that's all I was stating. Just trying to help.


By the way, we have 4-way SLI. Quad-SLI is the term used for dual-GPU cards :)

Man, you're quite the charmer! Haha. Quad-SLI is the correct nomenclature, as well. I've never heard it referenced differently...I mean 4-way or quad...same difference.

EDIT: just ran Heaven at 5760x1080 and you're right - it doesn't scale. I'd guess it's a Heaven problem, though. Because my performance in everything else is quad-SLI (or 4-way-SLI :)) level. I just started using Heaven to bench a few days ago...never ran into it with my previous 680 2GBs. My numbers:


FPS:
31.3
Scores:
789
Min FPS:
7.2
Max FPS:
131.5
 
Last edited:
I read it. Didn't miss it. When I span displays with surround I get 4-way-SLI-like performance...that's all I was stating. Just trying to help.




Man, you're quite the charmer! Haha. Quad-SLI is the correct nomenclature, as well. I've never heard it referenced differently...I mean 4-way or quad...same difference.

EDIT: just ran Heaven at 5760x1080 and you're right - it doesn't scale. I'd guess it's a Heaven problem, though. Because my performance in everything else is quad-SLI (or 4-way-SLI :)) level. I just started using Heaven to bench a few days ago...never ran into it with my previous 680 2GBs. My numbers:


FPS:
31.3
Scores:
789
Min FPS:
7.2
Max FPS:
131.5
Hey, I would like to apologize for my previous posts (before the edits). Mornings are not my best time. Regarding 4-way vs Quad SLI, I thought the same thing until Dan_D pointed it out. I agree that there shouldn't be a difference, but he's saying otherwise :)

Thank you for the benchmarks! I'm afraid that it's not only a problem in Heaven, since I got exactly the same results in Dragon Age II and Heroes VI. No difference in performance between 2-way SLI and 4-way SLI! I submitted a ticket to nVidia about this last night, we'll see what they say.
 
Hey, I would like to apologize for my previous posts (before the edits). Mornings are not my best time. Regarding 4-way vs Quad SLI, I thought the same thing until Dan_D pointed it out. I agree that there shouldn't be a difference, but he's saying otherwise :)

Thank you for the benchmarks! I'm afraid that it's not only a problem in Heaven, since I got exactly the same results in Dragon Age II and Heroes VI. No difference in performance between 2-way SLI and 4-way SLI! I submitted a ticket to nVidia about this last night, we'll see what they say.

Cool. I stand corrected. I've got some research to do, myself. I ran 3DMark 11 Extreme Preset and only scored a X8129. WOW. Either I need to go back to my Z77 plan or something is up with my rig! :)

EDIT: and of course: apology accepted (but not necessary). I know what you mean about the AM. :)
 
Just wanted to say that pictures of both of your systems are next to definition of [H]ard.
 
Just wanted to say that pictures of both of your systems are next to definition of [H]ard.

Ha...thanks man. fomoz is doing this shiz right. Z77 + 670 4GB 4-way SLI is the best max setup today. I started my 4-way SLI setup back when the 680s were launched so no Z77 or 670. Now I kinda wish I wasn't swayed back to X79 when I actually had a Z77 board and 3770K proc in my possession!

Back to the topic:

My issues may be popping up due to the lack of backplate and the thermal pads that are on them. Getting some graphical corruption that is consistent with memory issues. Had to downclock my core by 50 MHz in order to complete another 3DMark 11 Extreme Preset run! But the numbers are back to "normal" at least:

X12059
http://3dmark.com/3dm11/3831189

EDIT: this is also with "Maximize 3D Performance" on and NV Surround disabled (so one 1080p monitor) - so I think you are on to something fomoz!!! Nice work. Where do I submit a ticket, too? I'll ping Jacob F. from EVGA as well maybe he can help. Just sent him a tweet with a link to this thread.
 
Ha...thanks man. fomoz is doing this shiz right. Z77 + 670 4GB 4-way SLI is the best max setup today. I started my 4-way SLI setup back when the 680s were launched so no Z77 or 670. Now I kinda wish I wasn't swayed back to X79 when I actually had a Z77 board and 3770K proc in my possession!

Back to the topic:

My issues may be popping up due to the lack of backplate and the thermal pads that are on them. Getting some graphical corruption that is consistent with memory issues. Had to downclock my core by 50 MHz in order to complete another 3DMark 11 Extreme Preset run! But the numbers are back to "normal" at least:

X12059
http://3dmark.com/3dm11/3831189

EDIT: this is also with "Maximize 3D Performance" on and NV Surround disabled (so one 1080p monitor) - so I think you are on to something fomoz!!! Nice work. Where do I submit a ticket, too? I'll ping Jacob F. from EVGA as well maybe he can help. Just sent him a tweet with a link to this thread.
No problem, the goal is to get this fixed ASAP so that we can finally play some games in Surround on our setups :) You can submit your ticket to nVidia here.

Thanks, I'm actually planning on replacing the GTX 670's with the new (hopefully "Big Kepler") cards that are supposed to come out in about 6 months, so the $600 I saved compared to getting GTX 680 4GB should come in handy there. Also, 3770K/Z77 is another $350 less compared to 3930K/X79, since I just don't need the extra threads for gaming and I decided to gamble on the PLX chip being as good as X79's 40 PCI-E lanes. We'll see, I hope that I'm not going to have an issue like micro stuttering or something. I'm planning on upgrading the entire setup at every Intel Tock, so I'm trying to save where I can in order avoid remorse when I sell it at a discount later on :)

In the beginning I thought about removing the backplates, too, but I decided against it since they're cooling down the VRAM. The VRAM OC's well, but be careful, I noticed that past a certain point I start getting a lot of micro stuttering without seeing any artifacts. That's at 3706 MHz when I had 2-way SLI with the cards spread out with a PCI-E slot in between. It ran with no micro stuttering up to 3606 MHz. The micro stuttering can be easily seen visually, but I also confirmed it by reading a FRAPS frame draw time log after a run in Dirt 3. I'm running it at a conservative 3506 MHz right now, since the hot VRAM is making my overall performance slower due to Kepler throttling.

Just to make this clear for those who haven't read the whole thread. Here are the results from two 3DMark11 runs, everything else being equal:

"Span displays with Surround"

x122_65_550_1175mv_4.5GHz_4-way_7895.png


"Maximize 3D performance"

x122_65_550_1175mv_4.5GHz_4-way_11352.png
 
Yes, your results above are very consistent with mine. I just got done putting my backplates on (God damn that's a lot of screws!!!)...I'll deal with the ~5C higher temps in other ways. I'm starting to worry these cards can't perform in 4-way SLI with their factory OC applied. :(

Jacob F. got back to me regarding the memory corruption stuff but didn't say anything regarding the NV Surround scaling issue. I know it's NVIDIA's issue but I was hoping he would help us push this forward to them since surely he is better connected than a couple dudes with a lot of coin invested in graphics cards. :)

Not much more time to tweak today (my son is about to get up from nap) but I will hit this some more tonight. I'm currently running at 131%/-50/0 (power/core/mem) in EVGA Precision X until I figure out this heat problem. May need to buy better and/or more case fans. It's nuts because with my reference 680s I had GREAT temps - http://1pcent.com/?p=43. So far my "upgrade" to 4GB has been troublesome. Needless to say it was a lot of coin and I'm not entirely pleased, yet. But that's what you get for early adopting/etc. heh.

EDIT: a couple interesting related articles:
http://www./whats-new/articles/nvidia-surround-on-the-geforce-gtx-680/#1

http://hardocp.com/article/2012/04/25/geforce_gtx_680_3way_sli_radeon_7970_trifire_review/4

I'm seeing similar performance to what [H] reported in the above article. So I wonder if it's an issue that depends on the application? Have you tried BF3 MP? I watch FRAPS almost more than I look out for the enemy, lol...so that'll give us an idea, too...
 
Last edited:
To the OP, awesome setup.

Just a question though, do those two center cards have any intake airflow whatsoever?
 
To the OP, awesome setup.

Just a question though, do those two center cards have any intake airflow whatsoever?
Thanks! Yeah, the top three cards have airflow. It looked scary when I put it together, btu it's actually not that bad. There's a side fan blowing on them, too. They're running at around 72C max in Heaven at 1220/3504 with the fans at 80%.
 
Quad 4gb GTX 670 would be one of the most badass setups money could buy.
 
This may be an isolated issue to certain tests. I ran the standard "P" 3DMark 11 test and my results were the same (essentially) between "Maximize 3D performance" and the "Span displays with Surround" settings.

fomoz - can you see if you get similar results? Obviously, we're not going to run at 720p with this setup...but still. :)
 
This may be an isolated issue to certain tests. I ran the standard "P" 3DMark 11 test and my results were the same (essentially) between "Maximize 3D performance" and the "Span displays with Surround" settings.

fomoz - can you see if you get similar results? Obviously, we're not going to run at 720p with this setup...but still. :)

It's possible, since 3DMark11's Performance preset is pretty much a CPU test with this hardware.
 
It's possible, since 3DMark11's Performance preset is pretty much a CPU test with this hardware.

You'd be surprised. I.E. I score nearly 20K P with my MSI Lightining 680 SLI and with my quad-SLI I score around 25K. My CPU is clocked higher on the quad-SLI rig (3960X @ 4.9 GHz versus 3930k @ 4.7 GHz).
 
sk3tch, thank you for your efforts in investigating this. I think that at this point there's not much left to do other than to wait and see what nVidia says. I had a slim hope that their support teams and labs are working on weekends, but I guess that's not the case.

Overall, although it's cool, it's a pain to be an early adopter. I was very excited when I received my 4th GPU and finally completed my setup on Friday and I was expecting to do some gaming this weekend. Oh well, I hope that they fix this soon.
 
Cool. Try the P test at some point. You'll likely have similar results. In other news my issues from earlier are likely a bad card (VRAM issue - problems didn't go away when I re-installed the backplates). Now I need to futz with the lot of them to see which one is bad. :-/

I really think SLI gaming is the sweet spot (2 cards) at 1080p/120hz - especially if you can get two MSI Lightnings. They're nothing but butter. I'd quit the 4-way game but I'm already pretty invested in it. By the time everything will be cleared up by NVIDIA the next gen will be out and I'll be upgrading to that anyway. Ha.
 
Last edited:
Cool. Try the P test at some point. You'll likely have similar results. In other news my issues from earlier are likely a bad card (VRAM issue - problems didn't go away when I re-installed the backplates). Now I need to futz with the lot of them to see which one is bad. :-/

I really think SLI gaming is the sweet spot (2 cards) at 1080p/120hz - especially if you can get two MSI Lightnings. They're nothing but butter. I'd quit the 4-way game but I'm already pretty invested in it. By the time everything will be cleared up by NVIDIA the next gen will be out and I'll be upgrading to that anyway. Ha.
I will. Have you tried Heaven at 720p, too? Curious to see if you get the same results.

I liked my 2-way SLI setup a lot, too, especially when the cards were spread out! If I wasn't gaming on three screens, then I'd be completely satisfied with it, it worked very, very well for 1440p.
 
Nah, didn't try Unigine Heaven at 720p. Isolated the bad card (very thankful for the Rampage IV Extreme's onboard dip switches to disable PCIe slots) - so I'm at tri-SLI. Interestingly enough, BF3 performs about the same. Heh. What a joke.
 
Just did a run in the Performance preset.

"Span displays with Surround"

p122_65_500_1175mv_4.5GHz_4-way_19721.png


"Maximize 3D performance"

p122_65_500_1175mv_4.5GHz_4-way_20789.png


The graphics score is still a bit lower in Surround mode.

Can you please post the screenshots of your 3DMark11 P results? It's crazy that your 3960X @ 4.9 GHz scored 5,000 points more than my 3770K @ 4.5 GHz, what a beast!
 
Last edited:
No need...I just link to Futuremark's site. Easier to compare/contrast that way versus screenshots.

3DMark uses CPU quite a bit...but also graphics quite a bit. I like it better than Unigine Heaven for balance. No matter what - if you're chasing benches (not something I recommend) versus actual gaming you'll want the best processor. Obviously, CPU is not that important for games as long as you have a nice quad-core! :)

P25257 - 3960x @ 4.9 GHz / 4-way SLI EVGA GTX 680 FTW+ 4GB (NV Surround: ON)
http://3dmark.com/3dm11/3779176

P19163 - 3930k @ 4.7 GHz / 2-way SLI MSI Lightning GTX 680 2GB (NV Surround: OFF)
http://3dmark.com/3dm11/3799769

Like I said...sweet spot...two Lightnings. :)
 
Last edited:
Thanks. I know that after a certain point CPU doesn't really matter in games, but I still find those benches very impressive. Our graphics scores are similar, but there's a huge difference between our Physics and Combined scores. The 3960X is definitely a monster compared to the 3770K.

By the way, do you have a result for 4-way SLI with surround off?
 
I think I had something else running during the P run with Surround off. I did it again and I got 20,789 instead of 20,159.
 
I think I had something else running during the P run with Surround off. I did it again and I got 20,789 instead of 20,159.

You're hooked. :)

Do you play BF3? If you ever play BF3 MP, I'd love to know what ballpark FPS you get (i.e. whatever you eyeball from FRAPS or Precision X OSD).
 
Sick system!!
Post some minesweeper benchmarks when you get a chance.
 
Back
Top