GTX690 vs. GTX680 sli - User Experiences

G

GMaxx

Limp Gawd
Joined
Jan 13, 2005
Messages
311
Hi guys, I'm building a new rig. I have two gtx680 cards in front of me but just got a great deal on a gtx690 from a client.

So, I wanted to find out from users who either went from gtx680 sli to gtx690 or from gtx690 to gtx680 sli their experiences and views on which was better.

I already do know about the tradeoffs from heat, power req., noise and marginal performance gain on paper for these cards. However, it's always better to see what others who had these setups think.

Of course, I'll be making my own observations once I get my rig up and running but I wanted to see what others thought first.

Oh yeah, I'll be gaming on a 30 inch monitor and probably won't be going to multi monitors. My new rig will consist of the Asus Rampage IV Extreme and an 3930k.

My goal is to have the best experience in the foreseeable future for gaming.

TIA!

GMaxx
 
GMaxx said:
My goal is to have the best experience in the foreseeable future for gaming.
Click to expand...

With 680 SLI, you're already at the end of your upgrade path unless you have a $$$ motherboard that can support 3 or 4 individual cards in crossfire. With a 690 you can just pop in another one down the road.
 
GotNoRice said:
With 680 SLI, you're already at the end of your upgrade path unless you have a $$$ motherboard that can support 3 or 4 individual cards in crossfire. With a 690 you can just pop in another one down the road.
Click to expand...

So true. I guess because of my ADD compulsion to upgrade things then my definition of "foreseeable future" is 1.5 years at most. My current setup of GTX580 sli is the longest I've keep cards. I'm just too addicted to this hobby. Thanks!
 
I think those things that you listed are about the only differences. The 690 is SLI in a card, so you will not be invulnerable to any micro stuttering that might be present. Other than that, the upgrade path is the only thing.

IMHO - I would go with the 690 just because of the lower heat/noise produced - but then again I actually care about that stuff while others might not.

/jealous.
 
Another vote for 690, for the same reasons. Should be less heat/noise, and you can always easily add another card in if you need it.
 
chimaxi83 said:
Another vote for 690, for the same reasons. Should be less heat/noise, and you can always easily add another card in if you need it.
Click to expand...

It looks like the gtx690 will probably win out. Going watercooling on the chip and just picked up a Seasonic Platinum 1000w psu so silence my be golden in this case.

Thanks for your comments.
 
Why would you watercool a GTX690? It has essentially the best OEM cooler on the market.
 
Doesn't 4-way have performance issues? I think 3-way would be a better idea, but I can dig having only two lovely looking 690s in my computer and even separating them vs stuffing three cards on top of each other.

I personally have 2 680s and I like the look of having two cards with a slot in between connected with a 3-way (solid, not ribbon) bridge.
 
The 690 is no more or less vulnerable to microstuttering than 680 SLi. nVidia supposedly implemented frame metering with the 6xx series which is supposed to help with microstuttering though. Microstuttering is a biproduct of Alternate Frame Rendering, having the chips on a single card will make no difference in that regard. I haven't really noticed microstuttering enough to bother me since the 7950 GTX days though.

Phishy714 said:
I think those things that you listed are about the only differences. The 690 is SLI in a card, so you will not be invulnerable to any micro stuttering that might be present. Other than that, the upgrade path is the only thing.

IMHO - I would go with the 690 just because of the lower heat/noise produced - but then again I actually care about that stuff while others might not.

/jealous.
Click to expand...
 
Particleman said:
The 690 is no more or less vulnerable to microstuttering than 680 SLi. nVidia supposedly implemented frame metering with the 6xx series which is supposed to help with microstuttering though. Microstuttering is a biproduct of Alternate Frame Rendering, having the chips on a single card will make no difference in that regard. I haven't really noticed microstuttering enough to bother me since the 7950 GTX days though.
Click to expand...

Frame times on the GTX690's have been measured to be lower than GTX680 SLi over at TR, which supports the argument that GTX690 Quad SLi is probably the best way to get four GK104's in your system. This is probably due to efficiencies related to having the trace lengths and controller latencies fixed and known on the GTX690's PCB, thus allowing for tighter integration between the GPUs.

Also, micro-stuttering is indeed a primary product of AFR, but micro-stuttering can occur in any system that produces inconsistent frame times, and can be caused by many different things, like the CPU, drivers, game optimizations, etc.
 
If it is an evga or galaxy I will straight up trade you an Intel 320 series 600gb ssd brand new in the box still sealed for the 690.
 
JayteeBates said:
If it is an evga or galaxy I will straight up trade you an Intel 320 series 600gb ssd brand new in the box still sealed for the 690.
Click to expand...

Wow. Go away for a little bit and see what happens. lol.

Actually, I wasn't too concerned with micro stuttering. I had 580 sli and if i could live with them then it shouldn't be a problem here.

I'll probably end up keeping the gtx690. Appreciate the offer, but SSDs is kind of my thing as I've already got Crucial M4 512gb, Vertex 4 512gb, Intel 510 480gb, and Crucial C300 512gb SSDs. :D
 
If you look at the Tech Report GTX article when they measure frame latencies by percentile (which is the measure for microstuttering); 680 sli and the gtx 690, the curve is pretty much identical.

http://techreport.com/articles.x/22890/7

And if you read this part of the article: http://techreport.com/articles.x/22890/3

You see that 590 and 6990 (both dual GPU single PCB cards) exhibit the inconsistent frame times. So being multi-gpu single card does not make you any less immune to microstuttering than a dual card solution. Seeing that the 690 is just 2 680s with the PLX PEX8487 bridge chip on it (you can find motherboards with this chip built on), the difference between 680 sli and 690 is negligible.

IdiotInCharge said:
Frame times on the GTX690's have been measured to be lower than GTX680 SLi over at TR, which supports the argument that GTX690 Quad SLi is probably the best way to get four GK104's in your system. This is probably due to efficiencies related to having the trace lengths and controller latencies fixed and known on the GTX690's PCB, thus allowing for tighter integration between the GPUs.

Also, micro-stuttering is indeed a primary product of AFR, but micro-stuttering can occur in any system that produces inconsistent frame times, and can be caused by many different things, like the CPU, drivers, game optimizations, etc.
Click to expand...
 
GotNoRice said:
With 680 SLI, you're already at the end of your upgrade path unless you have a $$$ motherboard that can support 3 or 4 individual cards in crossfire. With a 690 you can just pop in another one down the road.
Click to expand...

But 690 is so expensive! I shall shall get another 680 down the road. Can't afford 2x690. :eek:
 
Particleman said:
If you look at the Tech Report GTX article when they measure frame latencies by percentile (which is the measure for microstuttering); 680 sli and the gtx 690, the curve is pretty much identical.

http://techreport.com/articles.x/22890/7

And if you read this part of the article: http://techreport.com/articles.x/22890/3

You see that 590 and 6990 (both dual GPU single PCB cards) exhibit the inconsistent frame times. So being multi-gpu single card does not make you any less immune to microstuttering than a dual card solution. Seeing that the 690 is just 2 680s with the PLX PEX8487 bridge chip on it (you can find motherboards with this chip built on), the difference between 680 sli and 690 is negligible.
Click to expand...

Yup, but where there is a difference, it's in favor of the GTX690. It's small, but I highlight it because it will likely be magnified when using a pair of them at higher resolutions (probably 3x1080p, so ~6MP) and high detail settings. The difference between using a pair of GTX690s and four GTX680's isn't huge monetarily, but being able to pipe full PCIe x16 3.0 to each one on an X79 board might just be the best configuration available. This is especially true considering the build quality of the GTX690s and the potential micro-stutter-zapping benefits Nvidia gains with known trace-lengths and latencies when everything is spaced on a card according to their design.
 
Also, those GTX590 and HD6990 results should really be telling, shouldn't they?

Nvidia plainly admits that they've put effort into reducing micro-stutter with the Kepler series, and the benefits are profound enough for our editors here at the [H] to take time to comment on, as well as how bad GCN has handled Crossfire with all but the latest beta drivers, for which we're still waiting for real tests.

I've personally experienced using a pair of HD6950s in Crossfire, and while it provided the expected performance boost once the profiles were sorted and games were patched, the micro-stutter really dampened the experience. Those cards are technically faster than, or are at least as fast as, my GTX670 SC, and yet the single Kepler absolutely feels faster. I'd expect the same result out of a single GCN card.
 
I find the frame latencies by percentile method not as good as the raw data. Only there you can clearly see the "jumping around" of the frametimes. The other method destroys valuable information.
 
boxleitnerb said:
I find the frame latencies by percentile method not as good as the raw data. Only there you can clearly see the "jumping around" of the frametimes. The other method destroys valuable information.
Click to expand...

Yes, you should only use percentages (or averages, etc.) when you're summarizing. Frame-times are measured in time, so that's what should be used to compare them.

This relates to the discussion in that it's been shown that the maximum frame-times for the GTX690 have been measured to be lower than they are for GTX680 SLI, which means that if you're going to get two or four Keplers anyway, you had might as well just grab a pair of GTX690s, unless you're shooting for 7MP+ resolutions and/or need more than 2GB VRAM.
 
You must log in or register to reply here.
Back
Top