GALAXY GeForce GTX 660 Ti GC 3GB SLI Review @ [H]

FrgMstr

FrgMstr

Just Plain Mean
Staff member
Joined
May 18, 1997
Messages
55,634
GALAXY GeForce GTX 660 Ti GC 3GB SLI Review - We've got something special for you; SLI and Surround gameplay! Two GALAXY GeForce GTX 660 Ti GC 3GB video cards in SLI in NV Surround resolutions for your gaming pleasure. We will find out how this setup performs, compares, and even overclocks. If you want an affordable SLI solution with great gaming performance, you'll like this.
 
"There is a smoothness to SLI we just can't put in words. We know for a fact that NVIDIA uses an algorithm that smoothes "frametime" in SLI. We don't know what it’s called, or even how it works, but we know it exists, and we know NVIDIA employs some special sauce when it comes to SLI. It is something that can only be felt, as you play a game, it is not something that shows up in a framerate over time graph."

Magic?
 
wadec22 said:
Great review. Y no surround results from the 680?
Click to expand...

They said it just couldn't cope with the resolution under several of the tests. It's probable it could handle something like Skyrim but possibly not most of the other games.
 
Excellent review, it'd been a while since you had done some in depth Surround/Eyefinity testing, and of course, no one does this as well as you guys. I've been itching to upgrade my 2x6950, I know now what to get if these ever come down in price a bit more (or I spot a smoking 7950 deal), I'm just not sure that's gonna happen at all before next gen cards come out.
 
That was always with AA enabled wasn't it? It's still interesting to c playable settings on three screens on one card. When I tried dual 680s i still found sli micro stutter to be unplayable for me, even though it was better than dual 7970s. I suppose u can't please everyone, maybe it's just too small an audience to test for.
 
I am curious if they had used a 4gb 680 card if it would have fared better at the higher resolutions.
 
You guys really need to do a review where you look at frametimes on SLI/CF and stop making subjective evaluations about how CF/SLI "feel" when you don't do any objective testing to support your statements.

If that were difficult to do, I'd understand, but it's almost stupidly easy... it requires ticking one extra box in FRAPS. This would be AMAZINGLY helpful to have for users looking into buying multi-GPU configs... currently the only site that does it is techreport and they don't do nearly as many GPU reviews as a site like [H]. It's also always helpful to get data from multiple people/sites so that you don't have to make judgments based on a single set of info (currently techreport's).

The need for some objectivity is highlighted in an article like this, where essentially your award and recommendation is based on a subjective evaluation that doesn't line up with your presented data, as your raw data shows the 7950s offering more performance for less money (and we know that the 7950 offers equal or better OC ability as well).

I'm not disagreeing with anything you're saying - my experience with 7970 CF is that it can get a bit "stutter-y" under the 40-ish fps mark in some instances, though I don't have a pair of 680s sitting here that I can subjectively compare against.
 
Last edited:
BigMack70 said:
You guys really need to do a review where you look at frametimes on SLI/CF and stop making subjective evaluations about how CF/SLI "feel" when you don't do any objective testing to support your statements.

If that were difficult to do, I'd understand, but it's almost stupidly easy... it requires ticking one extra box in FRAPS. This would be AMAZINGLY helpful to have for users looking into buying multi-GPU configs... currently the only site that does it is techreport and they don't do nearly as many GPU reviews as a site like [H]. It's also always helpful to get data from multiple people/sites so that you don't have to make judgments based on a single set of info (currently techreport's).

The need for some objectivity is highlighted in an article like this, where essentially your award and recommendation is based on a subjective evaluation that doesn't line up with your presented data, as your raw data shows the 7950s offering more performance for less money (and we know that the 7950 offers equal or better OC ability as well).

I'm not disagreeing with anything you're saying - my experience with 7970 CF is that it can get a bit "stutter-y" under the 40-ish fps mark in some instances, though I don't have a pair of 680s sitting here that I can subjectively compare against.
Click to expand...

I'm not confident stating that frametime is equal to "feel" I think there are many more factors than just frametime. It is another metric, but I'm not confident in its ability to relate in a real meaningful way to SLI/CFX performance. You are asking for an objective, scientific result on a subjective thing that you just have to feel out as you play, the sensitivity of each person is different.

This is fact, NVIDIA uses an algorithm that smooths framerates, it is a real thing, and it does have a real-world experience difference that can be felt, but does not show up in framerate.
 
nice review guys.

subjective option on the smoothness but each to their own. Maybe sli does have some secrete sauce, i will remain skeptical. For the pepsi challenge, that would be a bad idea remember they did a test intel sandy bridge vs bulldozer on one of those challenges and people picked the amd machine lol.
http://www.kitguru.net/components/cpu/harrison/amd-claims-pepsi-challenge-double-whammy/

guess amd has some secrete sauce to. LOL

nice testing with the new 12.11 drivers.

I would still go with 7950 crossfire over 660 TI Sli. I will take the raw performance and the free game bundle.
 
ebduncan said:
nice review guys.

subjective option on the smoothness but each to their own. Maybe sli does have some secrete sauce, i will remain skeptical. For the pepsi challenge, that would be a bad idea remember they did a test intel sandy bridge vs bulldozer on one of those challenges and people picked the amd machine lol.

nice testing with the new 12.11 drivers.

I would still go with 7950 crossfire over 660 TI Sli. I will take the raw performance and the free game bundle.
Click to expand...

Thanks for the kudos.

You can remain skeptical on the "secret sauce" but there is in fact something being done as confirmed by NVIDIA, just not discussed in depth.

As for the Pepsi Challenge, I have no doubt in my mind that most if not all would easily see the difference. I took CFX out of my personal gaming box for this very reason.
 
Brent_Justice said:
I'm not confident stating that frametime is equal to "feel" I think there are many more factors than just frametime. It is another metric, but I'm not confident in its ability to relate in a real meaningful way to SLI/CFX performance. You are asking for an objective, scientific result on a subjective thing that you just have to feel out as you play, the sensitivity of each person is different.

This is fact, NVIDIA uses an algorithm that smooths framerates, it is a real thing, and it does have a real-world experience difference that can be felt, but does not show up in framerate.
Click to expand...

OK, but frametimes get at "smoothness" far better than average fps does, and take minimal effort to include.

I'm sure you've read this stuff, but it's worth linking anyways:
http://techreport.com/review/21516/inside-the-second-a-new-look-at-game-benchmarking
http://techreport.com/review/22890/nvidia-geforce-gtx-690-graphics-card/3

Benchmarking frametimes may not be the definitive, be-all-and-end-all way to measure smoothness and stutter, but I've not seen any better methodology set forth, and TR's frametime results definitely show something similar to what you guys at [H] have been describing for a while - that SLI tends to be smoother than CF-X and that average fps results can be deceptive.

It doesn't seem the least bit odd to you that you're giving awards/recommendations based to some degree on subjective analyses that appear to contradict your data? At the very least, frametime analysis would allow you to point to some data to show that there's substance to what you're saying.
 
Last edited:
I remember reading stuff about frame times, and that staggered frame times were more noticeable than framerate. If they are indeed doing something to buffer the different frame times generated by the two different cards, then it would make sense that it's smoother than the AMD crossfire. Hopefully AMD jumps on this bandwagon. I, myself, couldn't find a need for dual video cards though. I'd have to invest in multiple monitors to justify the cost of two or more high-end cards. Great review! :)
 
Brent_Justice said:
This is fact, NVIDIA uses an algorithm that smooths framerates, it is a real thing, and it does have a real-world experience difference that can be felt, but does not show up in framerate.
Click to expand...

i'm no mathemagician or anything but, wouldn't any sort of frame rate smoothing involve a decrease in total frames per second?

this just seems unlikely in an industry dominated by these types of performance measurements.

i still think it must be magic.
 
wirerogue said:
i'm no mathemagician or anything but, wouldn't any sort of frame rate smoothing involve a decrease in total frames per second?

this just seems unlikely in an industry dominated by these types of performance measurements.

i still think it must be magic.
Click to expand...

not really... if you have one frame rendered in 17ms and one in 47ms, that's equivalent to 30fps average but won't feel nearly as smooth as if you had both frames rendered in about 33ms even though that's also going to wind up being 30fps
 
BigMack70 said:
not really... if you have one frame rendered in 17ms and one in 47ms, that's equivalent to 30fps average but won't feel nearly as smooth as if you had both frames rendered in about 33ms even though that's also going to wind up being 30fps
Click to expand...

in theory, that is correct but the only way that would be possible is if the chip was underpowered/underclocked to begin with. how else would it be possible to make up the needed time to render the second frame? i can only see dropping frames as as viable solution. if the hardware could have rendered that second frame faster, it already would have.
 
I will say this, data and framerates do not always represent the gaming experience, this is why we do what we do the way we do it. :)
 
wirerogue said:
in theory, that is correct but the only way that would be possible is if the chip was underpowered/underclocked to begin with. how else would it be possible to make up the needed time to render the second frame? i can only see dropping frames as as viable solution. if the hardware could have rendered that second frame faster, it already would have.
Click to expand...

We're talking about multi-GPU, not single GPU, which is why this phenomena can happen. The fact that the system has to talk to both cards and both cards have to talk to each other introduces potential for latency and unevenness not necessarily related to the power of the GPU itself.
 
Error:

On Page 2 - Test Setup, the System Configuration image says i7-2600K even though you guys tested with a 3770K.
 
Thank you for writing this article. Dare I ask you follow it up with comparing 3x 660 vs 2x 680?
 
BigMack70 said:
OK, but frametimes get at "smoothness" far better than average fps does, and take minimal effort to include.

I'm sure you've read this stuff, but it's worth linking anyways:
http://techreport.com/review/21516/inside-the-second-a-new-look-at-game-benchmarking
http://techreport.com/review/22890/nvidia-geforce-gtx-690-graphics-card/3

Benchmarking frametimes may not be the definitive, be-all-and-end-all way to measure smoothness and stutter, but I've not seen any better methodology set forth, and TR's frametime results definitely show something similar to what you guys at [H] have been describing for a while - that SLI tends to be smoother than CF-X and that average fps results can be deceptive.

It doesn't seem the least bit odd to you that you're giving awards/recommendations based to some degree on subjective analyses that appear to contradict your data? At the very least, frametime analysis would allow you to point to some data to show that there's substance to what you're saying.
Click to expand...

I went and checked out one of those reviews after seeing them referenced multiple times in these forums. In my opinion it's a pile of data that's meant to look impressive but ultimately has no real-world meaning other than what the reviewer assigns to it. There are too many ways to slice the data and no real-world standard by which to determine the "correct" way. What percentage of frametimes being under what threshold constitutes a good user experience? Who knows? And those kinds of judgments fall into the same trap as framerates--global averages obscure spikes that impact gameplay.

That leaves aside the underlying issue that was addressed in this very article, and not for the first time--the TechReport/FRAPS method measures frametimes in the rendering stage, not at the display, therefore it's not a real-world measurement of what the user experiences.
 
wirerogue said:
i'm no mathemagician or anything but, wouldn't any sort of frame rate smoothing involve a decrease in total frames per second?

this just seems unlikely in an industry dominated by these types of performance measurements.

i still think it must be magic.
Click to expand...

Magic for noob like you, real thing for real users...

Nice review..
 
Thanks for this thorough and revealing review. A week or so ago, the forum Galaxy rep mentioned that he/she was experimenting with 660 (non-Ti) SLI and was seeing close to 100% scaling across the board. It would be awesome to see [H] do a similar article for the 660. Especially given that at today's prices, a pair of vanilla 660's would probably cost LESS than a 680!
 
Nice review. Subjective results are a necessary part of any review. Same way the car magazines do it.
 
wirerogue said:
oh you got me good on that one. i've certainly learned my place now.
Click to expand...

Well considering you are indicating that Kyle/Brent don't know their shit...you really need to learn your place unless you can provide some references that somehow put in you in the same league as them?
 
Well considering you are indicating that Kyle/Brent don't know their shit...you really need to learn your place unless you can provide some references that somehow put in you in the same league as them?
Click to expand...

Kyle and Brent are quite knowledgeable when it comes to enthusiasts platforms. I wouldn't call them hardware engineers knowledgeable about products, but they represent enthusiast gamers and overclockers rather well, and what others would likely expect or use of product. Kyle and company at Hardocp do a great job of explaining feature sets and components to average users.

When it comes to smoothness between Crossfire and Sli. Its a mixed bag. Nvidia has some secret sauce which applies to frame time in conjugation with muti gpu setups. Plenty of other things effect the user experience such as the type of monitor used, the game being played, and the card performance in that game. I am sure Amd has some sort of sauce which they use to help smooth the performance with muti gpu setups, and if they don't they likely will soon. It all starts on how sli and crossfire work, one gpu renders a frame and the next frame is sent to the other gpu. The biggest problem is that the 2nd or 3rd or 4th card has to send that frame back to the first card then to your monitor which causes a delay. Thus your stuck with frame times that alternate like A/C current does. You experience Jitter or mirco stutter when the gap between frame times alternates with a large gap. Fast response time monitors are more likely to present the problem to you, while slower response time monitors will mask the problem.

Its complex and its crazy, but more importantly Both sides are working to resolve the issue completely. Honestly i don't know why they make each video card output each video, then combine the signals at the monitor, or at the back of computer. Would be feasible using display port, or hdmi 1.4a or any other connection that supports 3d stereoscopic.

Honestly I use rather slow response time monitors and i'm not bugged by a slight stutter in my games.
 
Trepidati0n said:
Well considering you are indicating that Kyle/Brent don't know their shit...you really need to learn your place unless you can provide some references that somehow put in you in the same league as them?
Click to expand...

first of all i made no indication of kyle/brent "not knowing their shit"

learn my place? did i just commit some sort of blasphemy? i apologize if i don't share your worship of kyle/brent.

is the comments section limited to only positive comments? maybe you should volunteer to moderate the comments sections and remove anything but positive ones.
 
Back to topic, recent price drops would suggest that NV is beginning to recognize and accept what AMD's aggressive pricing has done to their value proposition. They are still small and long overdue, but at least they are happening. NewEgg currently has an MSI 670 for $319 after MIR. Throw a couple of those in SLI against the 7950's in this review and I think you'd have a significant shift in the results.
 
makes me wonder if the memory limitations are more than the GPU these days - takes some of the shine off the 690 I just got.
 
You must log in or register to reply here.
Back
Top