MSI GTX 660 Ti Power Edition OC Video Card Review @ [H]

MSI GTX 660 Ti Power Edition OC Video Card Review - MSI is offering a custom cooled and factory overclocked rendition on the recently released NVIDIA GTX 660 Ti. We examine how well MSI's Twin Frozr IV cooling technology allows us to overclock this new generation video card. We will compare it to a Radeon HD 7950, an overclocked Radeon HD 7870, and a reference GeForce GTX 670.

Great article!!!! The 660 Ti is a solid card even at higher resolutions. I've been running it on my 30" Dell now for a few weeks and have been very impressed since coming from the GTX 480. The noise, heat, and power efficiency is better, and the performance is overall smoother with about 15% more frames before the overclock.

I've just flashed the bios on my $319 Gigabyte Windforce 7950 using the Windforce 7970 bios, and it's running rock solid stable at 1000MHz core, 1375MHz memory without any software overclocking software (I did this because I don't want to use overclocking software if I can flash the bios, and since there are no known editors that work with the Southern Islands' bios, you must use an existing bios, like the Gigabyte 7970 bios). The Gigabyte 7950 also comes with a dual bios switch and excellent cooler/quiet fans.

I know this is a conservative overclock for this card, but I just wanted to put things into perspective and provide some user experience for anyone considering the 7950 vs. the GTX660Ti.

Can someone tell me what happened in the Batman Arkham City benchmark; it just looks fishy to me.
I see in the 'highest-playable-settings' section that with physX enabled the GTX 670 takes a sizable hit in performance, yet the 660Ti does not? Then in the apples-to-apples section David turns off PhysX to compare all cards equally, yet the 660Ti does not increase in performance all that much. However, the 670 does. I see

*660Ti*
PhysX On: 29 , 77 , 46.8 (100 , 100 , 100)%
PhysX Off: 31 , 85 , 48.0 (106 , 110 , 102)%

*660Ti OC*
PhysX On: 32 , 86 , 51.3 (100 , 100 , 100)%
PhysX Off: 26 , 85 , 52.1 (81.2 , 98.8, 101)%

*670*
PhysX On: 27 , 83, 48.6 (100 , 100 , 100)%
PhysX Off: 30 , 105, 61.2 (111 , 126 , 125)%

These are some funky numbers just by disabling the PhysX feature. The overclocked 660 actually performed worse than the stock numbers, and the 670 turned in a huge leap in performance. I only bring this up as I just purchased this card two days before this article went live and I am excited that [H] is covering something I purchased before researching. In this game's page conclusion it is only mentioned that the cards destroy the Radeon's yet the numbers posted contradict the opinion at the bottom of the page. So I ask again, what happened in that benchmark?

Great read. I opted for this one for the cooling solution, Military Class III and the Borderlands 2 deal available here in Denmark ($450). Will be nice to crank up the settings in Max Payne 3 MP, my HD5870 is struggling with that and stuttering tremendously. Hopefully Rockstar is going to use the same engine tech for GTA V.

"We also observed some coil whine coming from the card..."

I didn't need to read that. No way I'm knowingly walking into that mess. Coil whine drives me insane.

You think they would have eliminated this already.

Thanks for the in depth review. I've been waiting for the MSI version to crop up, so I'm happy to see it here at [H].

enben980 said:

Can someone tell me what happened in the Batman Arkham City benchmark; it just looks fishy to me.
I see in the 'highest-playable-settings' section that with physX enabled the GTX 670 takes a sizable hit in performance, yet the 660Ti does not? Then in the apples-to-apples section David turns off PhysX to compare all cards equally, yet the 660Ti does not increase in performance all that much. However, the 670 does. I see

*660Ti*
PhysX On: 29 , 77 , 46.8 (100 , 100 , 100)%
PhysX Off: 31 , 85 , 48.0 (106 , 110 , 102)%

*660Ti OC*
PhysX On: 32 , 86 , 51.3 (100 , 100 , 100)%
PhysX Off: 26 , 85 , 52.1 (81.2 , 98.8, 101)%

*670*
PhysX On: 27 , 83, 48.6 (100 , 100 , 100)%
PhysX Off: 30 , 105, 61.2 (111 , 126 , 125)%

These are some funky numbers just by disabling the PhysX feature. The overclocked 660 actually performed worse than the stock numbers, and the 670 turned in a huge leap in performance. I only bring this up as I just purchased this card two days before this article went live and I am excited that [H] is covering something I purchased before researching. In this game's page conclusion it is only mentioned that the cards destroy the Radeon's yet the numbers posted contradict the opinion at the bottom of the page. So I ask again, what happened in that benchmark?

Click to expand...

The differences between PhysX Off/Normal/High that you're seeing is fairly consistent with what I saw in my GIGABYTE GTX 670 review. At the end of the day though, the 660 Ti is playable in Batman at 2560x1600 as is the GTX 670. An overclocked 660 Ti isn't quite fast enough to handle High Physx (although the frame rate _looks_ playable it isn't), but an overclocked GTX 670 has no problems with it.

As for the conclusion, in this particular game, the GTX 660 Ti can run PhysX while turning in comparable frame rates to the AMD cards, which in my mind is a win for the GTX cards. We don't focus as much on frame rate as we do which card delivers the better in-game experience.

bonus said:

"We also observed some coil whine coming from the card..."

I didn't need to read that. No way I'm knowingly walking into that mess. Coil whine drives me insane.

You think they would have eliminated this already.

Click to expand...

F'in +1.

What the hell? MSI advertises their use of "MILITARY CLASS COMPONENTS!!11" just to have coil whine like any other manufacturer? F that noise.

EDIT: Lol, no pun intended (originally)

Nice article.

David_Schroth said:

The differences between PhysX Off/Normal/High that you're seeing is fairly consistent with what I saw in my GIGABYTE GTX 670 review. At the end of the day though, the 660 Ti is playable in Batman at 2560x1600 as is the GTX 670. An overclocked 660 Ti isn't quite fast enough to handle High Physx (although the frame rate _looks_ playable it isn't), but an overclocked GTX 670 has no problems with it.

As for the conclusion, in this particular game, the GTX 660 Ti can run PhysX while turning in comparable frame rates to the AMD cards, which in my mind is a win for the GTX cards. We don't focus as much on frame rate as we do which card delivers the better in-game experience.

Click to expand...

-Concerning PhysX:
So what you are saying, in reply to the number comparison, is that a 660Ti does NOT care if PhysX is turned on @ Normal setting; yet the GTX 670 does? Because on your Batman AC page you list all the GTX cards as using "Normal PhysX" levels for highest playable settings. Is that a type O on that page? Because here, you are saying a 670 can run at High PhysX and the 660Ti cannot, yet that is not reflected in the review. That would explain why the 670 sees a 25% FPS increase and the 660Ti only sees a 1% When PhysX = Normal @ On/Off. Clarification?

Borderlands 2 is going to use PhysX and I really really want to know what kind of impact that is going to have on this card. I could fiddle around with it myself when the game comes out, but that's why I read these articles. I get the info without the legwork per say.

-On an aside:
Does [H] use or plan to use frame time data to show irregularities in frame rate rendering such as micro stuttering and momentary freezes. I feel those are more important than what the average min-max frame rates are; or is that reflected in the line charts posted in these reviews. Since I see a lot of peaks and valleys when following the performance line over time. I would take a more solid 35-40fps card with no deviation than a card that can push and average of 75fps but with a lot of drops to the 25fps level, and information such as this can only help.

David, you say [H] reviews focus on in game experience. Wouldn't showing which cards suffer dropped frames and micro stutters at such different settings go to show which cards hold more performance per dollar in said games? This would help to reinforce those observed non-playable-performance levels you mention happen with the 660Ti when High PhysX is turned on. As an example.

enben980 said:

-On an aside:
Does [H] use or plan to use frame time data to show irregularities in frame rate rendering such as micro stuttering and momentary freezes.

Click to expand...

If my memory serves me correctly, Kyle has answered to this before in a few threads. Last time I read his comments on this, I think he mentioned that they were considering the option, but there is an expense connected to it.

nakedhand said:

If my memory serves me correctly, Kyle has answered to this before in a few threads. Last time I read his comments on this, I think he mentioned that they were considering the option, but there is an expense connected to it.

Click to expand...

Yep, a viable production solution is still being looked at. Hopefully soon.

Awesome! I look forward to that.

It's good to see another 660Ti review. Referencing the Galaxy reviews, it would seem the 3GB configuration on the Galaxy card doesn't provide quite the performance increase expected. The MSI looks very competetive against the 3GB Galaxy.

enben980 said:

-Concerning PhysX:
So what you are saying, in reply to the number comparison, is that a 660Ti does NOT care if PhysX is turned on @ Normal setting; yet the GTX 670 does? Because on your Batman AC page you list all the GTX cards as using "Normal PhysX" levels for highest playable settings. Is that a type O on that page? Because here, you are saying a 670 can run at High PhysX and the 660Ti cannot, yet that is not reflected in the review. That would explain why the 670 sees a 25% FPS increase and the 660Ti only sees a 1% When PhysX = Normal @ On/Off. Clarification?

Borderlands 2 is going to use PhysX and I really really want to know what kind of impact that is going to have on this card. I could fiddle around with it myself when the game comes out, but that's why I read these articles. I get the info without the legwork per say.

-On an aside:
Does [H] use or plan to use frame time data to show irregularities in frame rate rendering such as micro stuttering and momentary freezes. I feel those are more important than what the average min-max frame rates are; or is that reflected in the line charts posted in these reviews. Since I see a lot of peaks and valleys when following the performance line over time. I would take a more solid 35-40fps card with no deviation than a card that can push and average of 75fps but with a lot of drops to the 25fps level, and information such as this can only help.

David, you say [H] reviews focus on in game experience. Wouldn't showing which cards suffer dropped frames and micro stutters at such different settings go to show which cards hold more performance per dollar in said games? This would help to reinforce those observed non-playable-performance levels you mention happen with the 660Ti when High PhysX is turned on. As an example.

Click to expand...

For PhysX, a reference GTX 670 (which was used in this article) is not quite playable in my opinion in Batman with High PhysX. An overclocked 670, such as the GIGABYTE one that I reviewed was pushed up enough to make me satisfied at high PhysX. The goal of our evaluations isn't to get hung up on the frame rate performance - to do my run throughs, I start the game, get to the start point, get FRAPS running and simply play the game. The exact frame rate (even with a 25% difference between settings) is fairly meaningless if the game is smooth and responsive during the entire runthrough (as I'm sure you know that frames per second doesn't equal actual playability. As for what the 1% versus 25% means for the overall card's architecture, I'm not entirely sure - it would take far more testing of many games between PhysX on vs off to be able to draw any sort of reliable conclusion as to whether the difference that you see is a potential anomoly in test data versus something more interesting.

As far as Borderlands 2, I'd say it is impossible to determine how cards will fare with PhysX at this point as the game hasn't been released yet. The only thing that I would keep in mind, is that PhysX isn't necessarily intended to be useful on single card rigs with ~$300 video cards in them - if you dared to try using PhysX on a 560 Ti based system a year ago, you would have been less than impressed with the results. At the end of the day, I would imagine how a card handles running PhysX in game play is largely dependant upon how the game uses the GPU for graphics and I wouldn't rely on Batman figures to predict how Borderlands will run....

As for our ability for demonstrating microstutter in a correct manner, well, see Kyle's comment ;-). It's in the works, but in the mean time, our maximum playable charts and the related comments should document our thoughts/feelings about how a game is playable, as we play the game first and supply the graphs as a way to back up/quantify our opinion rather than writing a review based upon the numbers.

While the design of the Frozr IV cooling solution appears to be solid, its one fatal flaw is that the heat generated is not exhausted out of the case, which makes it far more important to have a well-designed case with plenty of airflow.

Click to expand...

Yeah, that's fatal as my case uses positive pressure via HEPA filters (I never have to clean inside, ever).

Total fail.

well either somone labeled them wrong, calculated the wrong averages and then called it apples to apples these tables/graphs make NO since.

sparks said:

well either somone labeled them wrong, calculated the wrong averages and then called it apples to apples these tables/graphs make NO sense.

Click to expand...

FTFY


But yeah, something seems off.

I'm also not surprised about the coil whine, MSI has been cheaping out lately.

settled then, i'ma get a MSI 660ti over a ref. 7950, not gonna take a chance with the ref, 7950 not hitting 1.1ghz core, let alone stay cool with the ref. cooling design, also crossfire/amd drivers still make me nervous.