(Tom's) Seven GeForce GTX 660 Ti Cards: Exploring Memory Bandwidth

[Keysplayr]

Batman:AC is using a deferred Renderer. So it's hunting for bandwidth when MSAA is applied. AMD optimized their 8xMSAA performance with the latest drivers.

Interesting.

 

[toyota]

I think its more of an ROP issue in this game for the 660 ti. look here how even with the memory oced to 7600 on the 660 ti, its still getting killed by the 670 and 7950.

 

[Keysplayr]

I think its more of an ROP issue in this game for the 660 ti. look here how even with the memory oced to 7600 on the 660 ti, its still getting killed by the 670 and 7950.

It's interesting how it is so much lower than the others in the beginning but then hangs right with them 1/3rd into the graph til the end. Wonder why that is. Especially when the most taxing part of the benchmark is the last 15 seconds or so. Unless they aren't using the built in benchmark?

 

[KompuKare]

Anything else you'd like to add about bumpgate to this conversation about the 660Ti?

Hm, no really happy to leave it at that...

Without having overclocked the vram on any of the gtx660ti's they used for this article, it went from interesting and informative to dull and repetitive. I'd like to know if the drop off in performance is more of a result of shaving off a memory controller and losing the ROP's, or if it's more from the lower bandwidth itself.

This is a good point. Although we do know that THG is hardly known for their scientific approach. It would be nice for someone to do a proper analysis of the effect of memory etc. of the current 28nm GPUs. But I can see that being a massive undertaking.

And while they're at it measure the the power usage too: I'm sure the amount of memory, bus width and memory clock have an effect on wattage too. As do the choice of PCB components: VRMs get rather hot and efficiency must be major factor there too. Difference cards from different AIB vary a fair bit in their power usage and I don't think that's just because of binned chips and their voltages.

A lesson for AMD's PR department would be to cherry pick stuff for reviews although I would personally like to see techsites buying retail cards to review...

A roundup like that would be very interesting but which website is willing to put in all that work for us?

 

[tviceman]

I think its more of an ROP issue in this game for the 660 ti. look here how even with the memory oced to 7600 on the 660 ti, its still getting killed by the 670 and 7950.

That could be it, but man that graph sure does look like there is a driver glitch going on more than anything. If you're right, then the benchmark [h] uses is essentially ROP heavy for the first 1/3 of the run, then it's not at ROP heavy at all. I don't know. That would be really strange within a game benchmark to be extremely reliant on ROP's then BAM, be normal. That looks like a driver issue to me.

 

[3DVagabond]

Listen, this is not meant to be all encompassing. It's just stressing the cards in a particular way. The same idea as cranking tessellation. It reveals weakness in a card in a particular area.

As far as O/C'ing anything goes. [H] already did that comparison HERE. The 660ti really falls off.

 

[exar333]

I don't see why anyone would get this card when you can grab a 7950 for ~$318AR plus a free game. *Shrug*

 

[SirPauly]

Imho,

AMD has always done well with x8 AA -- and goes as far back as the 2900XT. Actually, even with their x6 AA offerings with earlier architectures. After all, ATI/AMD offer wonderful competition and choice to consider.

AMD has always had strong price/performance and goes as far back as the 3870/3850.

http://www.computerbase.de/artikel/grafikkarten/2012/test-nvidia-geforce-gtx-660-ti/7/

Notice the HD 6970 offer more performance than the GTX 580 with x8 AA? None of this is really surprising and new.

I see 17 percent more performance over-all for a GTX 670 with 1920 x 1080/x8 AA over-all -- compared to a GTX 660 TI -- for a 33 percent difference in MSRP. There has to be some differences based on one doesn't desire to cannibalize sales of the GTX 670 and GTX 680, one may imagine.

The GTX 660 ti offered a well placed price/performance ratio when the HD 7870 had a 299 MSRP and the HD 7950 had a 349 MSRP -- the landscape pricing has changed and now AMD offers even more stiff competition and even more impressive 28nm price/performance.

IF sales suffer for the GTX 660ti -- the market may adjust pricing.

 

[3DVagabond]

[H] doesn't use canned benchmarks. They play the game. It's hard to know what was going on at that point to cause the 660ti to drop down so much.

 

[Grooveriding]

The Kepler lineup top to bottom is crappy in the memory bandwidth department and when using high levels of AA. At high resolutions the GTX 680 is not much faster with 8xAA compared to the GTX 580, pretty sad.

You see this trickle down to all the cards. The 660ti with it's emphasised bandwidth handicap really shows how bad it is. The card should of never been $300, ridiculous. $250 at most.

 

[Keysplayr]

I don't see why anyone would get this card when you can grab a 7950 for ~$318AR plus a free game. *Shrug*

Does the 660Ti come with Borderlands2? Which is also a PhysX title so a nice match. Also, I suspect there will be more price drops from both camps before this war is over.

 

[Rvenger]

Imho,

AMD has always done well with x8 AA -- and goes as far back as the 2900XT. Actually, even with their x6 AA offerings with earlier architectures. After all, ATI/AMD offer wonderful competition and choice to consider.

AMD has always had strong price/performance and goes as far back as the 3870/3850.

http://www.computerbase.de/artikel/grafikkarten/2012/test-nvidia-geforce-gtx-660-ti/7/

Notice the HD 6970 offer more performance than the GTX 580 with x8 AA? None of this is really surprising and new.

I see 17 percent more performance over-all for a GTX 670 with 1920 x 1080/x8 AA over-all -- compared to a GTX 660 TI -- for a 33 percent difference in MSRP. There has to be some differences based on one doesn't desire to cannibalize sales of the GTX 670 and GTX 680, one may imagine.

The GTX 660 ti offered a well placed price/performance ratio when the HD 7870 had a 299 MSRP and the HD 7950 had a 349 MSRP -- the landscape pricing has changed and now AMD offers even more stiff competition and even more impressive 28nm price/performance.

IF sales suffer for the GTX 660ti -- the market may adjust pricing.

Well said Sirpauly.

 

[Arzachel]

Just like extreme tessellation factors, 8x MSAA is a huge performance penalty for little to no IQ gain. Anyone making sweeping statements based on this kind of outliers is just making themselves look silly. The test itself is pretty interesting though, I wonder what makes the 670 score so poorly?

 

[SirPauly]

That could be it, but man that graph sure does look like there is a driver glitch going on more than anything. If you're right, then the benchmark [h] uses is essentially ROP heavy for the first 1/3 of the run, then it's not at ROP heavy at all. I don't know. That would be really strange within a game benchmark to be extremely reliant on ROP's then BAM, be normal. That looks like a driver issue to me.

I thought the same thing:

http://www.hardocp.com/image.html?image=MTM0NjA2MDA5M09tVGRTMlE0eDNfNF81X2wuZ2lm

 

[Keysplayr]

I thought the same thing:

http://www.hardocp.com/image.html?image=MTM0NjA2MDA5M09tVGRTMlE0eDNfNF81X2wuZ2lm

It kind of looks like it takes a performance hit because it takes longer for it to load the textures into the buffer. Could also be a memory management issue that could be handled with newer drivers. Just thinking out loud.

 

[tviceman]

It kind of looks like it takes a performance hit because it takes longer for it to load the textures into the buffer. Could also be a memory management issue that could be handled with newer drivers. Just thinking out loud.

I think this is what is going on. I don't think less ROP's are the primary cause for gtx660ti's drop off in performance when enabling high AA levels. I think it's a matter of memory bandwidth.

 

[SirPauly]

The Kepler lineup top to bottom is crappy in the memory bandwidth department and when using high levels of AA. At high resolutions the GTX 680 is not much faster with 8xAA compared to the GTX 580, pretty sad.

You see this trickle down to all the cards. The 660ti with it's emphasised bandwidth handicap really shows how bad it is. The card should of never been $300, ridiculous. $250 at most.

The irony of your statement based on Computerbase.de' findings. The GTX 660ti is much faster with x8 over the GTX 580 than x4 AA over-all.

http://www.computerbase.de/artikel/grafikkarten/2012/test-nvidia-geforce-gtx-660-ti/7/

 

[SirPauly]

Just like extreme tessellation factors, 8x MSAA is a huge performance penalty for little to no IQ gain. Anyone making sweeping statements based on this kind of outliers is just making themselves look silly. The test itself is pretty interesting though, I wonder what makes the 670 score so poorly?

That may be subjective -- when FarCry was very popular in the training level -- the air craft carrier from a distance, moving towards it, had very noticeable aliasing -- even with x4 AA. With longer detailed near verticals and horizontals, with strong color contrasts, this is where more than x4 AA was needed to me and added immersion. Would post about how x4 is not enough and x8 AA would be welcomed. Also would post about the need of innovation with AA due to shader aliasing, specular aliasing, alpha test aliasing, improvements in filtering and over-all IQ flexibility for gamers.

X8 AA is very welcomed -- so are the features like CSAA, CFAA, SSAA, transparency, FXAA, MLAA and even TXAA. Tools and features gamers may use to find the right balance for their subjective tastes and tolerances.

 

[Cerb]

I'm all for cranking up the quality, but am I the only person that finds 8x MSAA a silly setting to use? When 4x MSAA doesn't solve your jaggies, 8x isn't going to do any better.

Sometimes 8xQ (8x-no-Q is pretty much pointless) or performance-taxing SGSSAA could be needed, for some games. 4x can partly remove jaggies, but replace some angles with crawlies in the process, and won't well work for all moving edges. But, doing so is taking a sledgehammer to the IQ problem, like good old OGSSAA/RGSSAA.

That's where the shader AA really shines. It still looks worse in a screenshot, but playing a game at nominal framerates, the combination of light MSAA/SGSSAA and shader AA (I prefer FXAA, but if you like MLAA, go for it) gets rid of basically all jaggies, sparklies, and crawlies, and doesn't hurt performance much compared to no AA.

With both brands constrained by TSMC, and the 660 Ti having a few popular titles where it really does smash AMD (BF3, FI), it may not drop in price much for awhile, yet.

 

[blastingcap]

Maybe the 7870 isn't applying AA to the entire scene. Magic drivers and all. That would be some reveal. Might be worth looking into.

I'm going to hope you were joking...

Without having overclocked the vram on any of the gtx660ti's they used for this article, it went from interesting and informative to dull and repetitive. I'd like to know if the drop off in performance is more of a result of shaving off a memory controller and losing the ROP's, or if it's more from the lower bandwidth itself.

While I agree, I remember a lot of people defending Kepler saying stock vs stock was the better comparison, oc headroom be damned, because most people don't oc etc. etc. I would like to see consistency from those people: either advocate for oc in general, or don't. Not saying you were one of them.

 

[chimaxi83]

Maybe the 7870 isn't applying AA to the entire scene. Magic drivers and all. That would be some reveal. Might be worth looking into.

I'm pretty sure I've heard it all now. Maybe Apoppin will look into this for us.

FWIW, I also think going by 8x AA is kinda meh for single card configurations. Just because a card "wins" doesn't make gameplay enjoyable on it /shrug

 

[RussianSensation]

I don't see why anyone would get this card when you can grab a 7950 for ~$318AR plus a free game. *Shrug*

Every review that actually tested GTX660Ti with MSAA and mods across the board shows it's not worth the $300 asking price. Whether 660Ti is ROP or memory bandwidth limited (or both) are just technicalities, but do not change the final conclusion.

The fact of the matter is 660Ti OC gets hammered when you use higher resolution textures and/or pile on MSAA against a 7950 OC. When the competitor offers no compromises in this regard to image quality and performance with overclocking at a similar price level, 660Ti at $300-$330 seems like a very poor recommendation on an enthusiast forum.

Since NV is rumored to drop prices on 660Ti shortly, NV knows this too. As it stands, no enthusiasts should touch a 660Ti @ $300-330 price level. Since when do we recommend a PC product that's 20-25% slower with MSAA/texture mods against a similarly priced competitor and has worse overclocking headroom? It makes no sense since we overclock parts to maximize performance/$ and that allows us to increase image quality. Performance/$ is exactly the reason why GTX670 became so popular against the 680 and overclocking is the reason 5850/GTX460 cards were so popular on enthusiast forums. It's pretty obvious the HD7950 is the real value and 660Ti is just a poser.

660Ti is severely overpriced against the 7870 for non-overclockers since it costs 25-30% more for barely 10% more performance overall.

HD7950 at least gets a pass since HD7950 OC = GTX670 OC. That tells the whole story why it's worth $300-330. GTX660 fails in all key areas at $300: Lacking enthusiast performance and not offering good price/performance. It can't handle MSAA or texture mods and already GPU Boosts to very high levels leaving almost no room for overclocking. NV also charges $20-30 more for 3GB of VRAM while 7950 has that for free.

 

[SirPauly]

Worth is subjective to the individual and the market place to me.

 

[VulgarDisplay]

Exactly. I was spouting this a few weeks ago, but the folks I was in an argument with kept glazing over it. It seems 8X AA is becoming the new minimum performance metric, although the difference in image quality between 4X and 8X AA is all but non-existent during live action.

It's honestly all about pushing these cards as hard as they can be at this time because there is no game they have trouble running. What this reveals is that as games get more demanding the 660ti is going to start to suffer, as many people have been saying since its release.

Also, the thing I don't understand why it's all the sudden acceptable to do things like claim MSAA in deferred games is pointless, or that high levels of MSAA is pointless in a GPU review. The only difference I see now is that Nvidia is painted in a bad light. Ever since Battlefield 3 came out people have been stating that MSAA doesn't count because it doesn't work on the whole scene, and that AMD cards running FXAA only performed much closer to Nvidia counterparts.

No one accepted that argument back then. Now all the sudden it's the argument you are using to defend this overpriced Nvidia GPU?

If people want to buy the 660ti that's fine, but to be on some crusade to defend it as worth the money and a great card for future games is absurd.

 

[SirPauly]

The irony is I feel that the GTX 660ti did such a nice job with AA, even with x8 AA, that it's competition and performance helped reduce AMD's pricing. Also felt it was placed in a nice price position with a 299 MSRP HD 7870 and a 349 MSRP HD 7950. The technology pricing landscape changes swiftly.

Now, with different MSRP price-points from AMD -- GTX 660TI is really comparing more-so with a HD 7950 than a HD 7870 and this is very stiff competition, imho.

I think it is great that not only AMD may be forced to drop pricing but nVidia may be forced to drop pricing, too. Pricing adjustments -- nothing new, but when both are doing it potentially -- translates in consumers winning. What's there to complain about?