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

[Grooveriding]

So defensive...

660ti is crap @ $300 and that is why nvidia is cutting its price. This thread is going to wind up as bad as the one about said cuts.

 

[SirPauly]

mistake -- double post, sorry!

 

[SirPauly]

Market adjusts -- nothing new -- and great to see from a consumer stand-point, imho!

 

[Rvenger]

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?

IMO AMD is cutting their prices to undercut Nvidia. You can find a 7870 for $200 if you look good and that price blows away both the 660ti and the 660. I have a feeling Nvidia will give the 660ti a $279.99 MSRP and $20 is not good enough.

 

[NIGELG]

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.

Shakes my head but not surprised that some would hit below the belt.

Come on,man give credit where it's due.7870 does 8AA much better.

 

[SirPauly]

IMO AMD is cutting their prices to undercut Nvidia.

Indeed! Isn't it wonderful?

 

[chimaxi83]

Yes, AMD is pretty much cementing, once again, their price/performance domination (as well as single card performance, but we already knew that).

 

[OILFIELDTRASH]

Im stuck in Nvidia hell. I be needins me Nvidia 3d visions.. If not for that I'd dump these gougers a long time ago. Im almost tempted to give up on the 3d due to this gouging.

 

[ZeroRift]

Hm, some interesting information on the 660ti memory bus:

Rather than take the usual route NVIDIA is going to take their own 3rd route: put 2GB of memory on the GTX 660 Ti anyhow. By putting more memory on one controller than the other two – in effect breaking the symmetry of the memory banks – NVIDIA can have 2GB of memory attached to a 192bit memory bus. This is a technique that NVIDIA has had available to them for quite some time, but it’s also something they rarely pull out and only use it when necessary.

Article

This makes me think that the OEM that slapped 3GB of ram onto the 660ti did it wrong. They may have assumed they could load each of the controllers equally, when in reality the GPU architecture was not built to accommodate that....

THG may also be running into the "worst case scenario" with their test setup:

The best case scenario is always going to be that the entire 192bit bus is in use by interleaving a memory operation across all 3 controllers, giving the card 144GB/sec of memory bandwidth (192bit * 6GHz / 8). But that can only be done at up to 1.5GB of memory; the final 512MB of memory is attached to a single memory controller. This invokes the worst case scenario, where only 1 64-bit memory controller is in use and thereby reducing memory bandwidth to a much more modest 48GB/sec.

Regardless, knowing that the controllers are "hacked" to use a power of 2 memory quantity on a 192-bit bus, it's not suprising that the 660ti struggles in memory intensive situations.

 

[toyota]

if there was an issue with the memory bus and odd 2gb configuration then it would show up when exceeding 1.5 gb. that card is starved long before its exceeding 1.5 gb.

I can tell you from my own experience that the odd configuration has no impact in the real world. my card has 192bit bus and 1gb of memory and it scales fine when exceeding 768mb and there are no drop offs in performance at all.

the 3gb card in the review might just have issues or maybe its not boosting all that high. heck maybe having 3gb of ram is causing more latency than 2gb.

 

[tviceman]

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.

Well the article is called "exploring memory bandwidth" and they didn't really "explore" memory bandwidth. They just ran a bunch of benchmarks. Increasing or decreasing the memory bandwidth with all the same tests would have provided a more accurate look at what is going on.

 

[tviceman]

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.

CRUUSADE RAWRARARWWR where exactly did I defend anything Nvidia related in this thread? Let's see... nowhere. I never said the gtx660ti was great, good, or even bad for now or for the future. I never said it was a good deal. I have never made any discussions about AA, other than to say 8X over 4X is visually almost entirely not noticeable. Please show me where I have said otherwise. Until then, apologize for accusing me of doing things I obviously am not doing and stop being confrontational. The only thing I said in this thread is the article isn't meaningful w/o increasing the memory bandwidth to see if that alleviates AA bottlenecks, that there looks like a driver / performance glitch in the [h] benchmark, AND that 8X AA is not some new standard in which all benchmarks should be ran with.

 

[chimaxi83]

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.

Tell me about it. The NV defenders mobilize instantly as soon as anything negative is said. Unbiased, objective views or discussion on video cards is beyond their abilities, obviously. D:

 

[Tuna-Fish]

it still makes no sense for the 7870 to be beating the 670 with 8x AA.

Why? The test is all about rops. The 7870 has ~10% more raw rop throughput, and it's rops are probably more efficient.

 

[toyota]

Why? The test is all about rops. The 7870 has ~10% more raw rop throughput, and it's rops are probably more efficient.

because the 670 does not lose to the 7870 in this game on any other review. in fact the 670 beats the 7950 and matches the 7970 in Batman in every other review. and how do you figure the 7870 has has 10% more throughput? 670 and 7870 both have 32 ROPs and the 670 will always boost over 1000mhz which is what the 7870 is clocked at.

 

[Keysplayr]

because the 670 does not lose to the 7870 in this game on any other review. in fact the 670 beats the 7950 and matches the 7970 in Batman in every other review. and how do you figure the 7870 has has 10% more throughput? 670 and 7870 both have 32 ROPs and the 670 will always boost over 1000mhz which is what the 7870 is clocked at.

He didn't think it through. Yeah, it doesn't make any sense at all. As you said, 670 rivals 7970 in Batman AC in almost all other reviews out there. So 670 losing to a 7870 in this title isn't really plausible.

 

[blackened23]

Tangent time, I think all PC related silicon manufacturers should create a united front to crush Apple. I don't know why I feel compelled to state this but I really despise apple and their litigious ways. I find it hilarious that Apple attempted to invalidate HTC's 4G patent portfolio (so they could sell the iphone 5 royalty free, scumbags) , after all of the nonsense in the samsung trial. I really, truly hate apple and their idiot fans.

Don't mind me, carry on.

 

[Keysplayr]

Google
Google
::: Looks for Apple suits against AMD :::
Google
Google

 

[SlowSpyder]

Isn't the 660 the card that has unequal bandwidth to the memory chips? 2 x 64 bit controllers to 1GB, and a single 64 bit controller to the other 1GB? I wonder how that may play into it as more of that 1GB with just the single 64 bit connection has to be used.

 

[toyota]

Isn't the 660 the card that has unequal bandwidth to the memory chips? 2 x 64 bit controllers to 1GB, and a single 64 bit controller to the other 1GB? I wonder how that may play into it as more of that 1GB with just the single 64 bit connection has to be used.

it makes little to no difference and I just addressed that in post #60

 

[SPBHM]

Isn't the 660 the card that has unequal bandwidth to the memory chips? 2 x 64 bit controllers to 1GB, and a single 64 bit controller to the other 1GB? I wonder how that may play into it as more of that 1GB with just the single 64 bit connection has to be used.

they also tested a 3GB card, which have the same amount per channel...
doesn't seem to make any significant difference...

also older cards like the GTX 550 TI, GTX 460 V2, GTX 560 SE (192bit with 1-2GB) also work in the same way.

 

[SlowSpyder]

it makes little to no difference and I just addressed that in post #60

Lol... I was too eager to post, read the first 50 posts then skimmed the last page, my bad.

 

[SlowSpyder]

because the 670 does not lose to the 7870 in this game on any other review. in fact the 670 beats the 7950 and matches the 7970 in Batman in every other review. and how do you figure the 7870 has has 10% more throughput? 670 and 7870 both have 32 ROPs and the 670 will always boost over 1000mhz which is what the 7870 is clocked at.

I remember a lot of talk regarding AMD not being able to have a 32 ROP/3GB card before the 7970 shipped, but obviously that happened. Because of those posts, I remember a part of the 7970 review sticking out when I read it, how AMD put some extra magic pixie dust into their ROP's. I'm not sure if maybe that has something to do with it?

http://www.anandtech.com/show/5261/amd-radeon-hd-7970-review/4

As it turns out, there’s a very good reason that AMD went this route. ROP operations are extremely bandwidth intensive, so much so that even when pairing up ROPs with memory controllers, the ROPs are often still starved of memory bandwidth. With Cayman AMD was not able to reach their peak theoretical ROP throughput even in synthetic tests, never mind in real-world usage. With Tahiti AMD would need to improve their ROP throughput one way or another to keep pace with future games, but because of the low efficiency of their existing ROPs they didn’t need to add any more ROP hardware, they merely needed to improve the efficiency of what they already had.

The solution to that was rather counter-intuitive: decouple the ROPs from the memory controllers. By servicing the ROPs through a crossbar AMD can hold the number of ROPs constant at 32 while increasing the width of the memory bus by 50%. The end result is that the same number of ROPs perform better by having access to the additional bandwidth they need.

 

[Wall Street]

He didn't think it through. Yeah, it doesn't make any sense at all. As you said, 670 rivals 7970 in Batman AC in almost all other reviews out there. So 670 losing to a 7870 in this title isn't really plausible.

How many reviews use 12.3 drivers instead of 12.7? How many use FXAA (High) instead of 8x MSAA?

I agree that almost all other reviews show the GTX 670 ahead, however, I don't recall any using this fairly unusual combination of settings.

 

[SirPauly]

That is why I like the over-all gauge from Computerbase.de -- may not be ideal for all but offers a solid estimate on performance with x4 AA and x8 AA over-all.