tweaktown review GTX660Ti

[Silverforce11]

Another day dreaming.NV has way better relationship with game devs than amd.Funny thing is most of the games u consider worthwhile for benches are sponsored by NV.

Let's wait for more reviews shall we??

The final judgement will be in once good review sites have a go at it, and it's all going to come down on pricing.

 

[Jaydip]

Amen to that.I find it pretty funny people took TT so seriously.I never bother with their reviews.

 

[blackened23]

U honestly have no idea of vram usage.U tout 1Gb extra ram as a redeeming feature in all of ur posts.Look at the [H] reviw of 680,even the 580 was able to beat the 7970 in skyrim using 1.5Gb vram.Every Gpu architecture allocated vram in a different way.U can't compare them directly.For eg Bf3 used more vram in 7970 while still being slower.Unless u have access to the game code u just can't comment on how vram is being allocated.Another thing
"Actually with latest drivers GTX680 lost in 14 out of 17 tests at 2560x1600 to HD7970 GE"
So 680 can't improve on that?seriously check ur facts before posting.NV has always been able to squeeze tremendous performance gains from their drivers.
"Also, Kepler's performance in Dirt Showdown and Sniper Elite V2 is terrible. Who knows if more games will use directcompute and contact hardening shadows. Another knock against the 660Ti."
Another day dreaming.NV has way better relationship with game devs than amd.Funny thing is most of the games u consider worthwhile for benches are sponsored by NV.

Just so you know, this is actually not true. All GPUs allocate VRAM the exact same way. Keep in mind that nearly every game on the market has game assets not exceeding 500MB (game assets include textures and such), anything above that is anti aliasing, tessellation, and so forth and so on. No GPU does anything fundamentally different when allocating VRAM, that is impossible.

Regarding BF3, several game engines scale their VRAM cache depending on the amount of available VRAM, frostbite 2 is one of those and crytek 2 is another. Most engines do not do this, but basically more VRAM has a bigger cache which is more efficient for more VRAM but doesn't necessarily increase performance. Technically, cards with more available free VRAM are supposed to get higher LOD but I haven't seen a difference that is perceivable. Currently these are the only 2 engines that do this to my knowledge.

Finally, it is very possible to run out of VRAM but you won't in a single screen resolution unless you use override AA or override SGSSAA. Remember, AA is the largest chunk of VRAM use in all cases. You will definitely run out of VRAM in surround resolutions with 2gb if you use too much anti aliasing, you can be rest assured of that. Usually that means in surround you have to use no AA or FXAA, it just depends on the game engine and the amount of space that game assets takes. You can even do 2x MSAA in some games. But what you cannot do is go 8x SSAA in surround or even 4x MSAA in some games, you will run out of VRAM at 5760x1200.

So my points are two fold - you are absolutely incorrect that different GPUs do different things with VRAM. You can swap a 2gb 6970 for a 2gb GTX 680 and all of your games will use the same exact amount of VRAM as long as all of your AA / IQ settings are matched, try it. Furthermore, VRAM can matter but it doesn't for single screen. More than 2 GB is definitely beneficial for surround, it gives you more image quality options but again this depends on what games you play. One game may force you to use FXAA for 5760x1200 while another lets you do 2x MSAA with 2GB - it just depends on which game and what settings you use. Its also possible that some games will fail with 2gb at 5760x1200. And this is a 3rd point, but do you realize how stupidly annoying it is to read your posts with "U" and all that crap? I get it, you're on an iphone, I don't suppose you could add 30 seconds and type words out? I know i'm not the only one that hates seeing that crap.

Anyway, i'll resume eating my popcorn.

 

[Jaydip]

No it is you who is misinformed.Memory allocation just doesn't work that way.
"So my points are two fold - you are absolutely incorrect that different GPUs do different things with VRAM. You can swap a 2gb 6970 for a 2gb GTX 680 and all of your games will use the same exact amount of VRAM as long as all of your AA / IQ settings are matched, try it"
Absolutely wrong.I think you are confusing the vram usage shown by various software with the actual vram allocation algorithm.My point was you can't compare between architectures in the same way.

 

[blackened23]

You honestly have no clue dude, you're looking at benchmarks on websites as a support for your argument. You have no idea what is going on behind the scenes, some game engines scale the VRAM cache to max available VRAM but that doesn't change the fact that game assets and anti aliasing don't require different amounts of VRAM on different GPUs. Feel free to compare any AMD 2gb card to an nvidia 2gb card in various games such as metro 2033 if you don't believe that (exclude crytek 2 or frostbite 2, those engines scale their VRAM cache). If you were familiar with the intricacies of programming games and fitting them into that VRAM, you would be well aware that GPU architecture has absolutely nothing to do with VRAM usage.

If you're clueless about a subject stick to what you know. And you don't know, stick to your lane.

 

[Jaydip]

It seems it is u who have never written a single line of code in your life.I know what I am talking about.Have u even written device drivers at all?Because I have and I know what I am talking about.Stop copying from wiki.

 

[f1sherman]

Even different drivers will sometimes allocate VRAM differently differently, or have different memory management

For example, GTX 570 stuttering on BF3 with maxed settings, due to running out of VRAM, yet being fine the with next driver revision

game assets and anti aliasing don't require different amounts of VRAM on different GPUs


I have never written a device driver , but I have a hunch this is wrong.
For one thing, MSAA having different quality on different GPUs have been observed.

How about texture compression, filtering?
Is it exactly the same, no matter what driver, what GPU, AMD/NV...?
I kinda doubt it.

 

[RussianSensation]

U honestly have no idea of vram usage.U tout 1Gb extra ram as a redeeming feature in all of ur posts.Look at the [H] reviw of directly.For eg Bf3 used more vram in 7970 while still being slower. 680,even the 580 was able to beat the 7970 in skyrim using 1.5Gb vram.Every Gpu architecture allocated vram in a different way.U can't compare them

NV and AMD GPUs will use VRAM differently but if a game needs more than 1GB of VRAM, there is NOTHING in the world NV can do to stop a 1GB card from tanking in those games. SKYRIM, Max Payne 3 are recent examples where 1GB of VRAM is not enough. Games such as Shogun 2 on Ultra at 1080P will use up more than 1.5GB of VRAM and tank completely on something like a GTX570 1.28GB, which is why AnandTech's tests still don't use Ultra at 1080P for Shogun 2.

Don't mix and match HD6950 2GB vs. GTX560Ti 1GB threads with this one. You aren't even comparing apples and oranges here.

Maybe we should have a survey here who would take a GTX560Ti 1GB over an HD6950 2GB? I would bet you most people here wouldn't even touch a $180 1GB GTX560Ti with a 30-foot pole. You say I have no idea about 1GB VRAM bottleneck, and yet you dismissed this in another thread as irrelevant:

1GB of VRAM is becoming a serious bottleneck in some games. You can dismiss it all you want but there is no way the GPU in the GTX570 is 40-100% faster than GTX560Ti.

The poor performance in Dirt Showdown and Sniper Elite V2 has little to do with VRAM problem for the 680 aside from 3 monitors where ideally you'd want at least 3GB. Somehow though out of all the things in my post you picked up 3GB VRAM and missed everything else?

I honestly can't recall GTX580 beating HD7970 in SKYRIM with AA on, and with latest drivers it's not even close, especially at 2560x1600. Since we are discussing how cards stack up today, that's pretty relevant.

It's not my fault AMD's driver team has fixed most of their driver problems with Dirt 3, SKYRIM, Battlefield 3 with Cats 12.7Beta. I recommended GTX670/680 cards for about 3 months actually. Now, with prices drops on the 7950/7970 and Cats 12.7B, the tides have turned imo.

NV's driver relationships may be good but while AMD has made up the huge deficit in BF3, SKYRIM, Batman AC and Dirt 3, NV has done almost nothing to cut the lead AMD has in these titles:

- Anno 2070, Serious Sam 3, Bullet Storm, Metro 2033, Alan Wake, Crysis 1/Warhead, Dirt Showdown, Sniper Elite V2, etc. This is why the number of games in which GTX680 is faster is getting smaller and smaller.

Of course you can get a factory preoverclocked GTX680 it does well against the 1050mhz HD7970. But those early HD7970 vs. GTX680 reviews in March are pretty much irrelevant at this point. All the latest review from Xbitlabs, TechPowerup, TechReport, TechSpot, BitTech, KitGuru, Tom's Hardware, AnandTech have declared HD7970 GE the fastest GPU now.

Launch time:
GTX680 $500 > $550 HD7970 925mhz

Today:
Sapphire Vapor-X $450 > GTX680 $500

And NV charges $580 for reference GTX680 4GB. I think GTX660Ti is a good time for NV to drop prices $30-50.

 

[3DVagabond]

really? is your math that bad?

if the 660ti was $150 at launch then the 670 would cost TWICE (2x) as much.
2x 150 = 300

if the 660ti cost $250 the the 670 would be 1.6 times as much
1.6 x 250 = 400

if the 660ti cost 270 then the 670 would be ~1.5x as much

And for your specific 300 to 400 hundred inquiry:

if the 660ti cost 300 then the 670 would be around 1.35x as much


i am guessing that once the 660ti launches it could be bought for less than $299. Could be wrong. But i think it should be priced between the 250 and 300 mark. just over the half way point. But thats my figuring.

But even if it doesnt and is it is $299. You could pay 1.35-1.4 times more for a 670 which performs up to 20% better in highly stressed scenarios. I just dont see how you can make it sound bad???? it sounds pretty good to me. Especially considering the target mid-grade market. They will get out of the box ~7950 performance for a card costing the same as the 7870. How the heck can you paint this bad????? Even if the 660 ti is as slow as the 7870 in some cases, its *faster* than the 7950 in many others. This is at stock, without overclocks.

Its pretty amassing to see so many down play this card. It will be priced like the 7870 and can outperform the 7950. This is plan truth. Even in the highest AA 1600p test, it swaps licks with the 7950. A toss up. Does the 7870 ever win against the 7950 at 1600p with max AA piled on? Does it ever?

the 660ti will be priced like the 7870 currently is. bellow the 300 mark. This is not where the 7950 is priced. if you cannot see the potential for the 660ti then i feel for ya. Downplaying it takes some serious effort

Is my math bad? Whatever! It's not 160% more, and you know it. As far as your comparison to the 7950, I've already stated why I'd compare it to the 670, instead. You are ignoring that, because it doesn't fit your agenda. /end

 

[AnandThenMan]

It seems it is u who have never written a single line of code in your life.I know what I am talking about.Have u even written device drivers at all?Because I have and I know what I am talking about.Stop copying from wiki.

I assume a spacebar is not required to write device drivers. D:

 

[Arzachel]

Real men set their IDEs to insert several spaces with tab and just delete some off occasionally

 

[Jaydip]

NV and AMD GPUs will use VRAM differently but if a game needs more than 1GB of VRAM, there is NOTHING in the world NV can do to stop a 1GB card from tanking in those games. SKYRIM, Max Payne 3 are recent examples where 1GB of VRAM is not enough. Games such as Shogun 2 on Ultra at 1080P will use up more than 1.5GB of VRAM and tank completely on something like a GTX570 1.28GB, which is why AnandTech's tests still don't use Ultra at 1080P for Shogun 2.

Don't mix and match HD6950 2GB vs. GTX560Ti 1GB threads with this one. You aren't even comparing apples and oranges here.

Maybe we should have a survey here who would take a GTX560Ti 1GB over an HD6950 2GB? I would bet you most people here wouldn't even touch a $180 1GB GTX560Ti with a 30-foot pole. You say I have no idea about 1GB VRAM bottleneck, and yet you dismissed this in another thread as irrelevant:

1GB of VRAM is becoming a serious bottleneck in some games. You can dismiss it all you want but there is no way the GPU in the GTX570 is 40-100% faster than GTX560Ti.

The poor performance in Dirt Showdown and Sniper Elite V2 has little to do with VRAM problem for the 680 aside from 3 monitors where ideally you'd want at least 3GB. Somehow though out of all the things in my post you picked up 3GB VRAM and missed everything else?

I honestly can't recall GTX580 beating HD7970 in SKYRIM with AA on, and with latest drivers it's not even close, especially at 2560x1600. Since we are discussing how cards stack up today, that's pretty relevant.

It's not my fault AMD's driver team has fixed most of their driver problems with Dirt 3, SKYRIM, Battlefield 3 with Cats 12.7Beta. I recommended GTX670/680 cards for about 3 months actually. Now, with prices drops on the 7950/7970 and Cats 12.7B, the tides have turned imo.

NV's driver relationships may be good but while AMD has made up the huge deficit in BF3, SKYRIM, Batman AC and Dirt 3, NV has done almost nothing to cut the lead AMD has in these titles:

- Anno 2070, Serious Sam 3, Bullet Storm, Metro 2033, Alan Wake, Crysis 1/Warhead, Dirt Showdown, Sniper Elite V2, etc. This is why the number of games in which GTX680 is faster is getting smaller and smaller.

Of course you can get a factory preoverclocked GTX680 it does well against the 1050mhz HD7970. But those early HD7970 vs. GTX680 reviews in March are pretty much irrelevant at this point. All the latest review from Xbitlabs, TechPowerup, TechReport, TechSpot, BitTech, KitGuru, Tom's Hardware, AnandTech have declared HD7970 GE the fastest GPU now.

Launch time:
GTX680 $500 > $550 HD7970 925mhz

Today:
Sapphire Vapor-X $450 > GTX680 $500

And NV charges $580 for reference GTX680 4GB. I think GTX660Ti is a good time for NV to drop prices $30-50.

Rs I am not talking about this
"Don't mix and match HD6950 2GB vs. GTX560Ti 1GB threads with this one. You aren't even comparing apples and oranges here." here.I am not saying more vram is bad but more vram doesn't necessarily make 7970 a faster card.I was talking about the memory allocation algorithm on NV and AMD and how can't you directly compare them.
Check this
http://www.hardocp.com/article/2012/03/22/nvidia_kepler_gpu_geforce_gtx_680_video_card_review/6
580 is faster than 7970 here.
Now where it needs more than 1Gb vram on a 1Gb card it will swap the data from the scratchpad(for NV). As it is using vm it will be slow no doubt but it will run.I believe 660Ti will have 2Gb vram so 1Gb vram bottleneck doesn't apply here.I think Msi 680 lighning is fastest GPU according to a recent [H] review?But it needs insane tweaks so lets drop it.

 

[Jaydip]

Real men set their IDEs to insert several spaces with tab and just delete some off occasionally

lol:biggrin:

 

[RussianSensation]

http://www.hardocp.com/article/2012/03/22/nvidia_kepler_gpu_geforce_gtx_680_video_card_review/6580 is faster than 7970 here.

That review dates to March 22. Maybe GTX580 was faster but GTX680 also wasn't that fast in Shogun 2. Talking about today's performance though.

I think Msi 680 lighning is fastest GPU according to a recent [H] review?But it needs insane tweaks so lets drop it.

Ya, it is. Only 4 games tested though. It took a 1364-1392mhz GPU boost to do it. Too bad that card costs $600 vs. $450 for the Vapor-X. I don't call that winning. Also, for close to that you can get 2x HD7950 MSI FT3. That's probably not the point but $600 for a card that needs 1.367 Volts under load to get those clocks is just for benchmarking. I am having doubt people are going to risk ruining their $600 piece of hardware by putting 1.367V into it on air. In 2 of those 4 games, a 1392mhz GTX680 netted 1.8 fps in BF3 and 1.2 fps in Witcher 2 vs. an overclocked 7970. Would you run your $600 card at 1.367 on air at home?

3GB vs. 2GB of VRAM is basically the least important in the comparison of 7950 OC vs. 660Ti OC. 24 ROP / 192-bit setup is probably going to limit GTX660Ti's overclocking scaling. I hope AnandTech has an overclocking section in the back to put these 2 cards to the test.

 

[Jaydip]

Amen

 

[3DVagabond]

Jeesh! I guess the Swedes didn't get the memo on pricing. TechPowerUp has a small article. Lowest price listed in Sweden (ahead of official release so this might change) $393.50USD +VAT! Maybe this card's a lot faster than it appeared to be on TweakTown? Or, are they just counting on the nVidia fans not caring about perf/$?

 

[Jaydip]

Anything above 249-259$ would be a stupid price for this card

 

[Crap Daddy]

Jeesh! I guess the Swedes didn't get the memo on pricing. TechPowerUp has a small article. Lowest price listed in Sweden (ahead of official release so this might change) $393.50USD +VAT! Maybe this card's a lot faster than it appeared to be on TweakTown? Or, are they just counting on the nVidia fans not caring about perf/$?

In Sweden everything is waaaay more expensive. It's totally worthless information regarding the suggested retail price.

 

[RussianSensation]

Possible specs for GTX660 non-Ti:

GTX 660
GK106, not GK104 (?)
960 SPs
980/1033/6008 memory
GDDR5 1.5GB
192-bit

 

[SirPauly]

That's what I originally speculated at Rage3d that the GTX 660 TI would be a GK-104 derivative and the GTX 660 would be based on the GK-106. If the GTX 660 is indeed based on the GK-104, hehe, that's a lot of sku's and price-points based on a single core -- one may say resource efficiency.

 

[toyota]

That's what I originally speculated at Rage3d that the GTX 660 TI would be a GK-104 derivative and the GTX 660 would be based on the GK-106. If the GTX 660 is indeed based on the GK-104, hehe, that's a lot of sku's and price-points based on a single core -- one may say resource efficiency.

GF114 had at least 4 official cards based on it. gtx560 ti, gtx560, gtx560 se, gtx460 v2.

 

[SirPauly]

If the GTX 660 is GK-104, may make it five, early in its cycle life, with potential price-points from 249 to 999 -- one may say resource efficiency.

 

[toyota]

If the GTX 660 is GK-104, may make it five, early in its cycle life, with potential price-points from 249 to 999 -- one may say resource efficiency.

well there was the EVGA gtx560 2win gpu so that makes 5 for the GF114 too. plus there were some OEM cards too so its really more than 5.

 

[SirPauly]

If one desires to count AIB differentiation sku's! 5 for the GF-114 --one may say resource efficiency!

 

[ocre]

Is my math bad? Whatever! It's not 160% more, and you know it. As far as your comparison to the 7950, I've already stated why I'd compare it to the 670, instead. You are ignoring that, because it doesn't fit your agenda. /end

Agenda?

strong words there sir.

I am not trying to convince anyone here to buy anything. I am not trying to push or persuade, nothing of the like. I find it really interesting you would say such things when in this very thread there seems to be this a sort pitch being thrown over and over page after page. What if i was on a mission and would just repeat buy the 7950 and overclock it to 680 levels. Whatever you do dont the buy the 660ti when you can buy a more expensive card and overclock it to the 680 levels. What if i repeated this over and over, page to page. with charts and graphs, trying to prove the 7950 was a better buy for you. You probably would have no issue if i was doing these things now would you?

Look dude, I love tech, love discussing technology and seeing things progress and unfold. I like to see the companies play their hands and i try to imagine how their hand will play. The 660ti to me is Nvidia's next move. Every bit of anything i have said in this thread has to do with how well of a hand it is. If it will play well or not. I am not trying to convince "enthusiast" that the 660ti (or the 7950) is the card they should get. I am not repeating a pitch over and over in hopes that someone might actually do as i say or do. I am not telling anyone to buy anything.

I am trying to speculate on the new hardware that is coming into the market. Whether it will do well, or not. Whether its will be successful or not. Why? because i have followed the PC since as long as i could remember. Since before i was in high school i have had a great interest in the technology.

When i try to discuss the 660ti here, its about how the card may do. How well it could perform in the market its being molded to. Its a card priced like the 7870 but can outperform the 7950. It will do extremely well from where i see this. But i keep hearing how the more expensive card in the next price bracket can overclock to 680 levels and no *should* buy this 660ti. What? then i here from you that the 670 performs better, which is another card that will cost more, in a higher bracket as well. What? of course it will, why shouldnt it?

Why would you keep throwing up the 670 anyway? why is all this overclock the 7950 to 680 levels coming up now? In discussing the real market where the 660ti will be sold, i cannot see how any of this fits in at all. How is this an indication of the 660ti,and whether it will be a successful launch or not. In the real market 95% of people buy GPUs to play games. Thats it. They buy the performance they need. They dont care about benchmarks or even know how to see their fps. If they want 680 performance, they buy the gtx680. They sure in heck dont buy a 7950 and try not max overclock it to 680 levels. Very few people do this. Very few. The majority of GPU owners just want to have it work, play their games and not have to worry about it.

But i guess your right, the 670 is more powerful. So the 660ti just sucks and you can overclock the 7950 so there is no point in nvidia even releasing the 660ti or whatever.

i serously disagree and i will bet you money right now, real money, that the 660ti will be a success launch and will make a splash. That is what i have been speculating. I will bet that no matter how much you bring up the more expensive 670, or others try to sell the more expensive 7950 overclocked to the max, the 660ti will be very successful. If a person made it there night and day obsession to convince ppl on this forum not to buy the 660ti, they might could manage to get 50 people to buy AMDs 7950 instead. They might get people to try that overclocking thing. But in the grand scheme, do you really think these more expensive cards are gonna somehow stop the success of the 660ti?

agenda????
This is what i am discussing: Whether this new GPU from nvidia will be successful or a flop. I am not asking anyone here if i should buy it or not. I am not telling anyone if they should buy it or not. I am not wondering if the 670 is more powerful, or even if its 20% more powerful. I mean my god, the 670 is supposed to be more powerful. WOW! i guess maybe the higher number means its more powerful? maybe?