why does 660ti beat 7950 so often?

[amenx]

At 1920x1080, the most common res atm, and if you dont OC your GPU much (the vast majority of buyers), 660ti is indeed a solid choice vs a 7950 and beats it in many recent games (Crysis 3, Bioshock Inf, FarCry3, etc). If you are an enthusiast who likes to OC to the hilt, the 7950 is the faster card. And if you are on a higher res (2560x1440 and above), the 7950 is the more logical choice.

For my needs and purposes, and seeing how any and all games I've tested work and run very well at highest settings (not crazy levels of AA), I would still pick the 660ti to this day.

 

[3DVagabond]

Even with no O/C'ing at 1080 and "moderate" to no AA (0 to 4x) over a large selection of games (not just 3 where one is the poster child for crappy performance anyway, FC3) the 7950 is a little bit faster than the 660ti. This is the 800MHz 7950, BTW. The The 7950 Boost trades blows with the 670, and even with moderate O/C'ing, not O/C'd to the hilt, the 7950 will beat the 660ti pretty handily.

 

[boxleitnerb]

As for the OPs question:
GCN (at least in its current implementation) is less efficient than Kepler in terms of let's call it "unit efficiency". Why that is, is a good question. I often hear it's about the front end (that part of the GPU that allocates all the work when it comes in). At higher resolutions (more pixels per triangle) this limitation becomes less and less important, thus AMD GPUs can use their raw power better than at lower resolutions. This seems to be the reason why AMD GPUs are so good with OGSSAA. With SGSSAA on the other hand (which doesn't increase resolution but sample the rendering resolution multiple times) they lose even more efficiency.

Btw I don't know why people like RS feel so compelled to lead with their usual praising of AMD and bringing up stuff that has nothing to do with the topic. The 7970 "boost" and OC was not part of the question. It was about the specs of the "normal" 7950 vs 660 Ti. Oh well...

 

[amenx]

Even with no O/C'ing at 1080 and "moderate" to no AA (0 to 4%) over a large selection of games (not just 3 where one is the poster child for crappy performance anyway, FC3) the 7950 is a little bit faster than the 660ti. This is the 850MHz 7950, BTW. The The 7950 Boost trades blows with the 670, and even with moderate O/C'ing, not O/C'd to the hilt, the 7950 will beat the 660ti pretty handily.

As is well known, you get different results from different sites. Of course the variety of games tested plays a big part in that. Most people I guess tend to focus on newer games (more interest in Crysis 3, Bioshock Inf, Tomb Raider, Sleeping Dogs, Far Cry 3, BF3 rather than Crysis 1, Metro 2033, etc). And which the OP probably tended to do as well which may be reflected in his thread title.

 

[Lepton87]

When I say barely play I mean barely manage a minimum of 60 fps in the latest releases like TR, Crysis 2 and 3, Bioshock infinite, hitman etc

So Titan is too weak just as well, I've got dips from 60fps in Bioshock Infinite. Oh, well, you need Titan SLI/maybe TRI SLI to play those games to your standards.

no single card is fast enough for that, an average of 68fps clearly shows that there are dips below 60 fps. I know that empirically. And we all know that singe card fps>M-GPU fps. 40fps will feel smoother on a single card then on multi-gpu.

Crysis 3 even worse, no single GPU is good enough for what you want, and I bet those multi-GPU set-ups also occasionally dip below 60fps with an average of 65-78fps, and then I doubt they tested the most demanding scene.

 

[RussianSensation]

As for the OPs question:
GCN (at least in its current implementation) is less efficient than Kepler in terms of let's call it "unit efficiency".

In this case unit efficiency is hardly relevant. In the context of HD7950 vs. GTX660Ti, we want to look at performance per GPU clock at stock speeds (since 660Ti is clocked higher) and their respective overclocking headroom in case this is a consideration. At stock speeds, despite lower clock speed the 7950 boost is faster. At overclocked speeds, it is also faster since not only does it scale really well with OCing but it is just 5% behind HD7970 in overall performance per clock. This is because HD7950 is a slightly cut-down HD7970 while GTX660Ti is a lot more cut down vs. the 680, both the ROP and memory bandwidth side. This makes HD7950 more of a downclocked high-end card and 660Ti more of a real mid-range card. Overclocking reveals the 7950's true potential to eclipse 7970GE/680s.

Source

Btw I don't know why people like RS feel so compelled to lead with their usual praising of AMD and bringing up stuff that has nothing to do with the topic. The 7970 "boost" and OC was not part of the question. It was about the specs of the "normal" 7950 vs 660 Ti. Oh well...

Huh? Excluding this very post, where did I mention HD7970 or how it relates to 660Ti? I said HD7950 Boost because that is the OP's card - "Sapphire HD7950 Boost". Secondly, considering you cannot buy original 800mhz 7950s, it stands to reason that the most valid comparison today is 7950 boost vs. 660Ti, for available and price reasons. Thus the discussion about a 7950 925mhz / V2 vs. 660Ti is the most relevant, which is what I addressed. HD7950 versions for sale are not 800mhz models

My other comment about HD7950 @ 1035-1050mhz ~ GTX680 is based on countless HD7950 owners and reviews online.

You can check that if you don't believe me. I posted reviews/links with overclocked 7950 performance many times against stock 7970GE/680.

To say that "GTX660Ti beats HD7950 so often" is somewhat misleading per the OP. GTX660Ti is slower than existing HD7950 versions and slower than OP's 7950 Boost:
http://www.computerbase.de/artikel/grafikkarten/2013/test-nvidia-geforce-gtx-titan/7/

With overclocking, it is no contest since HD7950 is only 5-6% behind HD7970 per clock. Overclocking HD7950 to 1150-1200mhz is like having 95% of the HD7970 @ those clocks.

Also, in the hands of enthusiasts, what makes HD7950 special is that it can surpass both stock GTX680/7970GE in performance. GTX660ti cannot do this on air even at 1300mhz OC.

Even in latest games, the 7950 V2 is still on top:
http://www.bit-tech.net/hardware/2013/04/23/bioshock-infinite-performance/4

If the OP took 15 min of his time to overclock the 7950, then it would handily outperform the 660Ti OC anyway. Interesting how you felt compelled to skip over all the facts and find bias in my post as per usual. Nowhere in my post did I praise AMD. I simply laid out the facts that HD7950 boost/V2 is the faster card overall in stock and overclocked forms. So the OP's assertion that 660Ti beats HD7950 so often has little merit. Where is the evidence?

Even in the latest games HD7950 boost is actually closer to GTX670 than to a 660Ti; and that's a stock 7950 V2, which is like running a stock Intel K processor.

 

[smeg1161]

The 7950 is ridiculously under clocked @ 800Mhz, AMD did that to keep it some distance from the 7970 which only has about 250 SP's more, Clock for clock there is not much difference between them.

You also need to keep in mind that Anand quite obviously do not up date their GPU charts, http://www.anandtech.com/bench/Product/550?vs=647 look at the most well known game there, BF3. according to that the GTX 660TI is 20% faster than the 7950.

AMD have made vast driver improvements since then, if you look around other reviewers, who have updated their charts to reflect later drivers you can see how out of sync Anand's charts are.

So some things you need to know, the 7950 is clocked about 30% lower than it should be, you can fix that in seconds by adjusting a slider in CCC or MSI AB...
Base clock 7950's do 50% overclocks day in day out, some will do even more.

Also, AMD's GPU's are designed with game GPU compute in mind, while Nvidia strip those sort of things out to save money and reduce power consumption.

As a result when a game uses GPU compute to render advanced lighting, DOF, StressFX and the like Nvidia grind to a halt, they can't handle it, as demonstrated in BIOShock... TombRaider... http://www.techspot.com/review/645-tomb-raider-performance/page4.html

Look at that, 7970 GE faster than Titan, 7870 = GTX 680......

Call me a fanboy if you like, i don't even think Kepler is a proper GPU, its just a cost cut image rendering device, throw any real work at it and it chokes.
Whats more its holding game progress back, Nvidia have always involved them selves in game development which is why they had looked good in the past.
But not any more, AMD have finally pushed the GCN arch out there and developers love it. now games are free to progress!

 

[boxleitnerb]

Ok, but in the context of HD7950 vs. GTX660Ti, we want to look at performance per clock. This is because HD7950 is a slightly cut-down HD7970 while GTX660Ti is a lot more cut down. Since HD7950 boost > GTX660Ti, 7950 OC is faster 660TI OC.

No we don't. Why? Because he asked about specs in general, not clock speed. It's also not about what card is more cut down. It's about the specs themselves.

Huh? Where in my entire post did I mention HD7970? I said HD7950 Boost. The OP's card is "Sapphire HD7950 Boost", so obviously the discussion is about a 7950 925mhz / V2, which is what I addressed.

My bad, typo. I meant 7950 Boost. The OP wrote just 7950. But even if he had meant the 7950 Boost, that changes little since the specs of that card are even higher compared to the 660 Ti. We don't need to nitpick about a few percent here and there - the 7950 (Boost) doesn't perform as it should given its specs. The question is why.

If the OP took 15 min of his time to overclock the 7950, then it would handily outperform the 660Ti OC anyway. Interesting how you felt compelled to skip over all the facts and find bias in my post as per usual. Nowhere in my post did I praise AMD. I simply laid out the fact that HD7950 boost/V2 is the faster card overall in stock and overclocked forms. So the OP's assertion that 660Ti beats HD7950 so often has little merit.

Those may be facts but they are irrelevant here since OC changes the specs obviously (fillrate, compute power, bandwidth etc.). And of course it's bias since you seemingly cannot stick to the question at hand but instead compulsively bring up other SKUs and OC in almost every thread. If it's not OC, it's bitcoining. Let's stay on topic please.

 

[RussianSensation]

There is no point in discussing specs because usually you cannot compare specs of AMD and NV cards directly because it assumes the architectures are equally efficient. Once you do enough research and realize that it is a waste of time, you arrive at performance per GPU clock speed as the metric you may want to look it, especially for overclockers. This has been discussed 100x on the forums and proven countless times with examples like HD4890 vs. HD5770 or GTX285 vs. GTX470 or HD7970 vs. HD6970. Cross-comparison of AMD vs. NV GPUs is usually a waste of time. It can even lead to erroneous results when comparing different GPU architectures from the same brand. 925mhz 32 ROP 7970 vs. 880mhz 6970. On paper the former is barely faster in pixel fill-rate. In the real world, 7970 puts down 51% higher pixel fill rate performance despite both cards having 32 ROPs and 7970 being clocked just 5% higher (925mhz vs. 880mhz). If you cannot directly compare 2 different AMD architectures without looking at specialized benchmarks for each sub-components (ROPs, TMUs, etc.), then you definitely cannot do so with AMD vs. NV cards. You also cannot compare memory bandwidth of AMD and NV cards. If 660Ti has 144GB/sec and HD7950 has 240GB/sec, what does that tell you about their real world gaming performance? Absolutely 0.

The answer is not about specs but performance per GPU clock since that's everything we need to know. At stock speeds GTX660Ti performs slower than a stock 7950 Boost. I don't know why this is news to you since it's been this way since at least last summer and hasn't changed. If you really want to talk about specs, it's a lot more useful to compare GTX660Ti to 670/680 to see how cut down it is and then compare 7950 and 7970 to see how cut down it is. In the former case, the card has 50% less ROPs and 25% less memory bandwidth, or a significantly cut down 680. 7950 is a lot closer to 7970 with the entire 384-bit bus and full 32 ROPs. That makes 7950 a downclocked high-end card in disguise and 660Ti a fast mid-range card. 7950 is severely pixel fill-rate limited which is why to make it run fast, you have to overclock the GPU and 'unlock' the ROP bottleneck. Once you lift this bottleneck, 7950 flies past 7970GE/680. 660Ti is cut down in too many places.

 

[futurefields]

My clock is currently OC at 1050/1250

here's an example of why I make this thread:

http://www.guru3d.com/articles_pages/bioshock_infinite_graphics_performance_review_benchmark,6.html

660ti handily beats the 7950 Boost in Bioshock: Infinite even at 2560x1600 ultra settings, where the 7950 Boost spec wise should be much better at these settings.

 

[futurefields]

660ti beating 7950 GHZ at Crysis 3 even at 2560x1600 Very High Settings:

http://www.guru3d.com/articles_pages/crysis_3_graphics_performance_review_benchmark,7.html

Looks like either Nvidia simply has the better card, or Guru3d has some seriously flawed tests.

 

[futurefields]

Far Cry 3 at 1920x1200

http://www.guru3d.com/articles_pages/far_cry_3_graphics_performance_review_benchmark,6.html

660ti wins again over 7950 Boost. 7950 only comes ahead at 2560x1600, but there using a single card gives abysmally low framerates anyways.

So is the 660ti a better card when running single gpu @ 1080p? Definitely looks like it to me.

 

[bystander36]

Far Cry 3 at 1920x1200

http://www.guru3d.com/articles_pages/far_cry_3_graphics_performance_review_benchmark,6.html

660ti wins again over 7950 Boost. 7950 only comes ahead at 2560x1600, but there using a single card gives abysmally low framerates anyways.

So is the 660ti a better card when running single gpu @ 1080p? Definitely looks like it to me.

To be fair, it all depends on the games tested. There are others which go the other way too. There are 3 other games you could look at which would go the other way.

 

[futurefields]

Bioshock: Infinite, Crysis 3, and Far Cry 3 are probably the 3 most popular graphics heavy games currently.

 

[bystander36]

Bioshock: Infinite, Crysis 3, and Far Cry 3 are probably the 3 most popular graphics heavy games currently.

I have noticed a trend with Nvidia and AMD. Nvidia seems to have closer to full performance at release compared to AMD (barring any bugs). As time goes on, AMD will tweak their profiles, and often catch up and or surpasses Nvidia.

Then again, I'm only going off the review sites game suites and just what has stuck in my mind.

There are several possible reasons that pop in to mind.

 

[futurefields]

I've noticed that as well, and I've also noticed they only do that for the most recent line of cards ie. AMD improves performance with 7000 cards on new drivers, 6000 cards don't get the improvements. So if you want the AMD "tweaked driver profiles" you have to continually buy the latest series of card, because new optimizations stop for the old line.

 

[smeg1161]

There have been massive discrepancies between reviewers in GPU performance over these last several AAA titles.

What everyone needs to keep in mind is that most reviewers depend on revenue from hardware makers to exist, AMD and Nvidia will use them as advertising arms.

pcper for example are accused of working for Nvidia and base their reviews on what they tell them.

There are also some accusations laid at Nvidia for turning off or tone down things that depend on GPU compute with driver updates.

Toms Hardware have in the past been accused of working with AMD.

Whatever the truth or not, at the end of the day one is responsible for making ones own decisions.

And when i look at the architectural differences between the two, i am decided on which one is better.

 

[Fastx]

Far Cry 3 at 1920x1200

http://www.guru3d.com/articles_pages/far_cry_3_graphics_performance_review_benchmark,6.html

660ti wins again over 7950 Boost. 7950 only comes ahead at 2560x1600, but there using a single card gives abysmally low framerates anyways.

So is the 660ti a better card when running single gpu @ 1080p? Definitely looks like it to me.

Fwiw you are looking at a graph with older drivers from 12-4-12 (13.1) on FC-3. Compare your 12-4-12 chart at 2560x1600 to the test done on 3-4-13 chart graph at 2560x1600. You will see an improvement (4) with slightly newer drivers (13.1) with the 7950 . Edit ok here is the chart also for 1920x1200 on the 7950 boost.

Here is a test with slightly newer drivers 3-4-13 (13.1) on FC3 @2560 x1600

http://www.guru3d.com/articles_pages/his_radeon_7950_x_iceq_review,10.html

•Very High-quality DX11 mode
•2x MSAA (Anti-aliasing)
•16x AF (Anisotropic Filtering).

 

[bystander36]

There have been massive discrepancies between reviewers in GPU performance over these last several AAA titles.

What everyone needs to keep in mind is that most reviewers depend on revenue from hardware makers to exist, AMD and Nvidia will use them as advertising arms.

pcper for example are accused of working for Nvidia and base their reviews on what they tell them.

There are also some accusations laid at Nvidia for turning off or tone down things that depend on GPU compute with driver updates.

Toms Hardware have in the past been accused of working with AMD.

Whatever the truth or not, at the end of the day one is responsible for making ones own decisions.

And when i look at the architectural differences between the two, i am decided on which one is better.

Most of them still try to be unbiased. They may sometimes soften the blow in their words, but Pcper has all kinds of AMD advertisement, so they aren't reliant on Nvidia.

I have noticed that Tom's Hardware tries to be unbiased, but a lot of their viewers are always on their case when ever they don't show AMD in the best possible light, yet their review on the 7990 still was a big negative. They try to be unbiased, and try to deliver the news as to not piss off all their AMD fanboys (not that they are all AMD fanboys).

A lot of times sites start to look biased because the evidence shows an advantage to one side or another. Are they supposed to give equal reviews despite the evidence? What I always look for is an even distribution of brand favorable games, or at least ones that aren't specifically trying to cherry pick ones to favor a particular brand.

 

[boxleitnerb]

There is no point in discussing specs because usually you cannot compare specs of AMD and NV cards directly because it assumes the architectures are equally efficient. Once you do enough research and realize that it is a waste of time, you arrive at performance per GPU clock speed as the metric you may want to look it, especially for overclockers. This has been discussed 100x on the forums and proven countless times with examples like HD4890 vs. HD5770 or GTX285 vs. GTX470 or HD7970 vs. HD6970. Cross-comparison of AMD vs. NV GPUs is usually a waste of time. It can even lead to erroneous results when comparing different GPU architectures from the same brand. 925mhz 32 ROP 7970 vs. 880mhz 6970. On paper the former is barely faster in pixel fill-rate. In the real world, 7970 puts down 51% higher pixel fill rate performance despite both cards having 32 ROPs and 7970 being clocked just 5% higher (925mhz vs. 880mhz). If you cannot directly compare 2 different AMD architectures without looking at specialized benchmarks for each sub-components (ROPs, TMUs, etc.), then you definitely cannot do so with AMD vs. NV cards. You also cannot compare memory bandwidth of AMD and NV cards. If 660Ti has 144GB/sec and HD7950 has 240GB/sec, what does that tell you about their real world gaming performance? Absolutely 0.

The answer is not about specs but performance per GPU clock since that's everything we need to know. At stock speeds GTX660Ti performs slower than a stock 7950 Boost. I don't know why this is news to you since it's been this way since at least last summer and hasn't changed. If you really want to talk about specs, it's a lot more useful to compare GTX660Ti to 670/680 to see how cut down it is and then compare 7950 and 7970 to see how cut down it is. In the former case, the card has 50% less ROPs and 25% less memory bandwidth, or a significantly cut down 680. 7950 is a lot closer to 7970 with the entire 384-bit bus and full 32 ROPs. That makes 7950 a downclocked high-end card in disguise and 660Ti a fast mid-range card. 7950 is severely pixel fill-rate limited which is why to make it run fast, you have to overclock the GPU and 'unlock' the ROP bottleneck. Once you lift this bottleneck, 7950 flies past 7970GE/680. 660Ti is cut down in too many places.

So you are not interested or capable in answering the question of the OP? Why did you enter the thread in the first place, then? He posted an interesting question that could have lead to a interesting technical discussion. Instead we get the same old "AMD OCs great" stuff from you that doesn't help the OP at all.

Btw you can measure fillrates. Nobody looks at the number of ROPs and TMUs alone since as you've shown they can be differently efficient. Go to hardware.fr for this, they do this with every new gen.
And you still have yet to conclusively prove that pixel fillrate is any kind of serious limit in todays cards. Just picking one or two cases where it works, doesn't count.
Finally (again!): The OP asks about the 7950, not the 7950 Boost (even though he has one). Don't just make stuff up, but instead maybe ask the OP what card he really means.

 

[wand3r3r]

So you are not interested or capable in answering the question of the OP? Why did you enter the thread in the first place, then? He posted an interesting question that could have lead to a interesting technical discussion. Instead we get the same old "AMD OCs great" stuff from you that doesn't help the OP at all.

What exactly are you waiting to hear?

The OP picked the 3 games that he thought NV was faster at, but coincidentally was shown that 1 of them happened to have outdated drivers and the 7950B was indeed considerably faster. The other 2 run better on NV period.

Thread title could/should say (yet lose so often and overall) as it's completely misleading.

According to thread title, everything has been relevant. If OP still wants to wonder why NV wins in 2 NV leaning games, at least put it in the title.

What's next a thread on why the 7950B wins the 660 ti so often, or why not 670 or even 690 if there is a game or two it wins (but not overall)?

Considering that the 7950B is the faster card, has amazing OC potential, and is clearly not slower than the 660 TI, why not discuss the 7950B and show the facts in the thread with a misleading title as it is?

 

[VulgarDisplay]

I've noticed that as well, and I've also noticed they only do that for the most recent line of cards ie. AMD improves performance with 7000 cards on new drivers, 6000 cards don't get the improvements. So if you want the AMD "tweaked driver profiles" you have to continually buy the latest series of card, because new optimizations stop for the old line.

Uh, the same thing happens with Nvidia cards?

The main reason why the GCN driver improvements don't have any effect on older cards is because it is a brand new architecture.

 

[boxleitnerb]

What exactly are you waiting to hear?

The OP picked the 3 games that he thought NV was faster at, but coincidentally was shown that 1 of them happened to have outdated drivers and the 7950B was indeed considerably faster. The other 2 run better on NV period.

Thread title could/should say (yet lose so often and overall) as it's completely misleading.

According to thread title, everything has been relevant. If OP still wants to wonder why NV wins in 2 NV leaning games, at least put it in the title.

What's next a thread on why the 7950B wins the 660 ti so often, or why not 670 or even 690 if there is a game or two it wins (but not overall)?

Considering that the 7950B is the faster card, has amazing OC potential, and is clearly not slower than the 660 TI, why not discuss the 7950B and show the facts in the thread with a misleading title as it is?

And yet he specifically asks about specs and if Nvidia shaders are more powerful than AMD shaders. But this part of his question is completely ignored, in favor of the usual idiocy. This thing I just cannot understand.

 

[wand3r3r]

And yet he specifically asks about specs and if Nvidia shaders are more powerful than AMD shaders. But this part of his question is completely ignored, in favor of the usual idiocy. This thing I just cannot understand.

Perhaps if the thread title wasn't so misleading and had the relevant question it would gather attention to the actual question hidden in the thread (title says one thing, OP another). I scanned the thread the other day and then didn't remember that little detail.

Edit. After rereading the first two posts more the first answer points out you can't compare them directly.

To say it's shaders are more powerful is hard to prove, and most likely wrong unless you actually cherry pick games.

I'd simply guess that the certain game engines are optimized for certain architectures or at least certain logic which tends to align closer to one or the other.

 

[boxleitnerb]

Perhaps if the thread title wasn't so misleading and had the relevant question it would gather attention to the actual question hidden in the thread (title says one thing, OP another). I scanned the thread the other day and then didn't remember that little detail.

Edit. After rereading the first two posts more the first answer points out you can't compare them directly.

To say it's shaders are more powerful is hard to prove, and most likely wrong unless you actually cherry pick games.

I'd simply guess that the certain game engines are optimized for certain architectures or at least certain logic which tends to align closer to one or the other.

Of course you cannot compare the specs directly. That's the interesting question: Why?
I wouldn't put it on the shaders alone. Nvidia seems to utilize the raw power of their GPUs better than AMD when you look at average numbers (ratings across multiple games). What exactly the secret is, I cannot say. I can just speculate. With lots of MSAA or OGSSAA/higher resolution, AMD does a better job in tapping out their raw power (bandwidth, GFLOPs).