[Hardcorp] GALAXY GTX 660 Ti GC OC vs. OC GTX 670 & HD 7950

[notty22]

Hunting for optimum performance from a single review sample does not mean anything towards real world retail results. In the field, where AMD is pumping 1.25V in to 7950's to offer higher turbo results is considered eligible for AVG results. Not low leakage single chips characteristics.

 

[Skurge]

Hunting for optimum performance from a single review sample does not mean anything towards real world retail results. In the field, where AMD is pumping 1.25V in to 7950's to offer higher turbo results is considered eligible for AVG results. Not low leakage single chips characteristics.

And yet TPU show the The 7950 more efficient than all but 1 of the kepler parts. Top 3 Being All GCN parts. This overall includes BF3.

I wonder how many people there are that have reference 7950s and flashed these BIOS? I don't know a single person on this forum.

 

[RussianSensation]

Hunting for optimum performance from a single review sample does not mean anything towards real world retail results. In the field, where AMD is pumping 1.25V in to 7950's to offer higher turbo results is considered eligible for AVG results. Not low leakage single chips characteristics.

Not even close. Only 1 out of 19 7950s on Newegg is a reference HD7950 B card. So how is that an average? An average after-market 7950 has been shown to use around 1.05-1.08V. Don't spread false information please. We have a thread of many MSI TF3 users who will assure you that your average is way off. The reference 7950 1.25V B card AT tested is the worst case scenario. Not a single after-market 7950 will be like that, because all those Vapor-X, TF3, Gigabyte Windforce 3x designs use binned 7950s and 7970 PCB. So no, your average assessment is misleading.

Look, I am not going to beat around the bush. I favour price/performance over performance/watt. This is why I got 470s when the 5870 launched and overclocked them to 760mhz which made them faster and cheaper than a 5870. So I am biased against price/performance. Since 7950 OC goes head-to-head against a $380-400 GTX670 after-market versions, I would still take it over the 660Ti.

Tanking with MSAA, high resolution textures/Skyrim mods from an overclocked 660Ti is unacceptable to me when the cost is just 40-50W of extra power. I'll take the higher power and 15-20 fps more and a safe peace of mind that 7950 will do just alright since it still has 32 ROP / 384-bit bus from the 7970. GTX660Ti is completely outclassed when it's pushed hard. Unless every game coming out is WOW, Crysis 3, or Borderlands 2 with PhysX, I am putting my money that HD7950 OC will continue to be as fast as a GTX670 OC while there is a certain degree of risk that 144GB/sec bandwidth and 24 ROPs may not fair so well down the line. These results are not what one would call reassuring for the $300 660Ti. $18 more and you get a faster out of the box 950mhz 7950 with an awesome cooler, or save $60 and you lose just 10% of the performance.

 

[notty22]

What AMD ships to reviewers for a bios for testing/reviews is not misleading. Repeating nonsense about what you consider real world information based upon unconfirmed stable or not stable overclocks is misleading and ridiculous.

 

[Skurge]

What AMD ships to reviewers for a bios for testing/reviews is not misleading. Repeating nonsense about what you consider real world information based upon unconfirmed stable or not stable overclocks is misleading and ridiculous.

So you prefer we talk about a reference card you can't even buy? I asked again. How many people are there that have Reference 7950s with the new bios?

 

[zaydq]

What AMD ships to reviewers for a bios for testing/reviews is not misleading. Repeating nonsense about what you consider real world information based upon unconfirmed stable or not stable overclocks is misleading and ridiculous.

You're really trying to fight for Nvidia's life. I bet if this was the other way around you wouldn't be so negatively vocal. Facts have been posted in tons of threads. That bios is shit, it doesn't reflect what the consumer is actually buying, or would want to buy.

Its funny how if it was AMD on top this gen, then everyone would most like throw out the tdp argument and would vouch for insane overclocking for Nvidia.

Let's face facts if you're an enthusiast buying 300$ cards and you're crying about energy efficiency, then buy a Hewlett-Packard. Ferrari owners don't cry about fuel prices.

Overclocking should always be a factor. Whether the consumer wants to or not, that's up to them, and they will therefore form opinions on the stock clocks.

This argument has been exhausted with flaming, factual posts, graphs, quotes, reviews and data, and yet people are blind to all of it and simply cry "Geeefarce"

 

[RussianSensation]

What AMD ships to reviewers for a bios for testing/reviews is not misleading. Repeating nonsense about what you consider real world information based upon unconfirmed stable or not stable overclocks is misleading and ridiculous.

I disagree. You implying that 1.25V is the average for 7950 card is a number you pulled out of thin air to justify your viewpoint that 7950 is hot, loud and power hungry. You conveniently ignored all these cards that have quiet coolers, which do not run hot and none of them comes with 1.25V bios.

1) Gigabyte Windforce 3x - $318
2) Sapphire 950mhz Dual-X 7950 - $318
3) PowerColor PCS+ 7950 = $338
4) HIS IceQ 7950 = $330
5) MSI TF3 880mhz 7950 = $318
6) Sapphire Vapor-X 950mhz 7950 = $338

If you read the reviews, most of these cards overclock to 1100-1150mhz on 1.175V or below (*except the Vapor-X that comes with 1.212V I believe). While the Vapor-X comes with higher voltages, the cooling system on it can keep a 1200mhz 7970 TOXIC below 70*C. That card is built like a tank, with 9 premium black diamond chokes from the $700 TOXIC that guarantee no coil whine. Please let me know when I can buy a GTX660Ti with such premium quality parts and a cooler to match.

You keep downplaying overclocking on the 7950 and implying that 1.25V is the real average that would push the power consumption of the 7950 into some stratosphere level. Yet, I heard none of these things from you when GTX460 @ 850-925mhz dominated AMD from a price/performance viewpoint. That's right, GTX460 became one of the best cards because of its overclocking despite the power consumption for such a card rising exponentially in the process and as usual overclocking not being guaranteed.

Now that 7950 completely mops the floor with the 660Ti, all of that is mysteriously forgotten.

vs. just a 1.03ghz 7950 already comes close to a stock 680 for $180 less! And this is in one of the best games for GK104.

I see so when GTX460 delivered HD5870 level of performance at OCed speeds for less $, performance/watt was disregarded. Yet, when an 1150mhz HD7950 passes GTX680/HD7970GE and provides GTX670 OC performance for $60-80 less, suddenly performance/watt is THE most important metric, above performance, price/performance, compute features, bitcoin mining, 3GB of VRAM for mods, above everything really. You seem to be focusing on performance/watt so much actually, that I am amazed you even bought GTX460s yourself, and didn't have any problems with GTX470/480/570/580 cards from a performance/watt metric when they got dominated by 5850 OC/5870/6950 2GB in very same metric you deem as suddenly the most important. Back then performance was more important than performance/watt? So why aren't you touting 7950 OC's performance advantage over the 660Ti's OC in light of all these historical factors? :hmm:

My story is I recommend cards on price/performance and consider OCing as a bonus that can often save gamers $. Your story seems to be changing between top performance (Fermi, bonus more VRAM) to performance/watt (GK104) to overclocking should be a factor (GTX460 @ 875mhz SLI in your own system) depending on which of these puts NV in the lead.

This round 7950 OC provides better price/performance and overall performance than a 660Ti OC and 7870 provides better price/performance, but you are saying performance/watt should now be the deciding factor and overclocking is no longer a factor? Which is it?

You did buy GTX460 over the 5850, right? I see....a GTX460 @ 900mhz uses > 200W of power; and you have 2 of those! You've been criticizing the entire HD7950/7970 generation's performance/watt non-stop and your own card sucks power that is equivalent to a 1150-1200mhz HD7970!!!!!!!!!!!!!!!

vs.

From where I am looking an HD7970 @ 1200mhz beats a GTX590 and uses maybe 20W more power than 1 of your cards. Correct me if I am wrong but that would put HD7970 @ 1.15-1.2ghz at nearly twice the performance/watt of your GTX460 @ 875mhz SLI setup. The GTX680 is what 25% more efficient but costs $100 more than a good after-market 7970?

=======================

You still have not addressed why someone should spend $60-80 more for a GTX660Ti over the 7870. 10% more performance for 25-35% price increase.

 

[Final8ty]

Compute. HD7970 GE delivers 1.08Tflops of theoretical double precision compute performance vs. 0.16 Tflops for the 7870. NV is charging $3,000+ to give you 1Tflops of compute in K20 Tesla card. AMD is selling you that for $430-450.....That stuff isn't free; it costs transistor space and die size (just compare HD6970 vs. 6870 or 7970 to 7870 die sizes vs. their respective functional units).

Pitcairn = 32 ROP / 1280 SPs / 80 TMUs in 212 mm^2 die OR = 6.6 mm^2 per ROP, 0.16 mm^2 per SP, 2.65 mm^2 per 1 TMU
Tahiti XT = 32 ROP / 2048 SP / 128 TMUs in 365 mm^2 die OR = 11.4 mm^2 per ROP, 0.18 mm^2 per SP, 2.85 mm^2 per 1 TMU

It's clear there is some massive transistor inefficiency in Tahiti XT for actual games. This is because if you read the strategy behind GCN and Heterogenous System Architecture, AMD is counting on the fact that GPUs will perform or help CPUs accelerate programs on the desktop that normally would only be CPU-centric. It's a risk they are taking. Their CTO just published a presentation on this strategy called 'Surround Computing'. AMD is betting big that compute will become a huge factor for putting GPU at the forefront of making it the most vital component in a PC. It's a chicken and an egg scenario. Without putting compute functionality into a GPU to make it general-purpose, software will never come out to take advantage of GPUs beyond games. Thus far, Tahiti XT is paying a penalty for this since few programs on the desktop are taking advantage of the added compute, but there are some already. Thus far, in the 3 games that actually use DirectCompute, Kepler is way behind Tahiti. Whether or not this will pay off long-term for AMD is impossible to say right now without seeing more trends. Still this could be AMD's own "Tessellation moment" so to speak. NV appears to be behind in general compute and if more games start to use compute shaders for lighting, shadows and so on, AMD will have a 1 generation head start; or it could be end up a huge waste of $ and total industry adoption failure like PhysX is. The difference is the entire industry is shifting towards GPU compute while PhysX is a closed proprietary format. So I think AMD's risk is more justified.

AMD also admitted that Pitcairn was designed by a different team and it was a more compact/efficient design in the first place. They intend to use the lessons learned from that design for their future generations.

Here is the compute difference between 7870 and 7970.

If AMD just stripped out all that "fat" and made a 2048 SP Pitcairn chip, it would surpass a GTX670 in performance and performance/watt. I expect we will see a very lean and mean gaming chip that surpasses the GTX670 when HD8870 launches. In that regard, I agree with blastingcap that GCN by itself is very efficient, but the HD7970 "GCN SKU" is heavily penalized for being a general purpose + gaming chip vs. a pure gaming chip that is GK104. It'll be interesting to see if NV ends up segregating their product line into gaming vs. compute / general purpose chips as they have done this generation.

http://www.engadget.com/2012/08/31/samsung-amd-hsa-foundation/
http://blogs.amd.com/fusion/2012/08/31/expanding-the-hsa-foundation/

 

[brandon888]

RussianSensation


after all analysis ..... i think that 660 ti is very overpriced ;/ im not AMD fan but 7950 is at same price and it's faster ... has better ram and bit interface to .... so better get 670 or 7950 .... depends on what a side you are and what you prefer ....


660 TI has tooooooooooooo low pixel filtering ...

nvidia should cut cores not memory ..... imho 1152 cuda cores with 256 bit interface and 6008 memory clock could be more balanced

 

[SirPauly]

They already did this round. The compute parts are going to workstations and professional markets and GK104/106 parts are pure gaming parts. This may change once GK110 launches.

That's not true at all. The GK-104 are in workstation/Quadro and GPU compute markets/Tesla.

 

[RussianSensation]

Sorry, I should have made myself clearer. While GK104 is in the K10 card for professional use, NV has stripped GK104 of compute functionality that was present in the GTX470/480/570/580 cards. NV continues to run well in Folding@Home simply because that program doesn't use compute, but relies on CUDA cores. This explains why GTX680 is still fast compared to the 580 in F@H. It is also true that not all programs use double precision performance, which is where K10 comes into play.

When compute is discussed from AMD's point of view, it's not only DP performance, but an umbrella that encompasses the GPU's ability to perform different things and not rely so much on how a code is fed to the front-end of the GPU (which was critical in utilizing all the units in VLIW-4/5). Right now from a compute performance, Tahiti is in another world compared to GK104. Although, I have no doubt that NV's GK110 is a beast as well.

Either way you slice it, AMD is betting that DirectCompute shaders, OpenCL and general purpose compute will become important as the GPU moves even closer towards being a general-purpose computing device, and not just a device which is good for games. As I said AMD may be wrong on this but if they are right, NV will be playing catch up, much like when NV realized tessellation was the next big thing in DX11 and took full advantage of their head-start.

You also have to remember that AMD is lagging behind in the CPU space and they can use GCN or future AMD GPU architectures that are even more focused on compute to work together with the CPU to accelerate certain tasks. When the entire industry is moving towards GPGPU and compute, you can't fight it or you'll be left behind. NV surely knows this. Right now we seem to be in the infancy stages of this. Both NV and AMD have their video decoding engines and it seems the next step is to actually start using the GPU as a whole to support CPU-based applications that normally would never have run on the GPU.

To further cement this view, we should see virtual memory on the GPU as the next step forward in the evolution of the GPU. Virtual memory space will allow CPUs and GPUs to use the "unified" virtual memory and pre-emption to further enhance the ability of GPU to autonomously process the data without the help of CPU (Source). That will move the GPU even further outside of the gaming realm and closer towards a general-purpose device. NV's Maxwell is rumoured to incorporate these features which already tells us that NV is also slowly moving in that direction (and of course since they basically "invented" GPGPU industry with G80 in the first place and spent millions of dollars promoting it).

If we see a full-blown GK110 next year, it'll be almost a confirmation that NV delayed it because of costs, yields, profitability and manufacturing capability/wafer capacity reasons. If GK110 is in the cards, then the GK104 mid-range Kepler was just stop-gap that was necessary before the real compute and gaming flagship GK110 launches. That's the interesting part -- will we see NV bring back the full compute that GF100/110 had back to the consumer space, or is NV going to focus on performance/watt and keep consumer parts compute-crippled but focused on gaming?

 

[SirPauly]

If we see a full-blown GK110 next year, it'll be almost a confirmation that NV delayed it because of costs, yields, profitability and manufacturing capability/wafer capacity reasons.

Indeed, and believe this is the reason.

 

[RussianSensation]

AMD is getting aggressive with 7950 prices in Canada. Newegg.ca has MSI TwinFrozr III 7950 for $276. The cheapest after-market GTX670 on Newegg.ca is $410 CDN. In Ontario we pay a 13% tax on the total price which makes the price delta between these 2 cards even higher. NV is getting owned so hard in price/performance right now, it's unbelievable they are selling anything at all in US/Canada. Every week the deals on AMD cards are delivering unbeatable price/performance it seems. Seriously guys, you have to admit, NV's cards are slowly moving off the map into crazy overpriced territory at these price levels. Why would anyone pay $130 extra for a GTX670 when OC vs. OC they perform about the same?! At these prices the 660Ti is a total non-starter.

 

[chimaxi83]

AMD is getting aggressive with 7950 prices in Canada. Newegg.ca has MSI TwinFrozr III 7950 for $276. The cheapest after-market GTX670 is $410 CDN. NV is getting owned so hard in price/performance right now, it's unbelievable they are selling anything at all in US/Canada. Every week the deals on AMD cards are delivering unbeatable price/performance it seems. At prices levels 660Ti is a total non-starter.

Wow! I bought a TFIII 6950 last summer for the same price lol. Amazing deal.

 

[AnandThenMan]

AMD is getting aggressive with 7950 prices in Canada. Newegg.ca has MSI TwinFrozr III 7950 for $276. The cheapest after-market GTX670 on Newegg.ca is $410 CDN.

Keep in mind that [H] considers the GTX660ti the direct competitor to the 7950, not the GTX670. The 660 is about $300 on Newegg.ca.

 

[RussianSensation]

Keep in mind that [H] considers the GTX660ti the direct competitor to the 7950, not the GTX670. The 660 is about $300 on Newegg.ca.

The same website that has 2 GTX660Ti advertisements on the front page and a GTX660Ti Asus giveaway? Isn't that 3 GTX660Ti advertisements right there? Interestingly enough, one of those ads is for the Galaxy GTX660Ti GC card they reviewed against the faster 7950. Just sayin'......

 

[AnandThenMan]

That screen grab makes it look more like [NV]OCP. :\ I still like their review methods, but I feel the praise the 660Ti is getting to be especially unjustified.

 

[toyota]

The same website that has 2 GTX660Ti advertisements on the front page and a GTX660Ti Asus giveaway? Isn't that 3 GTX660Ti advertisements right there? Interestingly enough, one of those ads is for the Galaxy GTX660Ti GC card they reviewed against the faster 7950. Just sayin'......

That screen grab makes it look more like [NV]OCP. :\ I still like their review methods, but I feel the praise the 660Ti is getting to be especially unjustified.

hard is NOT biased. they go back and forth all the time. they quite often praise cards that most sites don't. they also have some odd conclusions such as when they claimed the 7770 had the same playability as the 6870. that made no sense because even their factory oced 7770 would not even beat the 6850. hard is just plain weird at times.

 

[RussianSensation]

Not saying they are biased but both HD7770 and GTX660Ti reviews are not their best pieces of work, let's put that mildly.

 

[toyota]

Not saying they are biased but both HD7770 and GTX660Ti reviews are not their best pieces of work, let's put that mildly.

or 5830 or 8600gt. really they have tons of odd reviews. they seem to praise almost any new card at release.

 

[Final8ty]

AMD is getting aggressive with 7950 prices in Canada. Newegg.ca has MSI TwinFrozr III 7950 for $276. The cheapest after-market GTX670 on Newegg.ca is $410 CDN. In Ontario we pay a 13% tax on the total price which makes the price delta between these 2 cards even higher. NV is getting owned so hard in price/performance right now, it's unbelievable they are selling anything at all in US/Canada. Every week the deals on AMD cards are delivering unbeatable price/performance it seems. Seriously guys, you have to admit, NV's cards are slowly moving off the map into crazy overpriced territory at these price levels. Why would anyone pay $130 extra for a GTX670 when OC vs. OC they perform about the same?! At these prices the 660Ti is a total non-starter.

AMD have had the right effect!

Today over 100pc of 7950 sold and over 250pc of 7800 series in a day - WOW!

How many 660Ti's?

Just 29pc today, so quite a bit lower, but not so bad if they sustain that daily.

Still at £30 or so less they'd fly!

http://forums.overclockers.co.uk/showthread.php?p=22624449#post22624449

 

[3DVagabond]

Tahiti is not as well balanced as Pitcairn or GK104. Tahiti is starved for front end resources which affects perf scaling as you add more shaders . the perf scaling from pitcairn to tahiti in most games is around 25 - 30%. It definitely should be better. Only in games where bandwidth is a major factor like Metro 2033 Tahiti comes close to 50% faster than Pitcairn. In those games Pitcairn is actually bandwidth constrained as it does not run at 6 Ghz like GK104. AMD needs to improve the front end in HD 8970 and get better perf/ shader . If they can extract another 15% out of the same 2048 shaders , then if HD 8970 comes with 2560 shaders you can expect a decent 30 - 35% improvement. I also expect AMD to improve shader efficiency with enhanced GCN.
As for GK110 it has a significant amount of resources dedicated to FP64 and features which help quadro and tesla. So its efficiency will not be as good as GK104. so HD 8970 vs GTX 780 will give a clear idea as to who has the better efficiency when an all round compute + gaming chip is considered.

+1 :thumbsup: We are comparing apples to oranges with Tahiti to GK104.

Of course there's nothing being said, so this is just a seat of my pants guess, but I believe that the biggest issue with GK100 was power usage. AMD is no slouch when it comes to designing efficient GPU's. Just look at Pitcairn for example. Considering the power usage for Tahiti and how big GK100 was, I don't believe they could make it work. At least not this early on in 28nm's life.

 

[Arkadrel]

http://forums.overclockers.co.uk/showthread.php?p=22624449#post22624449

Lol, it shows people notice price, and price/performance.
Selling more than 3 times as many 7950's as 660ti's.

 

[3DVagabond]

Lol, it shows people notice price, and price/performance.
Selling more than 3 times as many 7950's as 660ti's.

Now we know that can't be true! nVidia sells millions of cards while the AMD cards just sit there collecting dust. Steam says so! /sarc

 

[Skurge]

The same website that has 2 GTX660Ti advertisements on the front page and a GTX660Ti Asus giveaway? Isn't that 3 GTX660Ti advertisements right there? Interestingly enough, one of those ads is for the Galaxy GTX660Ti GC card they reviewed against the faster 7950. Just sayin'......

Good thing they don't do car reviews then with that VW Tiguan Ad on the right