GTX 670 Vs HD 7970 At Max Overclocks

[OILFIELDTRASH]

I prefer Nvidia but some of you people are just being overboard. here a gtx670 oced to 1330 is easily beaten by a 7970 oced to 1250 at 1920x1080. considering the gtx670 is only around 5% slower than the gtx680 clock for clock then even the gtx680 oced would be matched or beaten here by the oced 7970. an yes not ever 7970 will do 1250 but not every gtx670 will do 1330 either.

See that card way down there at the bottom? The gtx 580? My overclocked gtx 480 is about right there performance wise. Now look way up at the top there. My new card the windforce oc 670. Yep that's it. That's all I care about.

 

[aaksheytalwar]

The 7970 is the fastest not 670 in that chart

 

[BallaTheFeared]

The 7970 is the fastest not 670 in that chart

Who really cares when all are over 60, while none are near 120 at 1080p?

So either you need two cards with a 120Hz screen, or you need a input lag inducing IPS at 60Hz. You want to go over 1080p you'll still want two, going triple screen might as well buy 3-4. At this point AMD doesn't even really offer multi card support so there is only one option there.

Otherwise what exactly are we talking about here? A few fps over 60 that you won't notice? If that's the case let's bring out the min results since that's all that matters in this discussion.

 

[OILFIELDTRASH]

The 7970 is the fastest not 670 in that chart

See where the 580 performs? That's about where I'm at performance wise. See where the 670 oc performs at? That's where I will be with the 670 oc. Big performance difference.

You can throw a 690 or a 7970 crossfire or whatever in the chart and guess what? I will still look at the 580 and the 670 because that's what pertains to my situation.
For $400 there isn't a better bang for the buck.

 

[aaksheytalwar]

He said 670 was fastest which isn't true. It is one of the fastest, not fastest. And yea the diff isn't noticeable.

 

[SirPauly]

BF3 :

7970 wins by .9fps

Batman :

7970 wins by 6.5fps

Skyrim :

7970 wins by 1.4fps

Amd fans still think the 7970 is worth 50 dollars more then the GTX 670?

It's really not about fans but it may be about a gamers needs -- what if a gamer desires to game at 2560 x 1600 or multi-monitors with 1080p, 1400p or 1600p?

AMD is closing the gap with 2560 x 1600 and may be separating with multi-monitor resolutions -- there is some differentiation and some OC prowess.

 

[aaksheytalwar]

This is probably the fittest fight ever. Such a close fight is probably the first they have had, it is like gtx 570 vs 6970 had 580 never existed

 

[KompuKare]

Well then that makes sense, I guess we can stop comparing older cards to newer cards in power consumption too then since the platform can make a world of difference. I mean unless of course they actually go back and retest with the same system.

Let me know when [H] tests 7970 OC and 670 OC on the same platform with power draw numbers.

Or, stick to sites don't just measure system draw at the socket but who use one of these:

Like ht4u.net, behardware.com, hardware.fr, and techpowerup.com. Of course that does mean waiting for those sites to review the cards overclocked. Never understood why some sites cannot be bothered to give real numbers. Makes their power draw results incomparable since even if one sites keeps an old system with a constant CPU & mobo, other sites are unlikely to that same system.

And no, taking system idle with iGPU vs max draw is not accurate either since there's no way of saying which part of the difference is due to the maxed out dGPU being reviewed and which part is due the CPU having to do work to max out the dGPU.

On caveat I can think of: certain combination of cards and their drivers may offload different tasks to the CPU. So for calculating which PSU to buy the max total system power draw may still be (somewhat) relevant. 'Somewhat' since draw at the wall tells you nothing about how much 12V for instance your PSU has to provide.

Oh and one final thing: measuring at the wall, the actual component draw from PSU will obviously depend on the efficiency of the PSU but the efficient differences between 110V and 240V. And the PSU's efficiency at certain % load. And... Just too many variables really.

 

[Crap Daddy]

I hope you all do realize that the 670 is a 400$ card and is competing with the 7950?
And you know that it is like 18% faster for the same price point?

 

[OILFIELDTRASH]

I hope you all do realize that the 670 is a 400$ card and is competing with the 7950?
And you know that it is like 18% faster for the same price point?

Price and power consumption is now a non factor. Maximum ocing cherry picked cards is the only suitable way to determine cards now.

 

[aaksheytalwar]

These aren't cherry picked cards, these are realistic overclocks. Not every 670 reaches 1330 anyway, so the 670 could be cherry picked as well.

 

[SirPauly]

This is probably the fittest fight ever. Such a close fight is probably the first they have had, it is like gtx 570 vs 6970 had 580 never existed

Yeah, but this round, nVidia may offer more bang-for-the buck combined with their differentiation strengths, and still retained performance leadership with the GTX 680 and GTX 690. AMD lost one of their most important selling points -- value.

 

[aaksheytalwar]

It was the same last round, 6970 wasn't cheaper than 570

 

[BallaTheFeared]

Or, stick to sites don't just measure system draw at the socket but who use one of these

But that doesn't really matter, does it?

I mean technically speaking, sure if you really want to know what the card itself draws then yes, that's the way to go.

However if you want to see how the card plays with processors and the rest of the system you need to look at total system draw.

This played out visibly with Fermi. You can see how cpu hungry it was, and that plays into total system draw. We have more refined and better chips like SB and IB now, however it's still a part of the equation imo.

I could be wrong, but it seems the situation is a bit reversed compared to last gen. Last gen the 480 wasn't so power hungry it dwarfed the sun, but when you combined it's power draw, with it's need for more clocks than the 5870, on a platform that could use 500+ watts at 4.2GHz in Prime95, well things got a bit carried away.

It seems now the 7970 requires more cpu than Kepler, even if the 7970 only draws 30 more watts than the 680, if it requires 70 more watts from the cpu to make it's fps they should be accounted for and rest on the shoulders of the 7970 just like it did with Fermi.

 

[KompuKare]

But that doesn't really matter, does it?

I mean technically speaking, sure if you really want to know what the card itself draws then yes, that's the way to go.

However if you want to see how the card plays with processors and the rest of the system you need to look at total system draw.

I guess. I did end my post with 'Just too many variables really.' Thing is it would be nice to get figures to get to the bottom of 'card X compared to card Y' and to be able to quantify it. I was tempted to name something like Furmark as suitable because I though it doesn't use much CPU but 1) just ran it and it used up 100% of one of my cores and 2) if the card is being pushed, the driver is being pushed and that uses a certain CPU load too.

Maybe there it's possible that some gCompute application may be able to generate a GPU load without loading the CPU, but I think that's a bit too much work for any site to do just to satisfy some scientific curiosity about GPU power usage.

 

[AdamK47]

Oh, this is getting just as worse as politics. Some people post graphs showing their side winning while others post graphs showing the other side winning. This never ends, does it?

 

[Olikan]

These aren't cherry picked cards, these are realistic overclocks. Not every 670 reaches 1330 anyway, so the 670 could be cherry picked as well.

IIRC, there was a retail vs press card in some site...gtx 680
you know, nvidia turbo is totaly crazy...

the retail card end up beeing 2% slower....at stock setting :whiste:

 

[SirPauly]

It was the same last round, 6970 wasn't cheaper than 570

Indeed but this time it is priced the same as the HD 7950.

 

[shauni]

Or, stick to sites don't just measure system draw at the socket but who use one of these:

Like ht4u.net, behardware.com, hardware.fr, and techpowerup.com. Of course that does mean waiting for those sites to review the cards overclocked. Never understood why some sites cannot be bothered to give real numbers. Makes their power draw results incomparable since even if one sites keeps an old system with a constant CPU & mobo, other sites are unlikely to that same system.

And no, taking system idle with iGPU vs max draw is not accurate either since there's no way of saying which part of the difference is due to the maxed out dGPU being reviewed and which part is due the CPU having to do work to max out the dGPU.

On caveat I can think of: certain combination of cards and their drivers may offload different tasks to the CPU. So for calculating which PSU to buy the max total system power draw may still be (somewhat) relevant. 'Somewhat' since draw at the wall tells you nothing about how much 12V for instance your PSU has to provide.

Oh and one final thing: measuring at the wall, the actual component draw from PSU will obviously depend on the efficiency of the PSU but the efficient differences between 110V and 240V. And the PSU's efficiency at certain % load. And... Just too many variables really.

I wish I was french so I could read hardware.fr more. It seems like a really good site.

 

[Guovssohas]

The thing that stands out to me is the 7950, it's really a joke compared to 670. Both at the same pricepoint.

 

[blackened23]

Oh, this is getting just as worse as politics. Some people post graphs showing their side winning while others post graphs showing the other side winning. This never ends, does it?

Unemployment reached 3 year lows and US GDP went up upon release of the GTX 680.

 

[cmdrdredd]

The way I compare is assume you can't overclock either one. Which is better value? Reason I do it that way is because the clocks people on forums or review sites hit are not usually typical. As always your mileage may vary with overclocks so generally I don't suggest using overclocked performance as a guide.

 

[RussianSensation]

Amd fans still think the 7970 is worth 50 dollars more then the GTX 670?

I want to know where I can get $450 HD7970 with an after market cooler that doesn't have a 50% chance of coil whine?

What's next? HIS $485 is the cheapest 7970 with a quiet fan, that can hit > 1.2ghz and no reviewer that I am aware of has complained that it has coil whine like the Sapphire Dual-X card. So it looks like there is an $80+ delta not $50.

Regardless, AMD should release a fresh batch of > 1ghz 7970 cards. When that happens, people can revisit this discussion. Right now there is about an $80-100 delta between a GTX670 and a good version of the 7970.

What I disagree with is the main discussion is around the wrong cards. A person who was looking to buy HD7870 for $330-350 or HD7950 $400 might just as well get a 670. Going from $330-350 to $400 670 for a huge performance boost might be manageable. Going from $330-350 7870 to $450-500 HD7970 for that buyer likely means going way over of his/her budget. This is exactly where GTX670 shines. GTX670 is the perfect upsell from people who were on the fence thinking of getting HD7870 or even 7950. It made both of those cards irrelevant.

Essentially the market is now a 3-horse race:

$140 6870
$250 7850
$400 670

GTX670 is 28% and 17% faster than 7870 and 7950, respectively, making both of them overpriced. The 2% of the market who want the best can fight over 680/7970 but it changes nothing about 670 making 2 other AMD cards irrelevant imo.

I think GTX670 is like the 6800GT of X800Pro/X800XT/6800U generation. That card sold like hot cakes.

 

[IlllI]

no idea what the point of this topic is. Seems pretty dumb really

 

[3DVagabond]

AMD overclocking the GPU from the factory means nothing to me. It's pure marketing. It's just like how they underclock the 7850 and 7950. It's BS.

You really need to read what AMD said about why they are able to release the more recent 7970's at higher clocks. They are capable of clocking higher while using less power. The one area that the GK104 is actually markedly better than Tahiti is in efficiency. Especially at higher clocks. It's not pure marketing. The chips are getting better as the process matures. This isn't the first time this has been pointed out.