GeForce GTX 680 SLI Performance Surfaces

[btarunr]

NVIDIA's big GeForce GTX 680 launch is just around the corner, but performance figures are already trickling in. Last week, we were treated to a wide range of benchmarks covering a single GeForce GTX 680. Today, VR-Zone posted a performance-preview of the GeForce GTX 680 in 2-way SLI configuration. A set of two GTX 680 cards were put through 3DMark 11 in Entry, Performance, and eXtreme presets. It should be noted here, that the GTX 680 cards were clocked at 1150 MHz core, and 1803 MHz (7.20 GHz effective) memory.

In the Entry preset, GTX 680 2-way SLI scored E22878; it scored P16860 in Performance preset; and X6243 in eXtreme. 2-way SLI of GTX 680 should be fit for 2560x1440/1600 resolution gaming. The rest of the test-bench consisted of Intel Core i7-3930K six-core processor clocked at 5.00 GHz, with 16 GB of quad-channel DDR3-2133 MHz memory, and ASUS ROG Rampage IV Extreme motherboard.



View at TechPowerUp Main Site

 

[TurdFergasun]

ahh interesting so w1zard was involved?... no? ok no more 680 news until then. k thx bye

 

[MxPhenom 216]

that is an insane memory clock and amount of bandwidth

 

[DriedFrogPills]

But isn't the default clocks on the 680 1006? Or is the 1006 one of the steps with the new aotu OC system nVidia is introducing. Either way can't wait until Wizz's review hits almost at the point of building a new rig.

 

[bear jesus]

I am very excited to see the reviews, as ivy bridge E will not be coming soon, a GPU upgrade is all i am after for now.

that is an insane memory clock and amount of bandwidth

I agree the memory clock is really impressive but the amount of bandwidth is not as impressive as it's only a minor increase over a slightly overclocked 580

The bandwidth is way more impressive on this

 

[hhumas]

nice............

 

[HumanSmoke]

I agree the memory clock is really impressive but the amount of bandwidth is not as impressive as it's only a minor increase over a slightly overclocked 580...[link] The bandwidth is way more impressive on this [7970]

Because everyone knows that bandwidth is the primary indicator of performance.

GTX 280 : 512 bit / 8 x 2214MHz effective = 141.7 GB/sec

GTX 560Ti : 256 bit / 8 x 4008MHz effective = 128.27 GB/sec

I knew this whole upgrading and architecture refinement was bogus...

 

[bear jesus]

Because everyone knows that bandwidth is the primary indicator of performance.

GTX 280 : 512 bit / 8 x 2214MHz effective = 141.7 GB/sec

GTX 560Ti : 256 bit / 8 x 4008MHz effective = 128.27 GB/sec

I knew this whole upgrading and architecture refinement was bogus...

I never once commented on performance, i was just commenting on nvidiaintelftw's comment about the amount of bandwidth being insane as i think there are more insane amounts out there thus why i gave an example.

 

[Anusha]

is it me or couple of similar clocked 7970's would score higher?

 

[MxPhenom 216]

is it me or couple of similar clocked 7970's would score higher?

I don't know do they?

 

[HTC]

is it me or couple of similar clocked 7970's would score higher?

It's not you: 7950 crossfired


If a couple of 7950s can do this, so can a couple of 7970s.


The 680 can do way more then this (i'm assuming) and so can 7970s: which one will be on top? We'll find out very soon!!!

 

[crow1001]

is it me or couple of similar clocked 7970's would score higher?

About the same with 7970s at 1075 core, take in to account the CPU is faster with 680 and the 680 is clocked at 1150Mhz. Fairly lame for a new high-end 28nm Nvidia GPU is it not? The again the 680 by all rights should be a 670 and priced $200 lower.



Kudos to you Nvidia, I honestly can't believe Nvidia are trying to hoodwink enthusiasts eyes with a mid range part overclocked and rebranded high end.

 

[HumanSmoke]

About the same with 7970s at 1075 core, take in to account the CPU is faster with 680 and the 680 is clocked at 1150Mhz. Fairly lame for a new high-end 28nm Nvidia GPU is it not? The again the 680 by all rights should be a 670 and priced $200 lower.

Doesn't seem to be much in it. The 7970's are OC'ed 16.2% core and 20% on memory...the GTX 680 is OC'ed 14.3% core and 20% on memory....wonder what the percentages are for power consumption, noise level and performance/watt.

So, more or less matches AMD's top SKU (resolution dependant I presume) with 20% less die space...the very definition of "fairly lame"

Kudos to you Nvidia, I honestly can't believe Nvidia are trying to hoodwink enthusiasts eyes with a mid range part overclocked and rebranded high end.

Um, haven't AMD already set the price for this level of single GPU performance? Whose to say that AMD wouldn't be looking at a similar spec to the GTX 680 if they hadn't belatedly decided that a higher level of GPGPU functionality might be a good idea after all (I'm assuming that the wider memory bus and compute might come at a cost to die area)

 

[crow1001]

I could go out and buy a 7950 at £300 overclock and still beat out the 7970 and 680, this is what Nvidia have basically done with the 680. Fact is looking at the specs a 256 bit bus matching previous high-end card the 580s memory bandwidth, the 580s replacement should be owning it in bandwidth and not just being the same, small core size, high clocks, less ROPs, the 680 is a mid range chip clocked high to compete with the 7970.

The real 680-probably 780 later in the year will have a much bigger core and would have trounced the 7970 at much lower clocks. Nvidia obviously are not ready with that core so they get a highly clockable mid range core and bang it out as high-end with a hefty price tag to go with it, only mugs will look at this card.

 

[xenocide]

Tahiti seems to offer lower performance than it should when compared to Pitcairn. Consider that an OC'd 7870 can outperform a stock 7970 which has substantially more memory bandwidth, 40% more SU's, 50% more VRAM, and over 50% more transisters. Now consider that that GTX680 is supposed to be a 292 mm^2 die (versus 365 mm^2 for 7970), with 256-bit memory (vs. 384-bit), 2GB of VRAM (vs. 3GB), and will operate at a lower TDP of 190w (vs. 210w), and it's actually quite impressive that it can keep up with the HD7970 at any level.

 

[W1zzard]

ahh interesting so w1zard was involved?... no? ok no more 680 news until then. k thx bye

huh what? i didn't have anything to do with this. what makes you think so ?

 

[Crap Daddy]

I could go out and buy a 7950 at £300 overclock and still beat out the 7970 and 680, this is what Nvidia have basically done with the 680. Fact is looking at the specs a 256 bit bus matching previous high-end card the 580s memory bandwidth, the 580s replacement should be owning it in bandwidth and not just being the same, small core size, high clocks, less ROPs, the 680 is a mid range chip clocked high to compete with the 7970.

The real 680-probably 780 later in the year will have a much bigger core and would have trounced the 7970 at much lower clocks. Nvidia obviously are not ready with that core so they get a highly clockable mid range core and bang it out as high-end with a hefty price tag to go with it, only mugs will look at this card.

First, you can get a 7950 and overclock it to 7970/680 level but as always goes those cards will overclock as well so you'll never reach that performance. Second, in 3DMark the 680 has a fairly big increase in performance compared to the 580, we're talking about for a single card extreme preset at around 3200 for the 680 and 2300 for the 580. And third, you are right, NV is cashing in on a card because... they can.

 

[crow1001]

First, you can get a 7950 and overclock it to 7970/680 level but as always goes those cards will overclock as well so you'll never reach that performance. Second, in 3DMark the 680 has a fairly big increase in performance compared to the 580, we're talking about for a single card extreme preset at around 3200 for the 680 and 2300 for the 580. And third, you are right, NV is cashing in on a card because... they can.

According to techpowerup's 7950 review it can clock as well as a 7970 and at same clocks there is nothing between the cards

It is hardly surprising that 1000 core 680 gets a bigger score than the 780 core 580, I mean I know of many mid-high end cards that can trounce the 580 in 3DM11 when clocked up the ass. For a new chip on a new design it is undweling to say the least.

Nvidia should have had the decency to put the core out as the mid range chip it is regardless of performance, doing what they are doing just makes them look like the assholes they are.

 

[xenocide]

It is hardly surprising that 1000 core 680 gets a bigger score than the 780 core 580, I mean I know of many mid-high end cards that can trounce the 580 in 3DM11 when clocked up the ass. For a new chip on a new design it is undweling to say the least.

Nvidia should have had the decency to put the core out as the mid range chip it is regardless of performance, doing what they are doing just makes them look like the assholes they are.

How many of those cards do it with 50% of the ROP's, half the memory bus, a die that's 40% smaller, and with a substantially lower TDP? Exactly.

 

[HumanSmoke]

Nvidia should have had the decency to put the core out as the mid range chip...

How can the GTX 680 be a mid-range chip if there is nothing above it in performance ?

...it is regardless of performance

Yeah, I can see that catching on with AMD's and Nvidia's shareholders. By your logic, the HD 7950 and 7970 should be priced the same....same die, same hardware (I doubt the difference in BOM between the two extends more than a few bucks). Moreover shouldn't the HD 7870 be retailing at ~$260 ? (Pitcairn being 212mm² to Tahiti's 365mm². 1GB of GDDR5 not adding that significantly to the BOM)

doing what they are doing just makes them look like the assholes they are.

Or maybe they are following usual business practice. What in your mind seperates AMD's HD 79xx/78xx pricing structure from Nvidia's?

 

[the54thvoid]

Nvidia should have had the decency to put the core out as the mid range chip it is regardless of performance, doing what they are doing just makes them look like the assholes they are.

There will shortly be two single gpu cards in the market. Vendor A and Vendor B both release them. Vendor A is out first with a card that is x% fast. This card at present is the fastest single gpu card in the world. It is therefore, undoubtedly a high end card.
Vendor B releases the card after Vendor A. By all accounts the newer card is possibly a slight bit faster.
Vendor B has now released a card that (in cases) is beating the previous high end card. Vendor B now has the fastest single gpu in history.

Low-----Mid-----High.

If we look at the simple 3 words above it can be viewed as a scale of low to high. At present the card vendor B is releasing is going to be the card that sits right at the right hand side, i.e. the highest performing single gpu in history. At present, vendor B has made no inference of any higher spec'ed single gpu card in it's immediate future.

Vendor B has released a high end card. It is absolutely illogical to suggest GK104 is a mid range card when it is
A) The fastest (contentious issue) card out there
B) Priced as such
C) Has no immediate successor (NOBODY knows if/when GK 110 is coming)

People that keep on saying GK104 is mid range need a reality check. Any product that is the fastest in it's current line up is it's high end product, irre-fucking-spective of it's production cost, die size or anything else.

If I could create a gfx card that cost me 2 cents to make but blew everything away I would market it at the cost higher than my competitors best card. It's called business. People that don't grasp that ought to stick to charity work and idealistic tree hugging.

 

[20mmrain]

1st. It doesn't matter if it was supposed to be a mid range chip or not! If you can't get your real Highend chip (GK110) to work I still consider it a fail. Who cares how good it will be if you can't get it to work. Sounds like excuse making from the Nvidia camp to me!

2. Clocks speed of 1006: Everyone jumped on this band wagon because videocardz.com listed the GTX 680 as of having those clocks. It has always been a rumor stemming from there. I am not saying the final clocks won't be above 706 but what I am saying is NDA hasn't been lifted and everyone on every tech site is acting like these cards have already been released and everything about them is fact.

3. Yes the HD 7970 can clock higher and score higher.... But like everyone else points out.... the GTX 680 is still very new and unreleased. Give it till it really releases so we have a fair comparison. No point in arguing a card we have no idea what the real specs are.

The one thing that I do notice that is consistent from site to site reported about these cards is..... That the HD 7970 competes hand in hand with the GTX 680 so I would expect that.

 

[alwayssts]

I don't get the mudslinging...still.

1536 dual-issue FMADD/scalar + (1:6) 256 sfu is a very efficient 32 ROP design.

A 1792 design where a vec16 unit can do 4 special functions and each shader does 2 flops (which equals 1536 shaders if doing 1:6 special function ops) is an equally impressive option.

One calls it a 680. One calls it a 7950.

THEY. ARE. SIMILAR.

Want more texture units, a smaller card, or like lime green? Get a 680.

Want more bandwidth, larger buffer for high-resolution, or prefer red? Get a 7950.

Please W1z, do a clock-for-clock (and or clock/equal bw) comparison in your review. Actual power consumption in a game included ('peak') would be great as well since both companies do tdp differently. I know it's not TPU's style, and people will likely bitch about comparing overclocking of one architecture vs closer to stock for the other, but I imagine what it really boils down to (other than fine-grained differences) is what each sku is capable...especially at these price tags. 7950's stock core speed/performance is such a misnomer it's sickening. Obviously it is conservatively clocked to make 7970 look better and (imo) so the 8850 can perfectly replace it's performance at under 150w. A similar story I believe is true for 7970 (judging by actual power consumption and the tdp space between pitcairn powertune rating and the pci-e power specs). nVIDIA, intelligently so, is exploiting a marketing hole and it's (again, imo) kind of bullshit to to compare architectures in that fashion. nVIDIA almost appears to be counting on it and someone needs to out the fallacy they are going to use with 680.

I imagine overall, when it comes down to it brass tacks, it's very close.

 

[dj-electric]

Clock to clock compressions are wrong from the very point you even think of it.
It does not make any sense. Each core and the frequency the manufacture decided to run it in.

 

[20mmrain]

*Look everybody" I just made my own video card brand! Call it 20mm's video cards.... The bad news is I was going to bring out a card 50 times more powerful then either the GTX 680 or HD 7970 but I was so disappointed with AMD's and Nvidia's performance that I came out with one only at the same speed. Don't worry it is the (WTF104) chip which was supposed to be my low end chip. But if you keep waiting I promise I will bring out that other 50 times more powerful chip later!!!

*Cough BS Cough*