Asif1924

How much faster will the GTX 780 be to the Radeon 8970

28 posts in this topic

We just have to see when the time comes. It is too early to discuss about the specs.

0

Share this post


Link to post
Share on other sites

Its called speculation. And last time I checked, there's no law against speculation.

In addition to this, anyone who tracks the computer graphics industry for the last 5 years will be able to make an educated guess about the potential performance of these cards.

For instance, one article states that the 8970 will be 20% faster than the current 7970. Based on that metric, we can surmise that it will also be faster than the current GTX 680 at roughly the same amount (lets say 15% since a stock GTX 680 is faster than a stock 7970 by about 5% on average). Now, a tech report article stated that if nVidia were to make a GTX 780 based on the GK110, it would be about 25% faster than the current GTX 680. But I read an article recently which stated that the GTX 780 will likely be based on a derivative of the GK110, maybe a GK114 lets say. And if thats the case, I'm sure nVidia would squeeze out more performance from the chip, so lets say a GTX 780 based on a GK114 would be 25% faster than the GTX 680. By deduction, the GTX 780 will be faster than the 8970 by about 5%.

I said 10%, so we'll see how that turns out.

Disclaimer: This is my own educated guess. This is not fact. And I've specified how I've arrived to my conclusion. Everyone is free to make their own conclusion.

0

Share this post


Link to post
Share on other sites

The whole purpose of this forums was that the people could also speculate about upcoming graphics cards (in other case there wouldn't be a proper section for GeForce 700 series yet).

1

Share this post


Link to post
Share on other sites

we havent done with gtx 600 series and we already talking about gtx 700 series with no major leaks we all know that gtx 780 will not use all the smx of tesla k20 so just relex man dont rush enjoy the gtx 600 series XD

0

Share this post


Link to post
Share on other sites

I'd guess more like 15%, depending on how much of a GK11* chip we get. NV has a lot of TDP to play with when it comes to the 680 vs 7970 GE in terms of watt/performance, however it can't be too much faster than the 680 or Maxwell will be less appealing to potential buyers.

0

Share this post


Link to post
Share on other sites

What I liked about this round of graphics cards wars (GTX 600 series and HD 7000 series) was that nVidia's GTX 680 was actually supposed to be its mid-range card (i.e. performance card), and it turned out faster than AMD's high-end card. This also begs the question: Does it make more sense for GPU makers (nVidia, AMD) to wait until their competitor releases their cards first? Because essentially, after gauging the performance of the HD 7900 series, nVidia was able to tweak its mid-range chip to compete with AMD's high-end. Had they released the GK104 as a GTX 660 Ti before AMD released, what would have happened? AMD might have branded their Tahiti-based cards as HD7800 series, then probably unleashed their new GCN2-based cards as 7900-series? If that occurred we would probably still be waiting for the GTX 680 by around this time, and it would have been based on the GK110 or a derivative...

Who knows, I'm just speculating and analyzing...its interesting...

0

Share this post


Link to post
Share on other sites

I think so too.

Timing is important when it comes to computer products. Look at Bulldozer, they are supposed to be released in 2009, 2010 but not until 2011 we can see how it performs and it already too late for AMD since FX-8150 is absolutely terrible in comparison with Intel i5 2500k and i7 2600k..

The same principle applies with nVIDIA GTX 400 series. Well, they performs like a beast, especially the GTX 480 and the GTX 460 despite you will need a power nuclear plant to feed those monsters, but they were 6 months late. By that time the HD 5000 series already had been sold like hot cakes.

If you look at the Gk110, you will see it is not a high end chip at all, it is a SUPER high end chip. 7.1 B transistors is not a small number at all. Tahiti has 4.3B transistors and not all of those transistors are used for gaming, part of them are for GPGPU, computing stuff. GTX 680 has 3.54B transistors and nearly all those transistors are used for gaming. If you look at the computational capability of HD 7970, then look again at the difference of transistors between Tahiti and GK 110, you'd realize that the Gtx 680 is not a mid range at all 'cuz it only lacks of computational capability.

0

Share this post


Link to post
Share on other sites

Yea the 680 is to the 780 as the 480 is to the 580.

My nuclear plant is pretty small :ph34r:

0

Share this post


Link to post
Share on other sites

Hmmm... About this much.

thismuch.gif

0

Share this post


Link to post
Share on other sites

We Don't Have The Confirmed Specs Yet...So We Have To Wait And See Who Will Kick Ass....

0

Share this post


Link to post
Share on other sites

Hmmm... About this much.

thismuch.gif

ok-1.png

0

Share this post


Link to post
Share on other sites

Hmmm...looks like the GTX 780 may actually be much more faster than the 8970 if the following article is true:

http://videocardz.com/35422/sea-islands-to-feature-only-15-performance-increase

If 8900 series is to get only a 15% performance increase vs 7900 series, there's not going to be much competition for nVidia...unless...AMD is saving the 30% performance increase for Sea Islands 2 later in the year...

0

Share this post


Link to post
Share on other sites

Asif1924, I am pretty sure when the performance increases are taken into account, AMD uses its fastest single-GPU and that is not the HD7970, but HD7970 GE. Why? Because if you make a card that's only 15% faster than HD7970, it'll be just 3-4% faster than 7970GE -- no point of even releasing anything like that. I believe your starting point for 15% faster for 8970 is the wrong SKU. HD7970GE was poorly marketed and should have been something like HD7980, but it is a separate and flagship SKU for AMD's 7000 series, having MSRP of $449 against $399 for vanilla 7970.

Looking at the performance specifics:

HD7970 GE is 11% faster than HD7970 (and 5% faster than 680 at 1080P, 11% faster at 1600P). If HD8970 is 15% faster than 7970 GE, that means relative to GTX680, it would be 21% faster at 1080P (1.05 x 1.15) and 28% faster at 1600P (1.11 x 1.15).

If GTX780 is just 15-20% faster than GTX680, then the 780 would actually be slightly behind.

If GTX780 is 25-30% faster than GTX680, then 780 would have a 3-5% lead at 1080P/1600P.

The key unknown is if NV will increase memory bandwidth on 780 by including 7Ghz GDDR5 chips or enlarge the memory bus width. Increasing memory bandwidth will benefit GTX780 a lot more at high resolutions --> Even if it's just 15-20% faster than 680 on average, it could be 30% faster at higher resolutions as 680 is at least 10% behind 7970 GE at high-rez.

The other unknown is the relative breakdown of future game engines. Will more games use DirectCompute for global illumination and contact hardening shadows such as Sniper Elite V2, Dirt Showdown and Sleeping Dogs? If so, that would favor HD8000 series. Will future games use more extreme tessellation (i.e., Crysis 3), drop native MSAA in favor of FXAA/MLAA, include more PhysX, rely on FP16 textures? If so, that would favor GTX700 series.

Also, your comment that GTX680 was supposed to be a mid-range product has never been proven, despite this theory being widely popular for most of 2012. If anything more recent information highlights that it was the only high-end consumer chip NV intended to launch even before seeing Tahiti XT. This is because even now NV is having difficulty filling in corporate client orders for Tesla K20 (GK110) despite that chip initially slated for Q3 2012. In August 2012, NV confirmed early sample shipments would be delayed to October 2012 and volume production for GK110 would not even pick up until December 2012. Keep in mind corporate clients are a high profit margin segment for NV and it's in their best interest to deliver the product to them as soon as possible.

"Second generation NVIDIA Maximus-powered desktop workstations featuring the new NVIDIA Quadro K5000 ($2,249 MSRP, USD) plus the new NVIDIA Tesla K20 GPU ($3,199 MSRP, USD) will be available starting in December 2012."

Is GTX680 a mid-range chip in the entire "Kepler" family? Sure because there is a GK110 7 billion transistor ship. However, NV has finally split up their professional and consumer graphics, which means GK104 is the high-end GeForce chip this generation.

The fact of the matter is NV could not manufacture a GK100 size chip in sufficient volumes given the wafer constraints they faced most of this year. How do we know this? For starters NV discussed on their conference calls that they were wafer constrained and expected manufacturing to pick up later in 2012 as TSMC ramped up 28nm wafer capacity. Secondly, we saw NV's strategy play out in front of us when they chose to reallocate/prioritize their production for mobile Kepler chips first to fill OEM laptop design wins, which is why they delayed the entire sub-$300 desktop Kepler line-up. Yes, there were approximately 120,000 units of GTX570 that needed to be cleared but that only affected the delay of GTX660Ti, The delays for GTX650/650Ti/660 were all related to NV first filling in mobile Kepler demand for laptops - a higher growing market segment than desktop PCs are.

The other reason why GK100 style chip was never possible in the consumer space was the cost of 28nm die. NV wants to maintain 50% gross margins and the only way to do that by selling a 600mm2 die is to price at $900-1000. How do we know this? GTX690 is 2x 294mm2 design and it costs $999. Again, even if NV could have produced GK110 in volumes (and I think this was impossible to begin with), they wouldn't have done it anyway due to this chip being impossible to sell at $499 without drastically lowering their gross margins. The alternative could have been waiting to build up inventory of die-harvested 12-13 SMX failed GK110 chips and launch them as a high-end gaming chip but as you saw above, NV can't even deliver K20 (GK110) to the market before December 2012 in volumes. So where would they have gotten the die harvested GK110 chips in 2012?

In the beginning of the year it was fun to speculate that GK104 was just a mid-range chip that NV used because they could but the reality is they used it because they had to. Mounting evidence shows GK104 was never meant to be NV's mid-range GTX660Ti but was actually the only chip they had, which is why they were worried Tahiti XT would beat them. It's no wonder the company was "relieved" after seeing benchmarks of Tahiti because they had nothing faster than GK104 ready for consumer space.

Even as of November 2011, VR-Zone already denied the existence of a large monolithic GK110 chip:

http://vr-zone.com/a...dmap/14067.html

Instead they noted this:

"Following right after GK104 will be GK110 - a dual GK104 flagship, thus completing NVIDIA's line-up for most of 2012 - remarkably similar to AMD's sweet spot strategy."

GK110 was the wrong name to use in that chart but it was really a GTX690 all along serving as NV's flagship for all of 2012. VR-Zone got the codename for dual-GK104 wrong but a large monolithic GK110 was never in the cards as a consumer GPU in 2012 for NV, even before they knew how fast Tahiti XT would be. NV strategically re-balanced Fermi architecture to refocus on for performance/watt with Kepler and 28nm wafer constraints early on made sure GK100/110 was impossible to manufacture in volumes. Since GTX680 was made on the early stage 28nm node (much like HD7970), and it already uses up to 180-190W at peak in games, it would have been inconceivable how a chip like GK100 could have possibly stayed under 250W TDP in 2012 without serious clock speed reduction. HD8970 and 780 will be manufactured on a more mature 28nm node, which should in theory allow them to run at a similar clock at lower voltages/leakage (or higher clock speeds at the same voltage). This is also the same reason Haswell should be much better than IVB since it'll be made on a more mature 22nm node (revision 2 if you will).

As far as GTX780 goes, GTX680 is very memory bandwidth limited which is why the performance falls off sharply at 2560x1440 or higher against 7970 GE series. It's hard to say what GK114 will look like but NV is not exempt from the laws of physics. If HD8970 at 400-410mm2 is running into power consumption issues at 1.05-1.1ghz, then NV will have the exact same problem with Kepler since both 8970 and 780 are made on 28nm node. NV can still beat HD8970 but it's doubtful it will be more than 10%, strictly based on rumors we heard so far.

1

Share this post


Link to post
Share on other sites

hehe now we know the forum script supports pretty long posts! ^_^

0

Share this post


Link to post
Share on other sites

Yeah, it's called BestJinjoAnalysisMod :)

Biggest Post In VCF :) :)

0

Share this post


Link to post
Share on other sites

With so much speculation on GTX780 being GF114 or GK110, it's fairly difficult to estimate how much faster it would be over HD8970.

S / A now says NV will launch a consumer GK110 part next year as a GeForce:

http://semiaccurate....mer-gk110-card/

Are we going to have GK114 be 15% faster than GTX680 and GK110 another 20-30% faster on top of that? Who knows. If GK114 is 15% faster than GK104, will NV sell it as a GTX760Ti/770 for $349-449 and position GK110 GTX780 at $599? Will NV launch a GK114 at $500 to compete with HD8970 and position GK110 as an uber 8800GTX Ultra style $800 card? So much uncertainty now given that this series won't drop until March 2013 at the earliest it seems.

0

Share this post


Link to post
Share on other sites

so much rumors, so little rum. But rest assure the GTX 780 will outperforms the HD 8970 about ~ 15 % more or less. It's been a tradition for the past 7 years. Unless one of the companies takes an arrow in the knee.

0

Share this post


Link to post
Share on other sites

NV lost the single-GPU performance crown in 2006 to X1950XTX and this generation to HD7970 GE. That means NV had the lead from November 8, 2006 (8800GTX) to December 22, 2011 (HD7970). That's about 5 years. Not entirely sure how you arrived at NV having a 15% lead the last 7 years. I wouldn't say it's set in stone that GK110 will even launch as OBR stated only GK114 will and S/A states GK110 is still in the cards. If GK110 will launch, then I agree with you that it should have no trouble outperforming the HD8970.

As a point of reference, it seems NV has run into TDP / yield issues with Tesla K20 as not only was it delayed all the way to December 2012 for volumes production, resulting in backlog of pre-orders, but the final specs have been reduced significantly from the original 15 SMX clusters. The final specs for Tesla K20 are extremely underwhelming from what it was supposed to be. It's just 13 or 2,496 CUDA cores and clocks are just 700mhz or so:

http://www.cadnetwor...z_g13/system=88

If NV can launch a 1000 mhz + GK110 with those specs, it would be very impressive. Hopefully 28nm node will improve enough by Spring 2012 to allow them to get there.

0

Share this post


Link to post
Share on other sites

Jingo nice analysis....

But check out this new bit of information about the GTX 780. Does your analysis still stand?

http://videocardz.com/35830/new-nvidia-roadmap-reveals-maxwell-coming-in-2014-geforce-gtx-780-details

0

Share this post


Link to post
Share on other sites

Sorry for misspelling your name...I meant "BestJinjo" not "Jingo"...

0

Share this post


Link to post
Share on other sites

Asif, I honestly don't know which way NV will go - a 7 billion transistor GK110 with 12-13 SMXs or rather a pure gaming chip like GK104 but larger in size, without all the compute related features like ECC and double precision transistors that take up extra space. At higher resolutions with the latest drivers HD7970 Ghz is now 11% faster than GTX680 and after-market HD7970 cards are approaching 19-20%.

http://www.techpower...X_Turbo/28.html

or

http://www.techspot....rds/page12.html

That means GTX780 needs to make up at least 10% just to come to parity with HD7970Ghz and then whatever extra on top is what's needed to beat HD8970. You can see how hypothetical math would work:

GTX680 = 100%

HD7970Ghz = 111%

Let's assume HD8970 is 20% faster than HD7970 Ghz = 111% x 1.2 = 132%

If GTX780 is 50% faster than GTX680 = 100% x 1.5 = 150%

Then GTX780 will be just 14% faster than HD8970 which is not a lot (150% / 132%). Of course I have no idea if HD8970 will be 20% faster than HD7970Ghz or if GTX780 will be 50% faster than GTX680 or more. At 1920x1200 or below NV continues to do well and I expect GTX780 to beat HD8970 at those resolutions easily. However, the higher the resolution, the more monitors a gamer has, the larger the lead for the HD7970Ghz. That means GTX780 will need to make up a ton of deficit left by GTX680 for gamers on 2560x1440+ monitors or those with multi-monitors. In some of those cases GTX680 trails by 20%!

"That said, we have to wonder why anyone would bother with the GeForce GTX 680 4GB for extreme resolutions when the Radeon HD 7970 GHz Edition was constantly faster at both 5040x1050 and 7680x1600. In fact at 7680x1600 the 7970 GHz Edition was on average 20% faster than the GeForce GTX 680 4GB in the half dozen games that we tested with."

http://www.legionhar..._680_4gb,6.html

0

Share this post


Link to post
Share on other sites

Now that the GTX780 is out, and now the R9 290X which I'll categorize as the 8970, we can now confirm or deny my original predictions about these cards.

0

Share this post


Link to post
Share on other sites

Based on the early and preliminary reviews and benchmarks, R9 290X is about 20% faster than GTX 780, and apparently trades blows with Titan.

 

But to achieve this, the R9 290X is clocked higher, has more stream processors and needs to be overclocked to completely beat GTX 780. I vaguely recall the benchmarks indicating that it is operating in the so-called "Uber" mode in order to beat GTX 780 and trade blows with Titan. It also consumes more power than GTX 780. Also, according to some preliminary review by Legit Reviews, its Performance per watt is lower than GTX 780.

 

I think stock for stock it will equal GTX 780 and may be slightly faster by 5%---once we see real-world reviews. 

 

Nervertheless this has prompted GraphZilla to release the GTX 780 Ti to maintain the performance crown.

0

Share this post


Link to post
Share on other sites

20% faster in 1440+

should be smaller in 1080p

Also, NVidia could probably go 16-17SMX in ~550mm^2 (maybe even 18, not sure on the exact size of the DP units in GK110) if they took out the DP units.... However, that spin would cost at least 20-30 million dollars, it would 100% be not profitable.

0

Share this post


Link to post
Share on other sites

So the plot thickens. Graphzilla unleashes the 780Ti which is 30% faster than 290X (Quiet mode) at 25x16 rez.

 

http://videocardz.com/47587/nvidia-geforce-gtx-780-ti-official-gaming-performance

 

Lets say its 20% faster on average than 290X in quiet mode. And in Uber mode maybe its 5% faster. Thats still a better engineered card because it consumes just 250W vs the 290X's 300W, and runs cooler (avg 85 degrees vs the 290X's 94 degrees), and is quieter than the 290X.

 

So the AMD Fans may complain that it is more expensive---at $699 vs the $549 for the 290X, but you get far more; CUDA, PhysX, GPU Boost 2.0, ShadowPlay (I've used this myself on my 670FTW, works great), GeForce Experience, plus I think its bundled with games.

 

Not to mention nVidia drivers are better. They always work. Every AMD card I've ever had in the past I've always had issues with their drivers. And some AMD cards I've had issues with overheating...

 

So, in my opinion, the 780Ti is better all-round. 

 

That said, this entire community still needs to support AMD, because without them, nVidia would have a monopoly on this market. And despite all that I've said, I think AMD did a super job engineering a card that was equal to Titan, and gave nVidia a run for their money.

 

Now what I would like to see is AMD improve their driver situation, and add more value to their product stack. Mantle is new and has yet to even be adopted by the gaming industry much less become a mainstay. True Audio sounds like a great technology, but we're talking about graphics cards here. The fact that they absolved the need to have crossfire connectors for multi-card gaming is a great solution. Also, lets not forget they were the first to launch seemless multi-monitor gaming via EyeFinity. So I dont doubt their engineers can innovate. 

 

Anyway thats my 2 cents.

0

Share this post


Link to post
Share on other sites

*yawns* let me say this again: different cards are engineered differently.

Anyhow, the ti will be better on air cooling, hands down.  Liquid cooling (and overclocking hard) the 290x will probably match or beat it.


Oh, and on the price, if you want to play the games (and buy the shield) anyhow, the Ti WILL BE A BETTER DEAL as the r7/r9 cards are currently not in AMDs bundle lineups last I checked (lmfao)

 

CUDA is worth less to games, just leave it out... It is like saying one of AMD's advantages is that it has GCN.  It doesn't mean anything outside of professional purposes.

PhysX is still used in a few games, if you play those games IT IS worth it generally speaking.

GPU boost is just a version of powertune (and vise versa).

Shadowplay is something AMD does not have, the usefulness varies.

AMD has a counter to geforce experience coming according to Hawaii annoucement day... although I cannot say if it will match it.



Still don't see where AMDs drivers issues are, please explain... I honestly don't see any issues outside of CFX, and, well, both friends of mine, and personal experience dictates that AMD is just as good as Nvidia in majority of places.  Of course, 100% parity is what would be best :)

AMD is much better at engineering a solution around problems (historically) while Nvidia does software around them.
Both are catching up to each other (Nvidia is getting better hardware and AMD better software) but currently it looks like AMD is catching Nvidia faster than NVidia is catching AMD (Maxwell will be the determining factor for this)

0

Share this post


Link to post
Share on other sites

Please sign in to comment

You will be able to leave a comment after signing in



Sign In Now