Nvidia unveils the GeForce GTX 780 Ti

[FlanK3r]

guys, few wats its nothing, specialy in enthusiast segment. I dont care if is 100W or not. Its "nothing for me" and Im not the biggest enthusiast in GPUs....

[boxleitnerb]

Ok... Where was all these "even footing" "we need to change our testing/benching methodology" the past 2-3 years?
Why do you need to shift the parameters of the test? Why can't you test stock vs stock?
That's right because the results don't show what you want.

I don't know where it was in other reviews, but hardware.fr, PCGH, ht4u and Computerbase have been preheating Kepler cards from day one. Not finally several other sites begin to do it, among them hardwarecanucks, techreport and Anandtech iirc.
And it IS a stock vs stock comparison. Quiet mode is the stock setting, uber mode isn't.

Look at any GTX480 review.
No I don't know what average means... please explain.

I bet I can because those results look extremely strange. GTX780 and Titan pulling less than 200w?
There is something screwy going on there because those results don't reflect what other sites have found.

The power consumption numbers there are not strange for one very simple reason. 3DCenter compiles values mostly from realistic gaming scenarios, i.e. with pre-heating (at least HT4U, PCGH and hardware.fr). This leads to all Nvidia Boost 2.0 cards clocking lower due to the 80C temperature target being hit, thus they also consume less power.
I don't need to look at "any" GTX 480 review, because this compilation of measurements is far better. Most reviews test the whole system which is very prone to errors. Where else on the web do you have 5-6 measurements of the consumption of the card itself? Nowhere.

Btw still waiting on proof on your claim that Fermi could use 300+W under gaming loads...

[PedantOne]

1 GHz 780 is better then Titan, and Uber moded R9 290X in many games, with AA. Thats fact! In average it is the same performance, in 1080 and 1600. Not in Star-Trek 4K resolution of course.

780 Ti has DP power disabled for ever, here is Titan still king, with his price.

Whats i know, Nvidia sent 780 Ti to all german reviewers this week. Many ppl have cards at home, some leak are on the way!

PS. I dont know which revision of GPU 780 Ti has, i think it is B1, but will see. I am out of office now, cant to check it. But try to ask my coleage to dismatle that card.

[[XC] Oj101]

I can confirm that NVIDIA has killed DP performance with the 780 Ti. Using 1/3 DP doesn't make financial sense, as either it means selling the card for $1500+ or hurting their own workstation market. I'm disappointed, I use 64 bit floating point operations and the old GTX 580 offers more power than the 780 Ti. Ah well.

[LordEC911]

I don't know where it was in other reviews, but hardware.fr, PCGH, ht4u and Computerbase have been preheating Kepler cards from day one. Not finally several other sites begin to do it, among them hardwarecanucks, techreport and Anandtech iirc.
And it IS a stock vs stock comparison. Quiet mode is the stock setting, uber mode isn't.

I don't know about the rest of them, which I doubt, but hardware.fr definitely did not. If they did, they didn't mention it.
Uber is stock. Maybe you meant to say Uber isn't default?

The power consumption numbers there are not strange for one very simple reason. 3DCenter compiles values mostly from realistic gaming scenarios, i.e. with pre-heating (at least HT4U, PCGH and hardware.fr). This leads to all Nvidia Boost 2.0 cards clocking lower due to the 80C temperature target being hit, thus they also consume less power.
I don't need to look at "any" GTX 480 review, because this compilation of measurements is far better. Most reviews test the whole system which is very prone to errors. Where else on the web do you have 5-6 measurements of the consumption of the card itself? Nowhere.

Btw still waiting on proof on your claim that Fermi could use 300+W under gaming loads...

I didn't see them mention the list of games they used for power consumption.
I don't see them testing the card by itself...

I thought you didn't need to look at GTX480 reviews. You supposedly know it all.
I don't recall you asking for 5870 to be clocked up so that it could be at the same temperatures as GTX480... or the same power consumption.

Shifting the parameters. Tsk tsk.

[boxleitnerb]

I don't know about the rest of them, which I doubt, but hardware.fr definitely did not. If they did, they didn't mention it.
Uber is stock. Maybe you meant to say Uber isn't default?

I meant Boost 2.0 Kepler cards since those are the ones mainly affected by temperature. Boost 1.0 cards like the GTX 680 are barely affected. Sorry if that wasn't clear.
Yes, default. Stock is an unmodified card, I guess. Like no bios flash, no watercooling etc.

I didn't see them mention the list of games they used for power consumption.
I don't see them testing the card by itself...

Well, you need to go to the respective reviews of course.
TPU tests with Crysis 2
PCGH tests with Battlefield Bad Company 2
hardware.fr tests with Anno 2070 and Battlefield 3
HT4U tests with Tom Clancy's HawX
I forgot what Heise tests with, I'll write them an email.

It is obvious by the values that these are only the cards themselves. The values are way too low for power consumption of the whole system. It is also well known that these sites do test that way.

I thought you didn't need to look at GTX480 reviews. You supposedly know it all.
I don't recall you asking for 5870 to be clocked up so that it could be at the same temperatures as GTX480... or the same power consumption

What? The GTX 480 was measured as well by the sites I mentioned. What do I need other reviews for, especially with inferior testing methodologies?
And what's up with your second comment? Relevance?
The point is, quiet mode of the 290X is the default setting. Either you test both cards at default or not. You cannot have it both ways.

[Ace123]

I had 4 gtx 480s overclocked with 1.2v each with the CPU at 5.2ghz and my rig was pulling 1700w during 3d mark
http://hwbot.org/submission/2358521_...80_21354_marks

[GoldenTiger]

But speaking of stagnant CPU innovation, a lot of people have completely overlooked just how far along peripherals have come quietly in the background. I just got a Corsair K95 and holy is a mechanical keyboard a massive upgrade over the previous ergonomic one. It's interesting that it's almost more comfortable to type with, because even though I sacrificed the wrist support, the keys are just so damn responsive. Or you could look at the Korean 1440p monitors and how SSDs are finally affordable. Just in general I think there's a lot of nifty stuff that's been rolling out without much noise

I agree on this one... went from a 2560x1600 CCFL 30" panel to a much less expensive 27" 2560x1440 LED panel that can do 110hz and LOVE it. Also grabbed a mechanical keyboard with Cherry MX Brown switches, added O-ring dampers and it's amazingly comfy to type on (I use a wrist wrest with it). Been using SSD's since 2008 I think it was, but they've gotten to really good pricing and capacities at this point. Peripherals and less-glorified parts have definitely come a long way, no question.

[marten_larsson]

Techreport has 5870 using 319 and 253 W in their reviews (first being the one of GTX480 and the latter being review of 290X). Fermi is using 424W and 290X 346W, differences: 105W and 93W. So the pendulum swings both ways and the conclusion could only be: the difference isn't big enough to argue about. Fermi and 290X has about the same power draw. As for acoustics, according to techreport, the winner is GTX480 by a fraction (49,9dB to 50,1dB). However, the 5870 had a load noise level of 51,0dB in that review and only 49,1 in the 290X review. In all comparisons I've used ?ber mode. If not the clear winner in power would be 290X and in acoustics it would also be a slight victory for AMD.

Ref:
http://techreport.com/review/25509/a...rd-reviewed/12
http://techreport.com/review/18682/n...-processors/13

[Ace123]

Dude single worst card I have ever heard is Radeon 4890.... Mine sounded like a diesel turbo spinning up
290x can't be nearly that bad

[Olivon]

I don't know about the rest of them, which I doubt, but hardware.fr definitely did not. If they did, they didn't mention it.

In fact they did and they mentionned it :

Notez enfin que compte tenu de l'influence de la temp?rature sur les r?sultats, et du fait que nous mesurons les performances sur une table de benchs en laissant la temp?rature/fr?quence des diff?rentes cartes se stabiliser, la temp?rature de la pi?ce a ?t? contr?l?e et fix?e ? 26 ?C pour l'ensemble des tests.

http://www.hardware.fr/articles/910-...cole-test.html

They wait temperatures/frequencies stabilization to launch the bench. Room temperature is fixed @ 26 degrees celsius also, in order to give fair results accuracy.

[kuroikenshi]

Dude single worst card I have ever heard is Radeon 4890.... Mine sounded like a diesel turbo spinning up
290x can't be nearly that bad

some of my cars have a very silent turbo.... you must've had a bad experience with diesel cars.

[Zipelgas]

It boosts over 900mhz consistently in gaming situations. So that is false.

- show me a gtx 780 that cant to 1250mhz

[Tha Last Meal]

NVIDIA GeForce GTX 780 Ti has 2880 CUDA cores

You read that right! TITAN ULTRA is actually GeForce GTX 780 Ti. It packs 2880 CUDAs, 240 TMUs and 48 ROPs (ROPs are not mentioned though). NVIDIA GeForce GTX 780 Ti is clocked at 876 MHz for base and 928 MHz for boost. Card is equipped with 3GB GDDR5 memory running at 7 GHz. The GTX 780 Ti is using 384-bit interface.

[bamtan2]

I can confirm that NVIDIA has killed DP performance with the 780 Ti. Using 1/3 DP doesn't make financial sense, as either it means selling the card for $1500+ or hurting their own workstation market. I'm disappointed, I use 64 bit floating point operations and the old GTX 580 offers more power than the 780 Ti. Ah well.

please remind the class exactly what you use the titan advantage for

[STEvil]

Techreport has 5870 using 319 and 253 W in their reviews (first being the one of GTX480 and the latter being review of 290X). Fermi is using 424W and 290X 346W, differences: 105W and 93W. So the pendulum swings both ways and the conclusion could only be: the difference isn't big enough to argue about. Fermi and 290X has about the same power draw. As for acoustics, according to techreport, the winner is GTX480 by a fraction (49,9dB to 50,1dB). However, the 5870 had a load noise level of 51,0dB in that review and only 49,1 in the 290X review. In all comparisons I've used ?ber mode. If not the clear winner in power would be 290X and in acoustics it would also be a slight victory for AMD.

Ref:
http://techreport.com/review/25509/a...rd-reviewed/12
http://techreport.com/review/18682/n...-processors/13

My roomie is using my old GTX480 so i'll have that and a 290x when it gets shipped..

[Lightman]

some of my cars have a very silent turbo.... you must've had a bad experience with diesel cars.

I like my Turbo LOUD! It let's me know that I'm about to be catapulted forward very rapidly

[kuroikenshi]

^^ my early warning sign was the G's pushing me hard into my seat, that was followed by the speedo going off the dial..... oh the days...
Overclocked my cars, now I overclock my PC's....

wonder who in the forums had the sig on the overclocked testicles, that one made me laugh hard.

[Iconyu]

please remind the class exactly what you use the titan advantage for

Demoed on Titan, I don't think it would do so well on a 780GTX.

[kuroikenshi]

I never say that I am impressed,

I am F'ING IMPRESSED!

[Ace123]

I should probably stay on topic to

[[XC] Oj101]

NVIDIA GeForce GTX 780 Ti has 2880 CUDA cores

Oh wow, they found the post I made a few weeks ago

I'm sorry, I just find it amusing

[To(V)bo Co(V)bo]

I can confirm that NVIDIA has killed DP performance with the 780 Ti. Using 1/3 DP doesn't make financial sense, as either it means selling the card for $1500+ or hurting their own workstation market. I'm disappointed, I use 64 bit floating point operations and the old GTX 580 offers more power than the 780 Ti. Ah well.

Workstation cards don't even really utilize DP. DP is great for crunching numbers and scientific stuff, not so great for running CAD/CAM or rendering. That is why Nvidia segmented their professional VGA's. Quadro uses single precision and Tesla uses double precision. Thats how they force you to make 2 seperate costly purchases now. Fermi actually had a good balance of both.

[LiquidReactor]

FXAA is what most people use and maybe 2x (4x on older games) MSAA at high res like 2560x1440 or better, more isn't performant enough to be worth using (low fps no one would actually want to play at, and basically useless in terms of visual quality in motion anyway). I game at 2560x1440 110hz, many use 1080p 144hz or 2560x1440 and many of them with high refresh rates with how inexpensive 27" has gotten for PLS/IPS panels.

Nvidia has indeed pretty much cut Hawaii's show short I'd say I have to agree. Who on Earth is going to buy a $550 card that runs at 95c (consistently hotter/louder than Fermi), has little oc headroom, and is loud, when you can buy a $500 one with lots of oc headroom, generally the same performance when taking into account pre-heated benchmarks rather than short sprints before the 290X throttles (which happens during normal gaming, clocks are 800-900mhz not 1000mhz+), and a quiet heatsink (or custom designs for $510)?

290X has had tiny amounts of units come in-stock and sell off, that doesn't have anything to do with them selling well, just minute quantities. As he said, so much for tens of thousands...

Competition is good, but there's a clear and obvious choice for buying unless you have a huge favoritism towards one brand: at $500 ($510-520 with custom coolers) and barely any slower if at all during actual gaming except 4K res which is so tiny a market as to be nonexistent so far, while being quiet and cool, the GTX 780 post-pricedrop is that.

What $500 card right now can compete with 290x?

[[XC] Oj101]

Workstation cards don't even really utilize DP. DP is great for crunching numbers and scientific stuff, not so great for running CAD/CAM or rendering. That is why Nvidia segmented their professional VGA's. Quadro uses single precision and Tesla uses double precision. Thats how they force you to make 2 seperate costly purchases now. Fermi actually had a good balance of both.

Sorry, I should have specified GPGPU use. I wasn't even thinking about Quadro when I made that post.

Fermi had a good balance of both but can't hold a candle to the GTX TITAN for either.