Test of the Real Nvidia Gtx Titan, single-sli-tri Sli, 4 WAY SLI
-
@PGS:
Come on, this is what Nvidia has been doing for years and years. What now affects your favorite company and it's no longer good for them to send engineers to "help" optimize games for their hardware, which used to be a vision of the future and great for the user, is now terrible.
This has always happened, ever since hardware companies started sponsoring software companies and vice versa, and it's always bad for the end user.
Sorry, but you reserve the customizations in the comment and the "your favorite company" thing for yourself, the tone is fair, we already know that it starts with pointing fingers and accusing of bias and ends up another way. I have proof of AMD's bad practices in several cases of "optimizations", and in a way that has never been seen in Nvidia, whose bad practices, if they have focused on something, it's blocking features in their hardware (proprietary use of innovations).
The worst case in recent times from Nvidia was with Crysis 2, with the issue of the "tessellation sea", something that barely burdened GPUs anyway (what really burdened it was the excessive tessellation detail in trivial details of sidewalks and similar, that did have some weight and was really questionable in its usefulness, the tessellation sea... an absurdity in comparison). But cases like HAWX2 or Lost Planet 2 with their tessellation were more than justified, another matter is that AMD's hardware had and has some pitiful tesselators, which have largely delayed the adoption of tessellation, the main innovative feature of DX11, for many titles.
Discrediting campaigns have been carried out against the use of "massive" tessellation by AMD (you'll find it reasonable that if I said "your favorite brand" it wouldn't make sense, right? I'm just telling you as an example), saying real absurdities like it's totally unnecessary to use triangles of 16 pixels or less area, when this doesn't make any sense (a triangle of 16 pixels has approximate dimensions of 6x6 pixels, more than visible with borders included on a current screen, generating smaller triangles still makes sense, even).
But there are implementations of effects that have been done in AMD that don't make any sense. It doesn't make any sense to use SSAA and add any other AA technology on top, it's ridiculous. It also doesn't make sense that the performance drops as it does with some effects totally similar to those implemented in non-GamingEvolved games or even TWIMTBP that perform adequately on Nvidia (and sometimes better on AMD).
It's clear, if a game comes out with a built-in benchmark in Gaming Evolved it's because it has been clearly manipulated to offer artificial results. This has been fulfilled in the last Gaming Evolved with benchmarks, all basically perform poorly on Nvidia. The reason for including benchmarks in such influenced games is simple, it's to affect review results since in these reviews they tend to adopt games with built-in benchmarks as they are convenient for reviewers. Then we have games that don't have built-in benchmarks but are considered very important at a technical level, like Crysis 3 or Far Cry3, which curiously perform very well on Nvidia, so we would have to talk about 2 levels of Gaming Evolved games, those that put AMD's performance at the service of the end user at the expense of quality, and those where the Gaming Evolved relationship is not posed with such servility, but as a plus for AMD and the developer to test and promote both products together.
There's a big difference between implementing a feature massively but improving graphic quality (LP2, HAWX 2) and implementing a feature in such a way that it sinks the performance on the rival even at the cost of losing something on their own platform (but much less).
The best proof of this can be found in the spontaneous improvement of Sniper Elite V2, a great game that is nevertheless burdened by a painful implementation of certain effects by GPGPU. I had this game in my sights because I already knew what was burdening the performance specifically in origin, with Nvidia. I'll dedicate a post to it on the blog and if necessary I'll leave some of the images here of the results because they have substance.
It's a very clear case of inserting features for an impressive burden on the rival. That wasn't even seen in the mentioned case of Crysis 2 with its "tessellation sea". And if Sniper V2 were an isolated case, I would let it go, but there are already several very similar cases. And it seems that Tomb is another case to add to the list.
This way I also gain reviews by 5-8%... :troll:
-
Let's see.. this is especially for maldo or wendi, since you know more about this than the rest of the mortals..
In Nvidia, at least in my case with TITAN, shortly after starting the game, you can no longer play with the "fregona" hair and Super Sampling activated because this happens:
The hair stops being hair and becomes a blurry spot that comes and goes… also watch out for the Vram consumed according to afterburner at my resolution and 4 SSAA, as much or more or less truth... let's see who has a graphics card with more than 4GB of Vram and also with the necessary power to run it... AMD certainly not..
However with FXAA, the "fregona" hair looks "normal"
Do you know if this is only with Nvidia, or does it also happen with AMD, some kind of incompatibility?
In both cases the rest of the options are at maximum.
Regards.
-
Let's see.. this is especially for maldo or wendi, since you know more about this than the rest of the mortals..
In Nvidia, at least in my case with TITAN, after a short time starting the game, you can no longer play with the "rag" hair and Super Sampling activated because this happens:
The hair stops being hair and becomes a blurry spot that comes and goes… also watch out for the Vram consumed according to afterburner at my resolution and 4 SSAA, as much as it's more or less true... let's see who has a graphics card with more than 4GB of Vram and also with the necessary power to run it... AMD certainly not..
However with FXAA, the "rag" hair looks "normal"
Do you know if this is only with Nvidia, or does it also happen with AMD, some kind of incompatibility?
In both cases the rest of the options are at maximum.
Regards.
As you say it's the vram and a bug in the game. Surely if you leave SSAAx4 but lower the resolution to 1080p the hair will already look good.
-
So what's going on, that I can't play at 1600P with SSAA? if I have 6 GB VRAM...I don't understand..
Well Maldo...I just did it at 1200p, and the same thing keeps happening:
-
Vamos a ver..esto vá especialmente para maldo o wendi,ya que sabeis mas de esto que el resto de los mortales..
En Nvidia,al menos mi caso con TITAN,al poco de empezar la partida,ya no se puede jugar con el pelo "fregona" y el Super Sampling activado pues ocurre esto:
El pelo deja de ser pelo y se convierte en una mancha borrosa que vá y viene…ojito también a la Vram consumida según afterburner a mi resolución y 4 SSAA,como sea mas o menos verdad...a ver quien tiene una gráfica con mas de 4GB de Vram y además con la potencia necesaria para moverlo...AMD desde luego no..
Sin embargo con FXAA,el pelo "fregona" se vé "normal"
Sabeis si esto es solo con Nvidia,o sé produce también con AMD,algún tipo de incompatibilidad?
En ambos casos el resto de opciones están la máximo.
Salu2.
A ver, ya he tenido mi primera toma de contacto sobre el rendimiento del juego con sus distintas opciones, y eso que nos muestras es muy anómalo, para empezar:
Supongo que activas el SSAA dentro del propio juego, por alguna razón extraña éste no me funciona con los 314.14 activándolo desde el menú de configuración del lanzador del juego.
Una vez activado el SSAA 4x, observo una anomalía muy fuerte y "similar" a la tuya en que afecta al pelo, pero de una manera distinta. Como tengo una kepler "pequeña" las diferencias pueden ser explicadas por las diferencias entre modelos, empezando por tener 2 GB de VRAM. En mi caso con 2x SSAA y todo en ultimate el test se comporta normalmente consumiendo unos 1,8GB de VRAM, pero al activar 4x SSAA, además de que el rendimiento se hunde mucho más de lo esperable, se pasa a consumir unos 1,5GB de VRAM, claramente imposible con estas codiciones. O sea, que no es que agote la memoria VRAM sólamente, es que además hace algo muy raro para de pronto dejar un hueco de más de 512 MB de VRAM sin usar, dañando más al rendimiento.
En tu caso la VRAM no es un problema, pero sin embargo tienes ese problema de emborronado, yo con el tema del 4xSSAA también he visto un artefacto gráfico con el pelo, y es que parpadeaba en tonos amarillentos el cabello de Lara cada poco (¿a punto de entrar en modo superguerrero?).
Son fallos extrañísimos, así que diré sólo qué se supone que deberían estar haciendo en ese momento las gráficas respecto al cabello:
Entrar en modo GPGPU, cargar los datos pertinentes y correr los computing shaders de simulación de cabellos (listas linkadas), calcular su posición en el momento dado de simulación del juego y guardar esos datos o en forma de una malla 3D, o incluso pasar de esto y calcular ya en este modo la proyección del cabello en la imagen 2D (e información de sombreado), y pasar a través de un buffer la información del cabello al modo 3D de la gráfica (vuelta a entrar en este modo) para componer la imagen final.
Sea como sea, hay un momento donde se renderiza el cabello, y que se guarda en forma de textura/buffer para uso posterior. El parpadeo es una anomalía muy importante, pero más es el "emborronado", que quizás apoya más a un posible uso de un buffer de imagen de los cabellos. Quizás se calcula y "renderiza" el cabello sin texturizar, y en mi caso debido a la falta de VRAM ocurre que simplemente se muestra a veces sin renderizar totalmente éste, con un color "plano" del buffer pasado desde el modo GPGPU (es posible que haya otros buffers para información adicional como info de brillos, etc), y en tu caso con VRAM de sobra el problema se transforma en algún problema bastante grave de cómo se implementa la imagen final renderizada del cabello al resto de la imagen (parece que han aplicado algún tipo de reescalado de esa imagen prerenderizada de baja calidad para el cálculo de la imagen a la resolución interna del SSAA, y después sufre un downsampling que acaba de joder la marrana).
No tengo ni idea de si pasa en las gráficas AMD, pero parece un problema de cómo se gestionan los pasos de renderizado del cabello en uso del SSAA 4x.
Y por cierto, ¿tú ves el campo de fuerza que cubre a Lara?, sí, el mismo que evita que el cabello se apoye de ninguna de las maneras en sus hombros o su espalda, dejando unos 3-4 dedos de distancia siempre entre cabello y cuerpo.
Está hecha toda una Jedi. Muy profesional el TressFX (Manquiña dixit).
–-----
PD: Qué curioso, al mirar mejor las capturas de ELP3 veo que la "mancha" es bastante simétrica pero NO cubre todo el cabello (se ven las puntas, etc, fuera de la mancha). Tiene un curioso colorido con zonas más claras y otras oscuras, ¿será un mapeado de sombreado/iluminación del cabello?. Pero debería usarse para calcular el valor de color/brillo de cada cabello final, no para aparecer así encima del mismo. :ugly:
Ni idea de qué pasa exactamente, pero vamos, bug gordo sin duda.
-
So what's going on, that I can't play at 1600P with SSAA? if I have 6 GB VRAM...I don't understand..
Well Maldo...I just did it at 1200p, and the same thing keeps happening:
Well that's funny. I can set it to 1200p with SSAAx2 and the hair looks good. I haven't tried more because the performance was no longer constant 60 and I can't play. Does SSAAx2 do it to you too? Yes I've seen other users that even with FXAA if they increased the resolution above 1600p they saw the hair badly.
-
Let's see, I've already had my first contact with the game's performance with its different options, and what you show us is very anomalous, to begin with:
I suppose you activate SSAA within the game itself, for some strange reason it doesn't work for me with the 314.14 drivers when activating it from the game launcher's configuration menu.
Once SSAA 4x is activated, I observe a very strong anomaly "similar" to yours in that it affects the hair, but in a different way. As I have a "small" Kepler, the differences can be explained by the differences between models, starting with having 2 GB of VRAM. In my case with 2x SSAA and everything in ultimate the test behaves normally consuming about 1.8GB of VRAM, but when activating 4x SSAA, besides the performance dropping much more than expected, it goes to consume about 1.5GB of VRAM, clearly impossible with these conditions. That is, it's not just that it exhausts the VRAM, it's also that it does something very strange to suddenly leave a gap of more than 512 MB of VRAM unused, harming the performance even more.
In your case VRAM is not a problem, but you still have that problem of blurring, I with the 4xSSAA issue have also seen a graphic artifact with the hair, and it's that Lara's hair flickers in yellowish tones every little while (about to enter super-soldier mode?).
These are very strange bugs, so I'll only say what the graphics are supposed to be doing at that moment regarding the hair:
Enter GPGPU mode, load the relevant data and run the computing shaders for hair simulation (linked lists), calculate its position at the given moment of game simulation and save that data either in the form of a 3D mesh, or even go beyond this and calculate the projection of the hair in the 2D image (and shading information) already in this mode, and pass through a buffer the hair information to the 3D mode of the graphics (back into this mode) to compose the final image.
Anyway, there is a moment when the hair is rendered, and it is saved in the form of a texture/buffer for later use. The flicker is a very important anomaly, but more so is the "blurring", which perhaps supports more a possible use of a hair image buffer. Perhaps the hair is calculated and "rendered" without texturing, and in my case due to the lack of VRAM it happens that it is sometimes shown without being fully rendered, with a "flat" color of the buffer passed from the GPGPU mode (it is possible that there are other buffers for additional information such as shine info, etc), and in your case with plenty of VRAM the problem turns into a pretty serious issue of how the final rendered hair image is implemented in the rest of the image (it seems that they have applied some kind of rescaling of that low-quality pre-rendered image for the calculation of the image at the SSAA internal resolution, and then it suffers a downsampling that just messes up the whole thing).
I have no idea if this happens on AMD graphics, but it seems to be a problem of how the hair rendering steps are managed when using SSAA 4x.
And by the way, do you see the force field covering Lara?, yes, the same one that prevents the hair from resting in any way on her shoulders or her back, leaving about 3-4 fingers of distance always between hair and body.
She's made a whole Jedi. Very professional TressFX (Manquiña dixit).
–-----
PD: How curious, upon looking better at the ELP3 screenshots I see that the "stain" is quite symmetrical but does NOT cover all the hair (the tips, etc., are seen outside the stain). It has a curious coloring with lighter and darker areas, will it be a hair shading/lighting map? But it should be used to calculate the final color/brightness value of each hair, not to appear like this on top of it. :ugly:
No idea what exactly is happening, but it's definitely a big bug.
That ELP3 bug and that other users with very large resolutions have is very likely posterior to the hair position calculation. When all calculations are done, there are post-processing passes to smooth the hair by applying anti-aliasing to each one. That post-processing is what I think is not well-sized for high resolutions and that's why it becomes blurry.
-
That ELP3 bug and that other users have with very high resolutions I think is much later than the calculation of the hair position. When all the calculations are done, there are some post-processing passes to smooth the hair by applying antialiasing to each one. That post-processing is what I think is not well-sized for high resolutions and that's why it becomes blurry.
Well, what can I say, it seems too blurry to me because of a bad application of a post-processing that probably just applies FXAA or little more. What surprises me is the symmetry of the final blur, and that it doesn't affect the hair tips at all, which makes me doubt about the post-processing or hair scaling issue (after all, supersampling consists of an upsampling and subsequent downsampling of the image). If that were the case, all the hair would have to be blurry, not almost all but not the tips.
Besides, the blurs would fit quite well in their tones with what would be the lighting to apply to the hair.
But whatever it is, it's a matter of how the issue has been done. Fine and good, "hoyga".
PD: By the way, it's curious that the scene of the integrated benchmark of the game, which occurs at the beginning just after leaving the cave, in benchmark it gives me with ultimate about 41-42 fps, but in the real game scene it runs around 60-70 fps with ease.
It will be post-processing issues, for sure… ¬¬
-
As it is related to the resolution, I repeat that I do not think it is anything related to physical calculations. ManuelG from nvidia says that they have already seen the error and are looking into it.
-
Sorry, but you reserve the customizations in the comment and the "your favorite company" for yourself, the tone is fair, bla, bla, bla, bla, ….
Nvidia and AMD have cheated whenever they could, you're not going to sell me the story at this point that AMD is doing something unprecedented. By the way, I haven't used any special tone, so you'd better save the superior attitude you're putting on.
Optimizations and deoptimizations are a mess whoever does them, but when they started with the "meant to be played" many of us warned that things would end up like this and many people denied it and even said that this was good for the user because their cards would work better and that the fault was with ATI for not doing the same… well here we have the briefcase war and the "superbenefited" users.
-
As it is related to the resolution, I repeat that I do not think it is anything related to physical calculations. ManuelG from nvidia says that they have already seen the bug and are looking into it.
It's just that I don't think it's resolution.
As I say, at 1200p with SSAA X4 it also does it... and 1200p, I don't think it's a super special resolution...
Best regards.
-
It's just that I don't think it's resolution.
Like I said, at 1200p with SSAA X4 it also does it... and 1200p, I don't think it's a super special resolution...
Best regards.
1200p SSAAx4 is a lot of resolution. Supersampling multiplies the resolution, it's not MSAA.
1920x1200 SSAAx2 would be like 2880x1600
1920x1200 SSAAx4 would be like 3840x24002560x1600 SSAAx2 would be like 3620x2260
2560x1600 SSAAx4 would be like 5120x3200But I'm not saying it's special because it consumes a lot of performance, that you would have more than enough, it's just that there's a bug that from a certain resolution the hair doesn't look good. As soon as they fix it you'll be fine.
-
@PGS:
Nvidia and AMD have cheated whenever they could, you're not going to sell me the story at this point that AMD is doing something unprecedented. By the way, I haven't used any special tone, so you'd better save the superior attitude you're using.
Optimizations and de-optimizations are a mess whoever does them, but when they started with the "meant to be played" many of us warned that this would end up like this and many people denied it and even said that this was good for the user because their cards would work better and that the blame was on ATI for not doing the same... well here we have the briefcase war and the users "super benefited".
What was left to read. Sorry, but the "superiority tone" is the one you started using when saying "your favorite company" making what is a thinly veiled accusation that is not relevant (as far as I know, you are the first to mention others and with that accusation), so first respect, don't come with those "tones" and then if you demand that from others, don't come with tones either.
And by the way, with the TWIMTBP plan, AMD has never been seen to have a game perform 50% less on AMD than on Nvidia with similar cards. And if you want an example of the mess AMD has done, look at the performance with Dirt Showdown, with "optimizations" activated and what happens to the performance.
AMD has never needed special help to have performance problems, in fact they have had them with Gaming Evolved programs like Dragon Age 2, and some more cases, which are the games that have gone worst for GCN graphics vs Kepler.
Now you'll tell me that this was an "Nvidia boycott". I insist that Nvidia, their "specialty" in these matters more than sinking rival performance has been to implement proprietary features, like PhysX, "exclusive" AA or the use of techniques that suit them.
Of course there have been cases of "boycott", but they are the minority compared to exclusive optimizations. And I don't know what a simple "you too" comes to when we are talking about the current situation caused by AMD in current games, not what panchito and benito did 2 days ago.
-
1200p SSAAx4 is a lot of resolution. Supersampling multiplies the resolution, it's not MSAA.
1920x1200 SSAAx2 would be like 2880x1600
1920x1200 SSAAx4 would be like 3840x24002560x1600 SSAAx2 would be like 3620x2260
2560x1600 SSAAx4 would be like 5120x3200But I'm not saying it's special because it consumes a lot of performance, that you would have more than enough, it's that there's a bug that from a certain resolution the hair doesn't look good. As soon as they fix it you'll be fine.
I also have the same issue with 1080p and 1200p X2 SSAA... a bug... but from Nvidia, the game or both?
By the way, the strings from Crysis 3, until there's a patch, nothing, right?
A slaudo.
-
@PGS:
Nvidia and AMD have cheated whenever they could, you're not going to sell me the story at this point that AMD is doing something unprecedented. By the way, I haven't used any special tone, so you'd better save the superior attitude you're putting on.
Optimizations and deoptimizations are a mess whoever does them, but when they started with the "meant to be played" many of us already warned that things would end up like this and many people denied it and even said that this was good for the user because their cards would work better and that the fault was with ATI for not doing the same... well here we have the briefcase war and the "superbenefited" users.
I can guarantee you that those of us following that thread are not very maniacal and we don't care about Ati or Nvidia and especially for a long time I've been reading Elp3 and he doesn't care either, he's always said what he thought.
Best regards
-
Well... it's official:
http://www.3dmark.com/hall-of-fame/3dmark-11-top-extreme-preset/
After the untouchable Kingpin, it's my turn...
I had to do it for the balls with the 2011... I don't like this platform at all, it has less performance than ivy, and besides, the micro can no longer go above 5.0GHZ... but the 12 threads do a lot in physics tests and I had no choice but to use it.
Of course the graphics are stock, air-cooled and without using voltage... why? if it's not useful for the moment...
I hope you like it... ;)
Best regards.
-
wOw Congratulations friend it looks super good ;D, I join the group of the F5f5f5f5f5f5.
-
Bueno…ya es oficial:
http://www.3dmark.com/hall-of-fame/3dmark-11-top-extreme-preset/
Después del intocable Kingpin,voy yo..
Lo he tenido que hacer por huevos con el 2011..no me gusta nada esta plataforma,tiene menos rendimiento que el ivy,y además,el micro ya no es capaz de subir de 5,0GHZ..pero los 12 hilos hacen mucho en las pruebas de física y no me ha quedado mas remedio que utilizarla.
Por supuesto las gráficas están de stock,por aire y sin utilizar voltaje..para qué? si no sirve de momento..
Espero que os guste..;)
Un saludo.
Enhorabuena.
La que me parece muy buena puntuacion es la de Hazzan,pues son 680,aunque esos van con LN2 y es muy distinto a hacerlo por aire,el LN2 te asegura mas frecuencia tanto en CPU como a las GPU y en esos benchmark se nota.
saludos
-
I won't have the Titan until next week :wall:
While I wait, see if you can resolve some doubts for me:
- Drivers, beta or WHQL?
- For overclocking, I see that EVGA Precision is used, I was used to MSI Afterburner, is it for some reason?
- The monitor's overdrive, the function to give more than 60Hz, has anyone tried it?
- As for overclocking, from what I read, it's best to leave the fan on auto, and link Power and Temperature to the top, without needing to touch the voltage, right? Also, I come from some GTX 580s, they didn't have Boost, so I'm not very clear if I need to touch the offset or if it's not necessary to touch it and just raising the base speed is enough.
Thanks in advance.
P.D. What a wild score, and what a pleasure it is to see it there, I remember when I put a score among the top ten years ago the joy it gave me....
-
Bueno…ya es oficial:
http://www.3dmark.com/hall-of-fame/3dmark-11-top-extreme-preset/
Después del intocable Kingpin,voy yo..
Lo he tenido que hacer por huevos con el 2011..no me gusta nada esta plataforma,tiene menos rendimiento que el ivy,y además,el micro ya no es capaz de subir de 5,0GHZ..pero los 12 hilos hacen mucho en las pruebas de física y no me ha quedado mas remedio que utilizarla.
Por supuesto las gráficas están de stock,por aire y sin utilizar voltaje..para qué? si no sirve de momento..
Espero que os guste..;)
Un saludo.
Enhorabuena ;D
