• Portada
    • Recientes
    • Usuarios
    • Registrarse
    • Conectarse
    1. Foro
    2. ELP3
    3. Temas
    Desconectado
    • Perfil
    • Siguiendo 0
    • Seguidores 0
    • Temas 4
    • Mensajes 298
    • Grupos 0

    Temas

    • ELP3E

      [Review by ELP3] AMD Radeon FURY X

      Siguiendo Ignorando Programado Fijo Cerrado Movido Tarjetas Gráficas
      126
      0 Votos
      126 Mensajes
      40k Visitas
      HandroxH

      Friends, could someone explain to me what those 2 features/resources of DX12.1 that NV talks so much about are about?

      Conservative rasterization e Rasterizer-ordered views

      Thanks. ;D

    • ELP3E

      Test of real proof-volcanic islands-amd 290x/cf

      Siguiendo Ignorando Programado Fijo Cerrado Movido Tarjetas Gráficas
      193
      0 Votos
      193 Mensajes
      57k Visitas
      F
      @Handrox: Lo estoy intentando hacer funcionar, pero no va la mia es una EVGA GTX 680 reference… Parece que debe tener estos VRM NCP4206 , aunque creo que una de las Titan de ELP3 usa otro distinto y lo hizo funcionar. Aquí esta la información lo que yo no me entero bien de todo lo que se dice. [UPDATED for AB Beta 16] Fully unlocked NCP4206 Voltage / LLC mod tool si que a uno le dice paciencia con las WF. Luego hablando de las 290 y 290X parece que tienen problemas de mantener la frecuencia, a mi me sorprendio mucho que tuviera un perfil tan bajo de ventilador, pero necesitan o sacar personalizadas o mandar bios con un perfil mejor. http://techreport.com/forums/viewtopic.php?f=3&t=90345&sid=e0f934cd238ba8a97f7fa3d513da1430 saludos
    • ELP3E

      Test of the Real Nvidia Gtx Titan, single-sli-tri Sli, 4 WAY SLI

      Siguiendo Ignorando Programado Fijo Cerrado Movido Tarjetas Gráficas
      851
      0 Votos
      851 Mensajes
      218k Visitas
      F

      @wwwendigo:

      The move will not be repeated as with kepler, because for starters they have begun with the most basic chip "without cap", the GM107, and also using an "old" manufacturing process.

      Afterwards, the issue is much more interesting than it seems, seeing that although the jump to 20 nm would imply only a reduction in die size (approx. 50%), zero benefit in consumption (which I doubt) and zero benefit in frequencies (there I neither say yes nor no), even under these conditions NVIDIA has the ability to create quite impressive chips:

      GM107 in 28 nm-> 148 mm2
      GM107 in 20 nm-> 74 mm2

      Power offered by a GM107 vs a GK104 at full power (GTX 770) -> greater than or equal to 50% of the latter.

      And it's just a chip with one GPC, 128-bit bus not squeezed, and 640 SPs (40 TMUs). And 60w of consumption. A simple GM106, if it were as it is this one compared to the GK106 and GK107, would already be beastly (multiply by 3x the design of the GK107, in this case the GM107, to offer… a performance quite better than the GM104).

      I predict that there may be two possible routes in mid-range and mid-high range chips with maxwell:

      Route 1, the "standard" following the same expansion mechanics seen in kepler:

      GM107 1GPC 640 CUs 40 TMUs 16 ROPs 128 bits. 148 mm2 or 74 mm2 (20nm),
      GM106 3 GPCs 1920 CUs 120 TMUs 48 ROPs 384 bits. 225 mm2.
      GM104 4 GPCs 2560 CUs 160 TMUs 64 ROPs 512 bits. 300 mm2.

      This would be replicating what we see with kepler, but I believe that the issue of buses can be cut (and ROPs), perhaps for versions of 256-384 bits at most in the GM104. But power, it can be done (but also save transistors and very complex PCBs by removing some controllers and ROPs in favor of perhaps strengthening other elements, or simply reducing size and consumption).

      I believe that the simple GM106 should be substantially more powerful than the current GK104, by a very wide margin. This creates a doubt and a possible different route of launches:

      GM106-> launched as a substitute for the GK104, perhaps called GM104. Up to 50% faster. 180W as maximum.

      GM104-> product ¿intermediate? of high range perhaps close to the GM110 design, or a new "top" based on chips for the performance market. Perhaps renamed to something like GM102.

      GM100 ¿?¿? Whatever it is, possibly important changes with respect to what we have seen in maxwell today with the GM107.

      I, who have tested maxwell intensively with its introduction card, have to say that I have been impressed with what they have achieved with the same manufacturing node and with such a Spartan chip. The potential it has is wide. I believe that if it came out without SLI support, it is simply to avoid being devoured by the sales of all GK104-106 with relatively cheap multigpu configurations and, above all, Spartan in consumption.

      But I refer to the fact that they will use a GM204 for the 880 and 870 and that later they will release a GM210 or 200, which will possibly be the 980, that's what I refer to when I say that the 880 will be placed as high-end just as they did with the 680, I don't think they will put it at 300€.

      It is clear that they gain a lot in performance per watt even using the 28nm, which is why it seems that it will give them a margin of tranquility to release the mid-range and release high-end when the manufacturing process gives them more guarantees, besides that it seems that this efficiency allows them to do many things.

      It is that if we compare the GK107 it was only up to a 650, now this GM107 has already been placed as 750Ti, using the same process, but that they can do many things like the whole range is equally efficient and if they can they will not sell it cheap.

      It is also very possible that this GM107 in 28nm will become a GM207 or GM117 in 20nm, with more SPs and they will make the 850,850Ti, graphics without connector and with good performance.

      That which you put of triple CC that GM107 in GM106 plus 384 bits should already be a beast, it is that the GM107 with how little it consumes and 28nm already surprises, it is difficult to know where they will come out but I believe they have released the GM107 just to give notice of what maxwell means.

      regards

    • ELP3E

      Nvidia Geforce Titan (gk110)

      Siguiendo Ignorando Programado Fijo Cerrado Movido Tarjetas Gráficas
      303
      0 Votos
      303 Mensajes
      76k Visitas
      F
      @Bm4n: Seguro que la siguiente generación rendirá mas, siempre sacan algún mastodonte para dar la publicidad. Pero me parece mas interesante lo que comentas de la optimización, el problema de todas estas "novedades" dependiendo del éxito que tenga (o de si los motores gráficos se optimizan para ello) seguirán o no y por desgracia no se suelen aprovechar hasta generaciones posteriores. Y no se hasta que punto lo que comentas sería como poner a un pura sangre a tirar de un arado, parece que nvidia decidió sacar un procesador de su gama profesional para juegos y eso no tiene porque ser una idea fantástica porque sí, solamente que el rendimiento/precio se dispare tanto ya deja que pensar sobre todo porque estamos hablando del hardware que más rápidamente se devalúa. A mi personalmente son estas las tarjetas que mas me gustan,parecen tractores y la 480 o 580 ya eran una idea similar solo que con los antiguos SP,han sido y siguen siendo tarjetas bastante buenas para jugar,aunque quiza podian haber rendido mas si no tuvieran para GPGPU,lo que si es verdad es que la 480 es la tarjeta que mejor a movido un API nuevo,DX11 me refiero,yo recuerdo las 8800 Ultra atragantarse con Dx10 y las 480 no solo lo movian bien al salir sino que hoy dia se siguen defendiendo muy bien. Yo creo que con Titan ocurrira parecido,aunque a ver como influye la nueva generacion de consolas. saludos
    • 1 / 1