G80 Random Speculation Thread

[tieros]

Since nVidia doesn't own anyone at the Inquirer, it's up to us to "leak" the "facts" we "know" about the G80 in order to get equal airtime.

Here's my list of "facts":


65nm process for G80, 55nm for G81
360M transistors
1.2V core for GTX, 1.0V for GT variant
800MHz GPU reference clock for GTX, 700MHz for GT
16 Geometry/Vertex processors
48 Pixel processors
32 Texture management units, shared between G/VS and PS
SM 4.0
AF improvements, including angle-independent option
True HDR output (16 bits/channel) for upcoming HDR monitor support
gDDR4 support


I'd list some more "facts", but dinner is ready.

Anyone else have a friend that knows a friend who has a cousin that used to date this guy that knows about the G80? :D

[-=DVS=-]

Wishful thinking but sadly unlikely :p to have all that.

[MUYA]

Well it is said to be some hybrid version...what that means I dunno, some fixed function piepline and some unified? More maths capabilities in each shader unit? HDR+AA? New AA method? PPP? XDR memory? increased memory interface 512 bits? ondie e-dram thinger?

[Mr_LoL]

1.2Vcore is a bit low if its going to run at 800mhz isnt it?

[jAkUp]

I don't think we will se it having a 512bit bus, HDR+AA is almost certain, DX10 is as well.

1.2Vcore is a bit low if its going to run at 800mhz isnt it?


It's 65nm though ;)

[Goran]

Embedded DRAM is also unlikely Ithink since it is very expensive. Bitboys, anyone?

[Muppet]

Specs, i don't know but it'll be fast ;)

[jolle]

Embedded DRAM is also unlikely Ithink since it is very expensive. Bitboys, anyone?
hehe yeah, 12mb on a 1024bit bus.. bit ahead of their time perhaps.

[tieros]

The "hybrid" part is that they are combining the geometry and vertex processors, since they are so close in nature. The geometry shaders can really be thought of as just vertex pre-processors. But they are leaving the pixel processors as separate hardware, so they can retain the current hardware optimizations for pixel processing, and add new ones. It will also let them set a much higher clock domain for the pixel shader portion of the GPU.

The other part of the hybrid equation is the separation of the texture management units from the shaders. They will be shared between the different shaders, and that section of the chip will be clocked in sync with the vRAM at a ratio that makes latency hiding more predictable and easier to manage.

By not going the fully unified route like the R600, nVidia will be able to save an additional 25-35M transistors that would be needed for a scheduler in a unified architecture.


Wheee, this is fun. What else should the chip do? :D

[Marvel_us]

Most of the stuff listed seems pretty doable.

What are the known speeds for GDDR4?

[Zelda_fan]

what is an HDR monitor?

[tieros]

http://www.brightsidetech.com/

Their approach is pretty cool, but I think the first company that gets to per-pixel backlighting control wins. Or maybe it'll be something like FED monitors.

HDR rendering is just a gimmicky special effect until you can push 16 bits/channel all the way to the display and have them displayed properly. Current LDR monitors are barely capable of 8 bit/channel output.

[coldpower27]

Since nVidia doesn't own anyone at the Inquirer, it's up to us to "leak" the "facts" we "know" about the G80 in order to get equal airtime.

Here's my list of "facts":


65nm process for G80, 55nm for G81
360M transistors
1.2V core for GTX, 1.0V for GT variant
800MHz GPU reference clock for GTX, 700MHz for GT
16 Geometry/Vertex processors
48 Pixel processors
32 Texture management units, shared between G/VS and PS
SM 4.0
AF improvements, including angle-independent option
True HDR output (16 bits/channel) for upcoming HDR monitor support
gDDR4 support


I'd list some more "facts", but dinner is ready.

Anyone else have a friend that knows a friend who has a cousin that used to date this guy that knows about the G80? :D

Depends if G80 is to come out this year I would say it would debut on the 80nm process which would be a mature product by the time they release this series, so low-risk. Then G81 would be the 65nm shrink. Considering that Nvidia has 0 65nm products now let alone any 80nm products I would say 80nm for the first revision is realistic.

The clock rates used for 800MHZ vs 700MHZ is too small for product segmentation unless you have some disabled functionality of the die.

[hemmy]

http://www.neoseeker.com/news/story/5116/

[retsam]

since we are all jumping on that yearly speculation bandwaggon, i going with a clockless async chip :retard:

[sabersix1]

Well I am going to say that the High End 8xxx GTX will have 1GB of GDDR4 .6ns RAM, and a Laser-Cut GPU made with the new Jizm-Meltdown .65 Fabrication Process. LOL It's gonna have a Turbo-Blower on a Peltier Themonuclear Heat Sink and it will Run at 1000Mhz Core/ 2.8 GHz Memory Clock Rates.

[walterman]

I don't think we will se it having a 512bit bus, HDR+AA is almost certain, DX10 is as well.




It's 65nm though ;)

Totally agree, a 512bit memory bus will make the PCB ultra expensive & complicated.