New NVIDIA GF100 Board Surfaces, Suggests New High-End SKU

[RavenXXX2]

Sources in the hardware industry leaked some interesting pictures of a new, supposedly reference-design NVIDIA GF100 GPU graphics card PCB, watermarked by board partner Little Tiger. The pictures reveal a PCB that's similar to that of the GeForce GTX 480, but with a stronger VRM that makes use of better high-C surface-mount capacitors (completely doing away with cylindrical capacitors), and draws power from two 8-pin PCI-Express power inputs. The design can deliver up to 375W of power (that's not the board power we're talking about).

This also opens up speculation about what NVIDIA would do with this design. The most talked about theory as of now points to a new high-end SKU by NVIDIA based on the GF100, that enables all streaming multiprocessors (SMs) physically present on the GF100, taking the CUDA core count up to 512, and ROP count to 64. The most likely marketing name for this SKU is GeForce GTX 485. Apart from higher CUDA core and ROP count than that of the GTX 480, slightly higher clock-speeds for the GPU are also on the cards. The memory subsystem remains untouched, at 1536 MB of GDDR5 memory clocked at 924 MHz (effective 3.7 GHz), over a 384-bit wide memory interface. NVIDIA could release this SKU this fall.

http://www.techpowerup.com/125011/New_NVIDIA_GF100_Board_Surfaces_Suggests_New_High-End_SKU.html

 

[Marvin]

Should be a monster card. 'The Fermi that was meant to be.'

 

[Tute]

I think this was always on the cards, just a case of when nVidia could get the yields to a stage where it was financially viable to manufacture.

 

[R088ieS86]

Looks no way near as clustered as the old PCB, hopefully this will mean easier to cool.
How close to the 5970 will it get

 

[RavenXXX2]

Monster is the word, I doubt it's a reworked core so I imagine the power consumption and heat will be extreme. Probably some of the top parts they have that can make the full core grade, mass availability it will not be IMO, and $20 over a 480..yeah right, they are charging well over $80 for overclocked 480's.

 

[Tute]

Looks no way near as clustered as the old PCB, hopefully this will mean easier to cool.
How close to the 5970 will it get

I think it'll be damn close. Then nVidia will have a competitive (at least in terms of performance) lineup right across the range.

Question is - where do they go from here? Surely the die can't get any bigger?

 

[HW91]

this PCB is old, and it's 3rd party iirc

looks like a gigabyte board

clues to it being third party are that nvidia can't actually sell it if it has dual 8 pins, only 1x6pin and 1x8pin

 

[Phoenixdancer]

Looks no way near as clustered as the old PCB, hopefully this will mean easier to cool.
How close to the 5970 will it get

not very...

a 480 is only 5% faster then a 5870...

a 5970 is 2x 5870.. sooo miles off

 

[Rroff]

Looks no way near as clustered as the old PCB, hopefully this will mean easier to cool.
How close to the 5970 will it get

It needs the full 512SP at atleast 1600MHz to even be viably performance competitive to the 5970.

clues to it being third party are that nvidia can't actually sell it if it has dual 8 pins, only 1x6pin and 1x8pin

Not sure where you got that idea from - but as to the PCB yes its a japanese AIB partner development and not an nVidia reference design.

 

[CreAtiv3]

Wonder what heatsink + fan design will be used on this one as there is no cut out on the pcb...

 

[HW91]

It needs the full 512SP at atleast 1600MHz to even be viably performance competitive to the 5970.



Not sure where you got that idea from - but as to the PCB yes its a japanese AIB partner development and not an nVidia reference design.

300w (or PCIe slot 75w + 6 pin 75w + 8 pin 150w) is the PCIe max spec, major manufacturers/builders like Dell won't buy the cards unless they meet PCIe spec

 

[Monkeynut]

What is the advantage of these different capacitors? Less heat output? Better smoothing of current? Longer Lifespan?

 

[james.miller]

What is the advantage of these different capacitors? Less heat output? Better smoothing of current? Longer Lifespan?

they look more expensive

 

[Rroff]

300w (or PCIe slot 75w + 6 pin 75w + 8 pin 150w) is the PCIe max spec, major manufacturers/builders like Dell won't buy the cards unless they meet PCIe spec

As I said not sure where you got that from... the PCIe spec does not set a hard limit on power useage.

 

[Rroff]

What is the advantage of these different capacitors? Less heat output? Better smoothing of current? Longer Lifespan?

Better quality current characteristics, longer lifespan.

 

[Monkeynut]

Better quality current characteristics, longer lifespan.

Does that help with squeezing more out of the GPU?

 

[HW91]

As I said not sure where you got that from... the PCIe spec does not set a hard limit on power useage.

just going on info I get on XS, which during my time reading this forum, XS has always given far more reliable information than on the OCUK forums

 

[Rroff]

Does that help with squeezing more out of the GPU?

Marginally... the gigabyte boards using them get 4% lower power consumption compared to the reference boards

just going on info I get on XS, which during my time reading this forum, XS has always given far more reliable information than on the OCUK forums

I'm not an expert on the spec, but I'm fairly well versed on it and nowhere, unless theres been an addendum I've missed, does it specifiy any overall hard limits, theres regulation on the limits of what can be drawn from each individual PCI-e standard power connector and various advisories but the rest is left to a guide for thermal and electrical management in desktop systems and aslong as the developer can show they've taken that guide into consideration there should be no problem getting ratification.

 

[3t3P]

looks nice for a nekid pcb imo

 

[drunkenmaster]

Marginally... the gigabyte boards using them get 4% lower power consumption compared to the reference boards

.

4% honestly is within the margin of error for one thing, and asides from that, each card will have different power usage, a working core from the centre of the wafer is likely to have a little lower power usage than those from the outside. Same goes for the memory and even a different fan can make a few watts difference(especially if its a tiny delta going at silly rpm) though I have no idea what fan the gigabyte fancy edition uses.

Same way [H] had a Galaxy 470GTX that uses 50W more than a standard one, with only 18Mhz higher clock speeds which does not account for a 50W difference. Different fan, different actual gpu core, different batch of memory probably can all add or take a few watts here or there.

As for the card, it sounds more like a fancy normal overclocked 480gtx pcb than a 485gtx one, because its not an Nvidia reference card judging by the looks, and frankly the standard 480gtx pcb is fully capable of running a full 512shader card anyway, the power difference between a 480gtx with one shader disabled and one not, will not be very big at all. Theres only a 40-50W difference between a 480/470gtx, despite less memory and far lower core clocks on top of a shader cluster changed.