|
The
Naked Truth:::...
 |
|
Card
- Memory Sink
|
The
shroud removed we can see that heatpipes are used
only to transfer heat from the top bank of memory
chips and transport it to the larger heatsink on the
right which sits above the vertically mounted bank
of memory chips. Air from the fan is fed first through
the sink that sits over the GPU before passing though
the memory sink on its way past the power circuitry
and away into the bowels of your case..
 |
|
Card
- Shroud Removed
|
Here
you can see the actual component layout on the PCB.
 |
|
Nekkid
|
And
the star of the show, the NV40 GPU. As you can see
in the shot above, the core is pretty large, which
ultimately means expensive to produce. All those extra
transistors needed to deliver shader model 3 better
deliver sales or this could be another case of jumping
the gun costing NVIDIA money, just as it did when
they made the slightly premature move to 0.13microns
and DDR2 with the 5800.
 |
|
The
GPU
|
It
looked like a little running adjustment had been added
to our reference sample which I didn't really explore
much. I'm sure the retail boards will have a slightly
neater job done.
 |
|
Amended
|
The
card is kitted out with Samsung GDDR3 memory which
is only technically rated to 500MHz according to Samsung's
site, which means at 1.1GHz it's actually running
beyond spec on this reference board.
 |
|
Memory
Chip
|
Unless
I'm reading it wrong, the -GC20 chips are limited
to 500MHz (1GHz DDR), and we should actually be seeing
-CG16 fitted.
Samsung
Memory Data
If
you're feeling particularly geeky I'm sure you'll
be thrilled to hear that "The 8Mx32 GDDR3 is
268,435,456 bits of hyper synchronous data rate Dynamic
RAM organized as 4 x 2,097,152 words by 32 bits, fabricated
with SAMSUNG’s high performance CMOS technology. Synchronous
features with Data Strobe allow extremely high performance
up to 6.4GB/s/chip. I/O transactions are possible
on both edges of the clock cycle. Range of operating
frequencies, and programmable latencies allow the
device to be useful for a variety of high performance
memory system applications."
Time
to move on.
|
