G96 Graphics Core

On paper, NVIDIA’s G96 GPU utilized in the GeForce 9500 GT looks awfully similar to the G84 GPU used in the GeForce 8600 series. Both GPUs feature the same number of stream processors (32), ROPs (8) and texture address and filtering units (16). Both chips also utilize a 128-bit memory interface to either GDDR3 or DDR2 memory. At first glance the only notable difference is the G96 GPU’s smaller manufacturing process.

However NVIDIA has incorporated several tweaks into their G9x GPUs in comparison to the G8x generation. For starters there’s PCI Express 2.0. PCIe 2.0 offers double the bandwidth of PCIe 1.1; 8.0GB/sec in each direction, providing a total of 16GB/sec of total bandwidth

In addition, G96 (like G92 before it), boasts improved color and z-compression over G84. This allows the GPU to make more efficient use of its available memory bandwidth, and should help the most at high resolutions, particularly once AA is applied.

Where did the rest of the extra transistors go? Besides the aforementioned improvements, G96 also adds support for DisplayPort displays. We’ve also been told that transistor counts can vary between each manufacturing process, so expect some variance there. All told G96 should offer clock-for-clock performance improvements over G84, and thanks to its smaller process the chip is cheaper for NVIDIA to produce, with lower power consumption as well. Here are the complete specs on G96, as well as its predecessors, G84 (GeForce 8600 series) and G86 (GeForce 8500 GT):

Mainstream GeForce Comparison
	GeForce 9500 GT 256MB/512MB GDDR3	GeForce 9500 GT 512MB DDR2	GeForce 8600 GTS	GeForce 8600 GT	GeForce 8500 GT
# of Stream Processors	32	32	32	32	16
Core Clock Speed	550MHz	550MHz	675MHz	540MHz	450MHz
Shader Clock Speed	1.4GHz	1.4GHz	1.45GHz	1.19GHz	900MHz
Texture Address/Filtering Units	16/16	16/16	16/16	16/16	8/8
Texture Fill-Rate	8.8 Gigatexels/sec	8.8 Gigatexels/sec	10.8 Gigatexels/sec	8.6 Gigatexels/sec	3.6 Gigatexels/sec
Memory Clock Speed	800MHz (1.6GHz effective)	500MHz (1.0GHz effective)	1.0GHz (2.0GHz effective)	700MHz (1.4GHz effective)	400MHz (800MHz effective)
Memory Interface	128-bit	128-bit	128-bit	128-bit	128-bit
Memory Bandwidth	25.6GB/sec	16GB/sec	32GB/sec	22.4GB/sec	12.8GB/sec
PCI Express Interface	2.0	2.0	1.1	1.1	1.1
Max Board Power Draw	50W	50W	71W	43W	N/A
HDCP	Yes	Yes	Yes	Yes	Optional

Notes

As you can see, NVIDIA is providing two different GeForce 9500 GT SKUs: one with GDDR3 memory, and one SKU with DDR2 RAM. The graphics core and shader clocks are the same on both SKUs, with the GDDR3 board shipping with faster memory.

NVIDIA has also unofficially developed a third reference design SKU with OC’ed clocks. This is the first time we can recall NVIDIA doing this. The so-called GeForce 9500 GT OC runs at 600MHz core while the stream processors operate at 1500MHz. The board’s memory is then clocked at 1.0GHz (2.0GHz effective), matching the clock speed of the GeForce 8600 GTS. This is the board NVIDIA provided for our testing today, although we also underclocked the board to the factory settings for our performance benchmarks as well.

Normally we’d be hesitant to test an OC’ed reference design (retail cards are obviously different), but considering the enormous overclocking results we obtained with our G96 sample, we’re actually a little surprised the OC’ed clock speeds aren’t the stock speeds. We managed to hit OC’ed speeds of 750MHz core, 1200MHz memory, and 1875MHz on the stream processors! Based on this, we’re pretty confident that all the usual suspects should have factory OC’ed GeForce 9500 GT cards, and some of them may actually go well beyond NVIDIA’s own GeForce 9500 GT OC.

Board analysis

NVIDIA’s GeForce 9500 GT reference board design is quite a departure from previous GeForce cards we’ve tested lately like the 9800 GTX+ and GTX 260/280. The card measures in with a diminutive 7” PCB and single-slot cooling. Since the board draws only 50W peak power consumption, no external power connector is necessary, instead the board draws its power solely from the PCIe interface.

For cooling the GPU, NVIDIA borrows the same heatsink/fan design that was first integrated on the GeForce 8600 GT. The cooling unit consists of a simple aluminum heatsink which is paired with a near-silent cooling fan.

As always with lower-end cards, NVIDIA’s board partners are free to make modifications for their own final 9500 GT cards, so we expect all kinds of derivatives to pop up over the course of the next few weeks. Already Palit has announced that they will be producing a passively cooled GeForce 9500 GT that runs at the stock GeForce 9500 GT clocks. Based on their history, ASUS, Gigabyte, and MSI will likely have passively cooled GeForce 9500 GT cards of their own shortly as well. These passively cooled 9500 GT cards would obviously be geared towards the HTPC crowd.

So now that you know more about the GeForce 9500 GT and its new G96 graphics core, how does it stack up against previous GeForce offerings as well as the competition in terms of performance? Let’s find out shall we?