Home / Component / Graphics / Zotac GTX 480 Review

Zotac GTX 480 Review

First, a quick briefing on the card specifications when compared with the high end ‘last’ generation.

GTX 480
GTX 295
GTX 285
ROPs
48
2×28
32
Stream Processors
480
2×240
240
Texture Address/Filtering
60/60
2×80/80
80/80
Core Clock
700mhz
576mhz
648mhz
Shader Clock
1401mhz
1242mhz
1476mhz
Memory Clock
924mhz (3696mhz data rate) GDDR5
999mhz (1998mhz data rate) GDDR3
1242mhz (2484 mhz data rate) GDDR3
Memory Bus Width
384 bit
4×448 bit
512 bit
Frame Buffer
1.5GB
2x896MB
1GB
Transistor Count
3B
2×1.4B
1.4B
Manufacturing process
TSMC 40nm
TSMC 55nm
TSMC 55nm

The GTX 480 is based on GF100 which is the foremost family member. The GF100 is a 512 SP/core product running in a 4x16x32 configuration and the first series of cards do not have all the functional units enabled. The GTX 480 is a ‘480’ core product and the lower end GTX470 which we will address in a later article is a 448 core part. We would assume that a later model in the range will have all the functional units enabled.

The GTX480 has one of sixteen SMs disabled and it is worth pointing out that this disabled SM will affect the tessellation, texturing and shading performance of the card when compared with the ‘full GF100’ specifications, however it won’t affect the ROP’s which actually form a seperate functional section of the hardware.

Interesting we can see that nVidia haven’t pushed the GDDR5 to as high a level as many people expected initially, they are limiting it to around 3.7ghz when we have seen other configurations of hardware running at 5ghz. This means actually that the memory bandwidth of the GTX 480 is only around 10% more than the GTX 285.

nVidia have actually gone on record to state that their biggest limitation wasn’t the 384 bit bus but ‘deficiencies’ with their own I.O. controller which caused this apparent limitation in memory speeds. We can partly assume that AMD have a longer track record with GDDR5 and therefore are slightly more mature in their I/O development phase.

Comparing the Fermi hardware to the last generation is difficult because a new ‘front end’ has been created to deal with the inherant challenges of tessellation – focusing on more traditional pixel shading and geometry. This basically means that a direct comparison of specification sheets is a little more tricky than before.

nVidia physX support has been optimised to allow the GPU to efficiently switch between compute kernels and physics requirements with additional benefits being offered by executing multiple small physics kernels in parallel. Technically this means that performance hits would be less when changing between GPGPU instructions like GPU AI rendering and physX.

The card supports 2x, 4x and 8x Transparency Supersampling AA via the control panel. There is also a new 32x CSAA capability which many people will be salivating over.

We can look at rated power consumption, with nVidia quoting a maximum draw of 250W, in comparison the HD5870 requires 188W. The card is specified to idle at 47w. We will delve into this in greater detail later, but the figures are already rather concerning. nVidia recommend a 600watt power supply with one of these cards. The GPU thermal threshold is 105c and as we find out later, the temperatures can get high.

Another new feature is added support for Nvidia 3d Surround. This unites stereoscopic 3d vision technology with tri monitor configurations in a similar fashion to ATI’s Eyefinity. Unfortunately you are required to run two cards in SLI to get the three outputs required – nVidia’s cards have a 2 concurrent limitation.

On the GTX 480 there are twelve 128mb GDDR5 chips (Samsung K4G10325FE) giving a total of 1536MB of VRAM which means, finally nVidia have increased the reference memory on their boards.

The card is formed around a five nickel plated copper heatpipe solution with the 5th remaining within the PCB area and the other four protruding outwards. There is also a fully exposed heatsink on the front which means the appearance is immediately very distinctive. The shroud incidentally is easily removed giving full access to the heatsink area underneath. The PCB design allows for air from both front and back sides – this isn’t a new concept, it was already used on the dual GTX 295.

The GTX 480 measures 10.5 inches which is only slightly larger than the 10.45″ of the GTX 285. We would strongly suggest if you are buying one of these cards to use a chassis with good airflow as even nVidia recommend to space the cards out as far as possible in an SLI configuration. This is a fine idea however many motherboards do not offer three PCIe slots with x16 bandwidth to them all.

The power connectors are positioned on the top side of the card which also helps fitting into slightly more cramped conditions. The GFX needs 1×6 pin and 1×8 pin connectors to operate, pretty much as would be expected in this price range. Power is delivered to the card by a complex power system of six phases for the core and a single heavy duty phase for the memory. nVidia have opted not to use digitial power switching, a system used by AMD, however in theory the six chokes should generate less heat as they are spread out across the full PCB area.

The card layout is a single full slot design and they offer a mini HDMI out with a pair of DVI ports. You can only use two of these outputs at the same time, so you can’t power 3 displays. Many people might be immediately disheartened to see that there is no displayport option on the card at all. It is surely an omission that could lose a sale to a high end customer.

The card offers nVidia’s VP4 for video decoding (H.264/VC-1/MPEG-2/MPEG-4 ASP) with an internal passthrough for audio … the same as the GT200 series which was launched a while ago. AMD still have the edge as they offer DTS-HD and Dolby True HD and will also mean that the high end media center users may very well avoid the card entirely.

Become a Patron!

Check Also

Sapphire is reportedly selling RDNA 2-based crypto mining graphics cards

Sapphire appears to have begun selling dedicated crypto mining graphics cards based on AMD’s RDNA …