
Review: Viable alternative to HD 5870 cards
Today we’ll be talking about Point of View’s GTX 470, a card that by now certainly needs no introduction. Both the GTX 470 and GTX 480 are aimed at high end market and are the HD 5800 series’ direct competitors. Nvidia launched the GTX 470 and 480 more than a month ago, and the cards are based on the largest and fastest Fermi GPU
We expected the GTX 480 to get the full version of the GF100 GPU but Nvidia thought differently and the GF100 ended up with one disabled Streaming Multiprocessor (SM). This means that out of 512 available stream processors (16 SM x 32 cores) or CUDA cores as Nvidia likes to call them, the GTX 480 lacks 32 to reach the full number. Going downstream to the GTX 470, this card has two disabled SMs and 448 stream processors in total. Naturally, this raises questions of what happened to the full version of GF100, but at this point we can only speculate
The GF100 is a complex chip that’s made of four Graphics Processing Clusters, four Raster Engines, six memory controllers, six ROP clusters and, as we’ve said before, 16 SMs. We’re not sure of exactly which SMs are disabled on the GTX, but we do know that the GTX 470 also comes without one ROP cluster containing 8 ROPs. At the same time, this means that the GTX 470’s memory bus is 320 bit, rather than 384-bit like on the GTX 480, which comes with all available ROP clusters. Each individual ROP cluster is assigned with one 64-bit memory controller
In order to make the performance difference evident, the usual practice is to lower operating clocks. This is the case with the GTX 470 where the GPU runs at 607MHz, compared to the GTX 480’s 701MHz. Stream processors on the GTX 400 cards run at 1401MHz and 1215MHz for the GTX 480 and GTX 470, respectively
The memory on the GTX 470 runs at 837MHz (3348MHz effectively). Coupled with the 320-bit memory bus, the card’s bandwidth totals at 133.9 GB/s
Nvidia opted on using GDDR5 memory, but the company did run into some problems with the speed. We see that the memory is clocked lower than on AMD’s high-end cards, but AMD uses 256-bit memory bus. As a result, the HD 5870 scores 153.6 GB/s with GDDR5 memory clocked at 4800MHz, whereas the GTX 480 (384-bit memory bus) is capable of 177.4 GB/s despite featuring memory clocked at 3696MHz
Point of View’s GTX 470 comes with 1280 MB of GDDR5 memory while the faster, GTX 480 packs 1536MB
The GF100 is a fast and powerful graphics processor, but cooling it on GTX 400 cards has proven to be quite tricky. Nvidia uses dual-slot cooling on both cards and while it does get the job done, both cards get pretty loud. The GTX 480’s TDP stands at 250W and the GTX 470’s at 215W. Idle consumption for the GTX 480 and 470 is 47W and 33W, respectively
Point of View GTX 470 is no different from the reference GTX 470, except for the stickers on the fan and the cooler. So, the card comes with reference dual slot cooling, which is a smaller version of the GTX 480’s cooler. To transfer the heat from the core to the heatsink, the cooler uses 5 heatpipes. The heatpipes are short and managed to fit beneath the plastic hood, so you won’t see them like on the GTX 480. The pictures below show the difference between the GTX 480 and GTX 470.




Point of View’s GTX 470 is 241mm long, which means it’s 26mm shorter than the GTX 480 and similar to the HD 5850.
A small 70mm fan is the same one used on the GTX 480. It’s a Delta fan rated at 1.8A. The fan, located at the far end of the graphics card, blows through the aluminum block and hot air leaves the case via the I/O panel’s outlets. In order to provide as much fresh air, the PCB features holes behind the fan. This trick was first done by Nvidia on its dual-GPU GTX 295 cards, and has now found it’s way to the GTX 400 cards

The GTX 470's cooler is made of four parts. The fan and the plastic hood used to direct air are visible but the plastic hides the third and the fourth component – a large aluminum block and metal block

Like on the GTX 480, our today’s card also features 5 heatpipes, but they’re much shorter and don’t exceed the GTX 470’s height. Since the heatpipes on this card are not visible, Nvidia decided not to “pimp” them the way they’ve done on the GTX 480.

The heatpipes are in direct contact with the GPU

The heatsink with the heatpipes cools the GPU only, while the memory is cooled by the metal block which extends the entire length of the card and cools other hot components

Nvidia’s Fermi GTX 400 series is in many respects the company’s first. Namely, this is the first time that the green team used GDDR5 memory on their high-end cards. The GTX 470 comes with 10 128MB GDDR5 memory chips (1280MB), whereas the GTX 480 comes with 12 memory chips (1536MB).

On their GTX 480, Nvidia uses 348-bit memory bus with memory clocked at 924MHz (3969MHz effectively), while the GTX 470 uses a 320-bit bus and memory at 837MHz (3348MHz effectively).

The GTX 470 comes with a 215W TDP and the card requires two 6-pin power connectors. In comparison, the GTX 480’s TDP is 250W and the card needs one 8-pin and one 6-pin connector Recommended PSU for the GTX 480 is 600W whereas for GTX 470 Nvidia recommends a 550W PSU. Power connectors are located on the top of the card

The I/O panel features two standard DVI outs and miniHDMI. We presume Nvidia opted on mini-HDMI to leave more space for ventilation, as the GPU can get pretty hot. The GTX 480 and GTX 470 feature the same I/O panel, as you can see from the picture

It’s worth noting that the GF100 supports DisplayPort, but Nvidia opted on HDMI as it’s much more common. However, if partners deem it appropriate, they can easily add DisplayPort out on the GTX 400 cards’ I/O panel
Nvidia included an HDMI sound device within the GPU, which means there’s no need to connect an external audio source to the graphics card when trying to bring both video and audio to your HD device via one cable only. In this respect, the GF100 features the same audio/video capabilities like the 40nm GT200 cards
Testbed:.
Motherboard: EVGA 3xSLI
CPU: Core i7 965 XE (3.6 GHz / 1.270 Volts, Intel EIST and Vdrop enabled).
Memory: 6GB Corsair Dominator 12800 7-7-7-24.
Power Supply: CoolerMaster Ultimete 1100W
Case: Corsair Obsidian 800D
Fan Controler: Kaze Master Pro 5.25"
Operating System: Win7 64-bit
197.41 WHQL
CCC 10.4.

Vantage
Vantage reports that GTX 480 is currently the fastest single-GPU card and it’s clearly faster than the GTX 470. The GTX 480’s advantage over the HD 5870 melts from 10.5% in Performance to 6.2% in High and ends up at about 3% in Vantage Extreme, but it still manages to beat the HD 5870 in every single test, albeit with a rather tight margin
In Vantage tests, the GTX 480 constanlty beats the GTX 470. Performance test reports 24.7%. better results, High settings report 26.6%, whereas Extreme settings result in a 27.2%. advantage, naturally in favor of the GTX 480.
Comparing the GTX 470 to the HD 5870 shows that the latter card packs more punch in more demanding tests. The GTX 470 ends up 12.7% slower in Vantage Performance, 19.3% at High settings and 23.4% at Extreme settings

Far Cry2.
In FarCry 2, Nvidia’s high-end GTX 400 series comes out on top and it’s even more evident when 8x antialiasing is turned on. FarCry2 results prove that by redesigning ROP subsystems in the GT100 chip, the company managed to improve compression and rendering efficiency as well as rendering smaller primitives that can’t be compressed. Throughput was also much improved compared to GT200 architecture, which resulted in overall better antialiasing performance and significantly smaller drops in performance when switching from 4xAA to 8xAA
Radeon HD 5870 ends up about 50% slower than the GTX 480 in both tested resolutions with 8xAA. The difference ducks to 39% at 1920x1200 and 4xAA and to 33.6% at 2560x1600, which proves that Nvidia’s GF100 is more efficient with 8xAA. With antialiasing turned off at 2560x1600, the performance difference ends up at only 15.5%.
Point of View’s GTX 470 outpaces the HD 5870 in few occasions, and the highest gap Nvidia’s ace made was at 1920x1200 and 8xAA – 17%. By increasing resolution and turning on antialiasing, the GTX 470 loses the fight versus the HD 5870.


Dirt 2 is one of the newer games that utilize certain parts of DirectX 11. At 1920x1200, the GTX 480 beats the HD 5870 by about 20%. This difference melts to about 7% at 2560x1600, with and without AA. Point of View GTX 470 fares well versus the HD 5870 at 1920x1200, but as soon as we hit 2560x1600, the HD 5870 proves that it’s faster. At 1920x1200 and 2560x1600 (no AA), the GTX 470 is slower by 1.6% and 15.9%, respectively. After turning antialiasing on, the difference melts to 3.4% and 16.7%, respectively


Overclocking
Point of View GTX 470 runs at 607/1215/837 MHz for the GPU/Shaders/Memory – meaning we’re talking about reference clocks. Nvidia strapped the GTX 470 with a dual-slot cooler, but the card easily hits 90° in 3D. The cooler is similar to the one used on the GTX 480, and it uses the same Delta 1.8A fan
The cooler is capable of keeping temperatures in check, but it does get pretty loud when the GPU heats-up. Still, it’s quieter than the one on the GTX 480. At reference clocks, maximum GPU temperatures didn’t exceed 93°C and the fan spun at 3400 RPM. In idle mode, Point of View’s GTX 470 ran pretty quiet
Overclocking always requires some kind of sacrifice – thermals, noise or both. In this case, we had to push the fan to maximum RPM, but the fan was unbearably loud. We didn’t meddle with voltages but managed to push the GPU from reference 607MHz to 700MHz. Truth be told, we expected more. We managed to hit 740MHz for the GPU, but the card didn’t run stable in all the tests. Overclocking the memory on GTX 400 cards is a bit tough, and we only managed to reach 880MHz (3520MHz, reference is 837MHz). Overclocking brought 13% better results in FarCry2.




