Key Specifications
- Review Price: £319.99
The idea that one IT company’s fortunes blossom while those of another dwindle is an ancient one in the business. Trends change, technologies advance, and some out-of-the-box, novel ideas succeed while others fail. Therefore, it is not surprising that the two largest graphic card makers of the past ten years have experienced ups and downs in their business. Thanks to its Radeon HD 5xx0 family of graphics cards, ATI has recently seen success. Due to their class-leading performance, features, and power consumption, as well as the fact that they are the only DirectX 11 compliant cards on the market, they have been the obvious choice for just under six months. Finally, Nvidia has introduced the GTX 480 and GTX 470, two cards based on its most recent chip technology known as Fermi, so ATI doesn’t have the DirectX 11 party to themselves.
On paper, both of these cards appear to be on track to elevate performance to a new level. The GTX 480 features twice as many stream processors as the GTX 285 top-of-the-line card from Nvidia, while the GTX 470 has 448. These cards appear to have everything they need to catch up to or surpass the best ATI has to offer, including an improvement in memory from GDDR3 to GDDR5, as well as a wide range of architectural upgrades.
Techpotamus typically examines the flagship component of a new line of graphics cards first and use that time to examine the underlying architecture. But in this review, we’ll actually be focusing on the slower GTX 470 card because Nvidia only had a few review copies available. However, we’ll still take a close look at the general Fermi architecture, which will run this card and all of Nvidia’s lineup for the foreseeable future.
Therefore, the processors inside the GTX 480, GTX 470, and future Nvidia GPUs will be designed on the Fermi overarching architecture. It shares some of the fundamental components of the most recent Nvidia design generations, but many of them have been rethought in light of DirectX 11’s requirements.
Starting with what is constant, the CUDA Core (Core), or Stream Processor as it was once known, continues to be the fundamental building element of Fermi. This little processor serves as the fundamental number-crunching unit that does all the calculations necessary to generate the beautiful graphics in your games or to process data for other GPU-accelerated tasks like video encoding and ray tracing. Moving up a level, things appear to be somewhat similar but are actually rather different.
These Cores were grouped together into eight-core groups in G80 and GT200, the chips that power the 8800GTX and GTX 280, respectively. This configuration is known as a Streaming Multiprocessor (SM). The Texture/Processor Cluster (TPC) was located above this and added texture units and more memory to the process. But with Fermi, an SM now has 32 Cores, four texture units, and something called the Polymorph engine, which we need to describe from scratch.
Contrary to popular belief, a graphics card does not do all of the rendering tasks for 3D scenes on your computer. The wire frame model is actually created by the CPU, and all the fancy effects you see are then applied to it. These wireframes must be kept very simple in order to retain a respectable degree of speed, though, as a CPU is doing so many other things simultaneously, such as AI, physics, and animation. Due to the computationally intensive nature of creating all the triangles needed to accurately represent the complex surfaces of a realistic world, despite recent advancements in graphics, you still see characters in games with pointy heads and corrugated iron that is completely flat when viewed up close.
Transferring some of the labor of building a scene’s fundamental geometry to the GPU is part of the solution. Tessellation and displacement mapping, two fundamental techniques that make their debut for DirectX-based games with the new DirectX11 API, are used to accomplish this.
(center)No Tessellation – Tessellated – Displacement Mapped(/center)
A simple wireframe model’s vertices are simply filled in with tessellation to provide a far more realistic, smooth surface. It merely brings the model closer to having a more natural appearance rather than adding more detail.
A displacement map, on the other hand, is a texture (2D picture) that represents height information and is used to change the model’s vertices’ relative positions when applied. This enhances the model with all the minor nuances that give it actual life. As a result, 3D models are infinitely more realistic and require less additional graphical tricks to make them appear realistic. Other graphical effects, such as applying shadows, are considerably enhanced because they follow the accurate lines of the complex model rather than the basic one, i.e. you don’t get pointed shadows, because it is actually influencing the core geometry as well.
All of this geometry manipulation has led to a change in the typical geometry interpretation section of a GPU, from a single monolithic geometry setup stage to several (or so Nvidia believes; ATI kept things simpler with its HD 5xx0 series). This brings us full round to the Polymorph engine, which controls how each SM’s geometry is calculated.
Similar to this, Fermi replaces its single rasterization engine with four, each of which is paired with four SMs to form a graphics processing cluster. The rasterization engine oversees the conversion of all those 3D triangles into 2D pixels (GPC).
After this, the layout returns to being more recognizable, with the main thread scheduler, the Host Interface, and memory controllers serving as the event’s judges.
In addition to these major adjustments, Nvidia has also made a number of smaller adjustments. First off, the CUDA Core now handles single and double precision floating point calculations better and complies with the new IEEE 754-2008 standard. Added improvements to texture units include a speed boost and support for new texture compression formats. By eliminating the need to address system memory, a new big L2 cache further reduces latency.
“Finally, we receive AA on vegetation in Crysis” is said in the centre. (/centre)
Edge blending is now even smoother and more realistic thanks to the ROP units’ new 32x Coverage Sampling Anti-aliasing mode. Additionally, a brand-new super-sampling AA option in Crisis now permits the first-ever smoothing of foliage edges. A number of compression and efficiency improvements have also helped to improve overall ROP performance. The bandwidth to the framebuffer is increased by the addition of bigger and better implemented caches, considerably boosting among other things the speed of ray tracing.
The 3D Vision Surround extension is a little more esoteric. Similar to ATI, this is an effort to provide players with a justification for purchasing numerous expensive graphics cards, as normal gaming may not necessitate such expensive hardware. Similar to ATI’s Eyefinity, it essentially allows you to play games across three monitors, each having a resolution of up to 1,920 x 1,080 pixels. However, Nvidia has taken things a step further by including support for stereoscopic 3D as well. So, the opportunity exists for real 3D gaming provided you’re prepared to invest in two GTX 480 graphics cards, three monitors with quick enough refresh rates for 3D, and a set of 3D spectacles (in fairness, you can also game in 3D on a single monitor).However, we lacked the necessary testing equipment.
Then, Fermi is initially integrated into the GF100 chip that powers the GTX 480 and GTX 470. It has 64 texture units, 16 Polymorph engines, four Raster engines, and 512 Cores overall using four GPCs. These are connected to 48 ROPs that are divided into six groups of eight, with a 64-bit memory controller supporting each group.
The important thing to note in this situation is that 512 Cores is more than the GTX 480 had when I first discussed it in this article. This is because Nvidia disabled one of the SMs, and there is only one explanation for this: Nvidia can’t reliably produce GF100s that function in full. This is a typical issue, and it’s the main justification for the existence of graphics cards like the GTX 470. You can still acquire a functional chip with a lower performance by turning off the non-working portions of the chip. But this is the first time we’ve ever witnessed a manufacturer use chip disabling for a flagship product. Making a processor with three billion transistors as big as GF100 comes with this danger. These cards’ clock rates are also rather low, so if Nvidia can boost its manufacturing prowess, a faster-clocked card employing the entire, unrestricted GF100 chip may be available soon.
Nvidia GeForce GTX 480″ is shown in the center.
But for now, there are two cards that you may order in advance (stock will be shipping come April 14th). A GTX 480 starts at £448.99, while a GTX 470 costs £319.99. The GTX 480 costs about 40% more than the HD 5870 at these pricing and is just £50 less expensive than the dual-chip HD 5970. The GTX 470 costs £60 more than an HD 5850 and the same as an HD 5870. These cards will require extremely exceptional performance numbers to even approach a recommendation at these prices.
Nvidia GeForce GTX 470″ is shown in the center.
As previously stated, we’ll be looking at the top-tier GTX 480 in due course, but for the time being, let’s focus on the GTX 470. The GTX 470 is quite underwhelming in person, especially given the hype surrounding its release and how technologically advanced its GPU is (though, in fairness, the GTX 480 is more of a spectacle). It should fit in most ATX-size PC cases without a difficulty because it is 9.5in long, which is the same length as a motherboard is wide. It should also be simple to accommodate because it is relatively light and has a cooler that looks fairly standard.
Due to the chip’s 40nm manufacturing process, which allows for relatively low power consumption given its complexity and “just” two six-pin PCI-E power connections, it should work with the majority of current power supplies without any issues. Having said that, this card will consume more power than even an HD 5870 with a total board power of 215W.
Four display outputs (two Dual-link DVI-I, one HDMI, and one DisplayPort) on the most recent high-end ATI cards necessitate that one of them encroach on the second card slot, which is typically utilized to vent hot air from the card. However, because Nvidia only allows for three display outputs (two Dual-link DVI-I and one micro HDMI), the entire second slot is reserved for hot air exhaust.
It’s finally time to start contrasting DirectX 11 gaming performance with the release of competing DirectX 11 hardware from both ATI and Nvidia. But before we began testing, we ran through our regular list of DX9 and DX10 games. As usual, we conducted a number of gaming benchmarks after adding the card to our reference system, the specifics of which are below. With the exception of CSS and Crisis, the results are manually logged using FRAPs as we play the same game part over and over again. We replay time-demos for CSS and Crisis, and framerate is automatically recorded. To ensure consistency, every result is verified again, and the results’ average is noted. All of the in-game detail settings in Crisis are set to High, while the maximum graphical settings are used in every other game.
Since we’re only examining the GTX 470 today, we limited our DX11 testing to the GTX 470 and HD 5850 since they should be priced similarly (although they aren’t at the moment). We played Battlefield Bad Company 2, Just Cause 2, and Colin McCrae: DIRT 2. All of the in-game graphical settings were at their highest, and we used FRAPs to record framerates while performing manual runs-through.
Games for the “Test System – DX9 and DX10”
- Microsoft Windows Vista Home Premium 64-bit Intel Core i7 965 Extreme Edition Asus P6T motherboard 3 x 1GB Qimonda IMSH1GU03A1F1C-10F PC3-8500 DDR3 RAM 150GB Western Digital Raptor
“Cards Were Tested”
- GeForce GTX 470 and GeForce GTX 295 from Nvidia
- GeForce GTX 285 by nVidia; ATI HD 5970 by AMD
- ATI HD 5870 and ATI HD 5850 from AMD
”’Games Tested”’
- Far Cry 2
- Crysis
- Race Driver: GRID
- Call of Duty 4
”’Test System – DX11 Games”’
- Intel Core i7 965 Extreme Edition
- Asus P6T motherboard
- 3 x 2GB Kingston KHX1333C9D3K2/4G PC3-8500 DDR3 RAM
- 2TB Seagate Barracuda XT
- Microsoft Windows 7 Home Premium 64-bit
”’Cards Tested”’
- Nvidia GeForce GTX 470
- AMD ATI HD 5850
”’Games Tested”’
- Just Cause 2
- Colin McCrae: DIRT 2
- Battlefield Bad Company 2
In terms of DX9 and DX10 performance, the GTX 470 continues to easily outperform the GTX 285 that it replaced. With the exception of Far Cry 2, it performs roughly on par with an HD 5850 and significantly worse than the HD 5870. At the current price, this is insufficient. Yes, it outperforms both ATI GPUs in Far Cry 2, but in our opinion, winning one out of every four games is insufficient.
In terms of DX11 gaming, we observe a very even split, with the HD 5850 maintaining a sizable lead in Just Cause 2 and the GTX 470 maintaining a steady performance advantage in Colin McCrae:DIRT 2. But if we factor in value once more, the HD 5850 or perhaps the HD 5870 would be a superior option.
The situation for the GTX 470 doesn’t get any better when considering power draw. It consumes 10W more power at idle and 66W more when under load than an HD 5870. Your energy cost might not go up by hundreds of pounds, but it’s still not ideal.
When we consider noise levels, the situation is similar because the GTX 470 is noisier than the HD 5870 and HD 5850 at both idle and under load. The GTX 470 isn’t noticeably more bothersome when gaming than the others, but none of these cards are exactly silent when idle either. All of these cards create an audible whooshing sound.
One advantage Nvidia does have is its own technologies, especially PhysX and 3D gaming. The former still appears as a bonus in select games, and when it does, it truly enhances the spectacle. Regarding 3D gaming, we still consider it to be somewhat of a gimmick, but Nvidia claims that it is already compatible with over 400 games, so it may be about to take off. All things considered, however, we don’t feel that any of these features is sufficient to warrant the additional cost of the GTX 40 cards. Are you not satisfied with Nvidia GTX 470 performance? Why not look for better cards? You must read a comparison of Gtx 1650 Vs Rtx 3050 to see how two geeks perform and this makes up your mind to save some more bucks and buy the best card to have an amazing gaming experience.
Verdict
On a technological level, it appears that the long wait for Nvidia’s genuine next-generation graphics hardware was worthwhile. The Fermi architecture has a tone of functionality and can undoubtedly improve the performance of both gaming and non-gaming applications. In reality, the architecture is entirely theoretical, and the finished product hasn’t yet met expectations. Simply said, the GTX 470 is overpriced and underwhelming.