Video Graphics Card Comparison

[absolute55]

You might be lost in the dozens of references offered by NVIDIA and AMD, made various renaming do not help. To see more clearly, we have harnessed the past few weeks at a fairly daunting task, since it is present you the performance of all (with some exceptions) solutions DirectX 10 NVIDIA and AMD / ATI since the release of the first. Specific outcomes for each game will be presented on a gross basis, with a page containing a performance index performance in three resolutions with the two highest settings, and all graphics for each of the nine combination of resolution and control chart. On the other side have gotten the Nvidia GeForce 8800 GT with DirectX 10 in addition to their SLI setup, turn the mid-range with the GeForce 8600 GTS and low-end GeForce 8400 GS TurboCache. On the ATI side, were added DirectX10.1 Radeon 3870 and 3850, also increase the power of the Radeon 2900 XT in CrossFire. In addition, we have added the Radeon 2600 PRO and 2600 XT, both individually and in CrossFire. Finally, in counterpoint to the 8400 GS, has joined the Radeon 2400 Pro Hyper-memory.

It has been shown that these comparisons are very valuable for people who want to compare graphics, especially for those who are thinking about buying one, either individually or in SLI / CrossFire. Here we will give the keys to decoding to understand how different manufacturers call their graphics cards, then try to compare a theoretical point of view the performance of cards based on their internal structure.

The critical component on a graphics card is the graphics processor, GPUs in English, which is associated with memory for storing data. In general, the higher the range, the card needs more power to operate. But often there are people finding themselves faced with a incomplete support and realize that there is lack of cables (8-pin PCI-E and / or 6-pin) or PC power off because even if you have a burning Food very poor quality.

There are two simple tips to follow:

• Refer to the manufacturer's specifications for the card you have purchased
• Avoid brands that are unknown, this do not perform as advertised and may even be dangerous for some models.

The entry-level sector is one of the most paradoxical markets for manufacturers of GPU. For if it is here that the overwhelming majority of volume sales - and therefore the turnover and financial results - the products available have no attraction and glitter of high-end cards or even mid-range. Often, these cards do in fact not even play in 1024x 768, or push the 3D options. In their defense, it must be said that the equation subject to manufacturers is not simple offer a product with honorable performance at a good price, three times lower than the first maps of the middle range. Considering the price incompressible due to the presence of at least 128 MB of memory (currently) of PCBs, some electronic components and radiator, this leaves very little or nothing for the GPU itself.

[absolute55]

However, recently there was change at this level, directly from the arrival of PCI Express. By the way, improvements brought by this new interface are surprising. Even if the gains from the AGP 8X remain unavailable in the game , PCI Express has already introduced two very different technologies but both interesting, much faster than the deeper changes that to cause in terms of game programming. The first is the SLI and the AFR, and the second and Hypermemory TurboCache interest to us today. The main purpose of these technologies is in effect to minimize the amount of memory onboard the graphics card to generate more money for the GPU, or at least the manufacturers. The ultimate goal remains increasing the price / performance ratio on these entry-level solutions, at least in theory.

At the below I had attached a post for the list of desktop and mobile cards. Look at the attached file.

Video Memory

To understand how these different techniques should have an idea of how video memory is managed. When a 3D chip must apply a texture to a primitive, it tries to access its local memory and if it is not present, it transfers it from the main memory via DMA (Direct Memory Access, this technique allows a device to transfer data directly from main memory without requiring CPU cycles. It is opposed to PIO mode where the CPU must copy the data itself on the bus).

But the amount of video memory is not infinite, when it is full it must replace a texture already present by the news. To determine which texture should be replaced most of the graphics card drivers use a LRU. That texture that is replaced is that which has not been used since the longest. This algorithm works properly most of the time but can also be catastrophic in some cases! Especially if the total amount of textures needed for a frame does not fit in video memory, the card is found to transfer all the textures in every frame, which is quick to put on the knees.

The failure of the Virtual Texturing is more surprising, the technical principle is fundamentally good, no loss of performance and solves several problems gracefully. Yet apart from 3DLabs, no other manufacturer will follow this path. It is difficult to believe that this is the hardware overhead to create a mechanism for virtualization of local memory that has hampered ATI and nVidia, so there is no logical explanation for the failure of this technology in all cases should be reborn in a much more advanced version with WGF
With the HyperMemory and TurboCache ATI and nVidia challenge therefore for every day techniques to save video memory, drawing on earlier initiatives in the field and improving them. This time the two manufacturers have realized that this was not on the high-end sector that should be shared as shown by the dramatic example of AGP texturing.

[absolute55]

The nVidia 6200 TurboCache for their part are divided into three versions, announced earlier and already available:

• GeForce 6200 TurboCache 128MB (with 16MB onboard 32-bit)
• GeForce 6200 TurboCache 128MB (32MB on-board 64-bit)
• GeForce 6200 TurboCache 256MB (with 64MB onboard 64-bit)

Difference between the old and new series

Older generations, series 1000 and 2000:

• The thousands digit characterizes the family of GPUs: a 1900 is older than 2900;
• The hundreds digit is connected to the power of the GPU, the higher it is, the GPU is more powerful: a 1950 Pro is more powerful than a 1650 Pro;
• Finally, a suffix comes complete reference. The most common suffixes were in ascending order of performance: GT Pro, XT.

New generations, series 3000, 4000 and 5000:

The thousand figures does mean the same thing generation: the higher it is, the map is more recent, for example, the 4000 series is newer than the 3000 series.

• A 4850 is much more powerful than 4650 the highest number of hundreds indicates a higher GPU, as before;
• A 4850 is a bit less powerful than the 4870. Here, ATI has changed the dozens, because it is the same GPU running at a higher frequency (as memory), instead of changing the suffix as before.
• You can compare cards from different generations, but not maps which reference is completely different 4850 is less powerful than a 5850 (single-digit difference), but nothing can be said a priori on the comparative performance a 3850 and a 4570.
• 4570 is the most recent (thousands digit = 4> 3), but the hundreds digit is less than (5 <8): he must go look at the structure of their GPUs to say more.

Nvidia

NVidia, the numbering follows as an almost logical order, but a bit more complicated. Indeed, in early 2008, Nvidia launched the 9000 series, but these maps are based on the same GPU as the 8800 release in the winter 2007-2008, the G92, mainly with frequencies up: this GPU is still found on Recent GT240 cards and 250.

The HD 4650 iSilence is indeed based on a Radeon HD 4650 clocked at 600 MHz which includes 320 stream processors and 8 ROPs. His memory is in turn the DDR2 128bit and are entitled to 512MB clocked at 1000 MHz. In addition, as the name implies, this card has a completely passive cooling (Zalman VNF100) and could therefore be suitable for those seeking a quiet card for a PC and / or home cinema. Side temperatures, despite the waning passive one does not exceed 85 ° C heavy use while in office and multimedia use are typically ranges between 45 and 55 degrees.

The gear includes two DVI ports and a TV output, while the bundle is fairly basic for him with a DVI> HDMI (which recovers the audio signal), another DVI> VGA, and finally a CD-ROM with pilots.

[absolute55]

For older generation, Series 8000 and 9000:

• The thousands digit characterizes the family of GPUs, so almost the year of release. For example, an 8600 GT, released in 2007, is older than a 9600 GT, released in 2008.
• The hundreds digit is connected to the power of the GPU, the higher it is, the GPU is more powerful (in the same family) is the determining factor. For example, an 8600 GT is much less powerful than an 8800 GT.
• Finally, a suffix is completed by the reference, it is related to the GPU and memory frequencies, the number of units activated GPU (or stream processors), it significantly increases the performance, from lowest to highest, were: LE G, GS, GT, GTS, GTX, Ultra. All suffixes do not exist for all cards.
• From June 2008, nVidia released the GTX 200 series, based on the GT200 GPU, composed mainly of GTX 260, 260 +, 275, 280, 285 and 295. Were added to this series of cards based on old GPU cards 8000/9000 series, for the GT200 bit difficult to be suitable for low power: the series appeared 100 and 200.

For older sets, you can compare maps of different generations, but not maps which reference is completely different a 8600 GT is less powerful than a 9600 GT (single digit difference), but nothing can offhand performance compared to a 8600 GTS and 9500 GT a run (all different): he must go and see the structure of their GPU power to rule. As for the new generation cards, this is not simple. GT 200 series for the tens digit increases corresponds to a more powerful GPU, but between 100 and GT Series GT 200, impossible to navigate you must go see the internal structure of the graphics processor (GPU).

The GPU is essentially composed of units of computation, called Flow Processors (Stream Processors, SP), and also called processing units or cores nVidia CUDA. More than a card Processor Flow (SP), more powerful it is. To get a clearer idea of theoretical performance card from the same manufacturer them, it must look at the structure of the GPU. This one is composed of stream processors (Stream Processors, SP), which is the essential calculations, texturing units (TU) and POR's (Raster Operators).
ATI is doing well here anyway, but the introduction of the 6200 TC 32 MB steals the show at X300 SE. The 6200 is 16 MB for its part once again dropped, but still less than the GMA 900.

Example

Between the HD 3850 and HD 4670, the comparison of the internal units of the GPU gives:

HD 3850 (GPU name: RV670) stream processors: 64 v5, often counted 320 (64x5), 16 texturing units and 16 raster units (POR's).
HD 4670 (GPU name: RV730) stream processors: 64 v5, 32 texturing units and 8 raster units (POR's). The same amount of stream processors, but a difference between texturing units and ROP's. As the stream processors are the most important, we can say that the analysis of GPU performance will be similar, but with a margin of error greater than in the case of the GeForce cited above.

To compare the performance of an nVidia graphics card and an ATI card is more complicated. There must be above compare the number of stream processors, texturing units and ROP's, but the architecture is quite different from nVidia and ATI GPUs, including on stream processors, requires home a little more detail . Since the 8800 first version, NVIDIA proposes an architecture he calls LMIS is to say, Single Instruction, Multiple Threads, which corresponds in practice to the implementation of an operation on multiple threads (son of execution) as opposed to the architecture used by ATI which is called SIMD "Single Instruction, Multiple Data", where an investigation of several operations is applied to an item.

ATI

For six months now from ATI 4870X2 was what was more powerful (more expensive too) in the field of graphics cards for gamers. Nvidia was able to respond immediately in only able to offer dual-gpu card with a 65nm GT200 too greedy, it's now done with the Geforce GTX 295 which has two chips in 55nm GT200 engraved. This map looks like a SLI GTX 260 +, although on some points it also draws on the GTX 280, but what about in practice, what level of performance expected? With the Nvidia GTX 295 does succeed in making the leadership and AMD 4870X2.

Five cards will therefore oppose: the GTX 295, GTX 285, GTX 280 for two cards to Nvidia and AMD 4870X2 and 4870. This little world will be refereed by Intel and its Core 2 Duo E8500. Nvidia will successfully assume leadership to AMD, but we will see that things do not stop there.

[absolute55]

GPU Specifications :

ATI Radeon HD 4770

The Radeon HD 4770 we had the hands for this folder is the same as that used in our test of the HD 4770 to find a sample directly provided by AMD. The map has been delivered naked, without box and accessories. Reference card required, which includes its GPU shaders and 16 ROPs 640 is clocked at 750MHz, while the 512MB of GDDR3 128bit are clocked at 800 MHz. To learn more about this first map engraved in 40nm GPU.

Gigabyte Radeon HD 4670

After the HD 4650 is logically a Radeon HD 4670 is in the spotlight. This one comes from Gigabyte and deserves to be equipped with Ultra Durable 2 components called for greater stability and shelf-life if we are to believe the claims of the manufacturer. The map meets the frequencies of origins, however, the Radeon HD 4670 GPU with a running at 750 MHz and 512MB 128bit GDDR3 memory at 1000 MHz. Like the HD 4650 GPU includes 320 stream processors and therefore differs from the latter mainly by its frequency and memory.

One big advantage of this card for home theater PC enthusiasts is the presence of an HDMI on the gear next to the VGA and DVI plugs. You will not need an adapter what we believe is a fair point, although in the bundle Gigabyte offers in addition to the manual and CD-ROM with drivers, DVI> VGA. However, the fan of the card is not thermo regulated and sounds even when the machine is at rest.

Gigabyte Radeon HD 4850 1GB Multi Core

Second card fan-less our comparison, the Gigabyte HD 4850 1GB uses a heatsink Gigabyte Multi Core which was the basis used on graphics cards with GPU slightly less powerful than Radeon HD 4850. On this map, if silence is part of the mode toaster is not as far this heating product once it is requested. Moreover, we could not even finish our assignment with temperatures making it because the PC crashed after a few minutes FurMark. In short, a card discouraged if you do not have a well-ventilated case.
Apart from that, she embarked 1GB memory clocked at 940 MHz while the classical models are entitled to 512 MB which operate at 993 MHz. The GPU runs at him for 640 MHz and thus enjoys a mild overclocking compared to 625 MHz recommended by AMD.

PowerColor Radeon HD 4890 PCS

For the evaluation of the most powerful GPU from ATI we were treated to a graphics card PowerColor Radeon HD 4890 PCS. This map is not really like all the others since PowerColor Radeon HD 4890 used a cooling system and has significantly boosted ZEROtherm operating frequencies of the GPU and memory. Indeed, while conventional HD 4890 operating at 850/975 MHz (GPU / memory), the PCS 4890 turns on her at 950/1050 MHz frequencies that should enable him to get some interesting results. Note also that we had the right to bundle in his edition BattleForge which, as you'll understand the game BattleForge includes in its full version. In addition to this game, we find in the box a manual, CD with drivers, a CrossFire bridge (and the only one comparison), the DVI> VGA and DVI> HDMI cable and a HDTV.

GeForce 9800GT

First graphics card based on NVIDIA GPU Gigabyte GV-N98TZL-512H is equipped with a GeForce 9800GT which is really nothing more than a GeForce 8800GT renamed. It is therefore the basis of a G92 engraved in 65nm, which includes 112 stream processors, 16 ROPs and 256bit GDDR3 memory on buses. The Gigabyte following specification with reference frequencies 600/1500/900 MHz for the core, shaders and memory. Cooling side she cons by using a cooler Zalman VF-830 which was already used on one of the 8800GT manufacturer. Unfortunately, this cooler is not what makes quieter and sounds even at rest, even if there is loudest on the market. The bundle includes two DVI> VGA, DVI> HDMI connector, a Molex> PCIe 6-pin, an HDTV, a manual and a CD-ROM with drivers for installation.

[absolute55]

List of Top 5 Graphic Cards :

• Sapphire Radeon HD 4850 512 MB
• Sapphire RADEON HD 5850 1024 MB
• XFX HD 5850 1024 MB
• Gainward 9500 GT 512MB
• ASUSTeK COMPUTER SILENT/DI/512MD2 EAH4350 512MB

Entry Level Graphic Cards

• Sapphire Radeon HD 5450 - 1GB
• Gigabyte GV-R567OC-1GI
• MSI N9600GT-T2D2G 2048MB
• Mid Range Graphic Card
• MSI R5750-PM2D1G
• Sapphire Radeon HD4870 1024MB
• Generic 512MB GTS 250
• Sapphire Radeon HD 5770 Vapor-X - 1GB
• Sapphire RADEON HD 5850 1024 MB
• Zotac GeForce GT 240 - 1GB
• Asus ENGTS250/DI/512MD3

High End Graphic Card

• Sapphire HD 4890 1024 MB
• Generic GeForce GTX 275 896MB
• Zotac GTX 295 1792 MB
• XFX Radeon HD 5850 XXX - 1GB
• MSI N470GTX-M2D12
• Sapphire Radeon HD 5870 Vapor-X - 1GB

Conclusion:

The choice between ATI and Nvidia does not only depend on the absolute performance. New Nvidia PhysX accelerator support, for example, and the rather less interesting for players of CUDA, which can offer not ATI. That Radeons are at eight times the anti-aliasing before the Nvidia counterparts, also offers the latest generation of DirectX 11 and Eyefinity (HD-5000 series), by which you can operate up to three screens. In the image quality, the two have little bearing.

Indeed, the GTX 280 is Nvidia card just goes everywhere, even with an Athlon X2 6400 AMD performance difference will be very little different from an Intel Q6600. Instead, with the ATI 4870X2 this difference is 50 %. But even if the ATi card shows less affordable for older configurations, the test showed the limits of the 4870X2 is still not reached! The next generation processors will surely exploit the advantage. The question often arises during the acquisition of a graphics card: How much memory do I need? And contrary to some beliefs of ancient times, the amount of memory is not what most affects performance.

The main question to ask: Under what conditions the game will I change means what resolution and what filters. If the question does not really arise for the latest generation of cards (nVidia or ATI), the question remains unanswered for a slightly older lines like the famous Ati HD4870 is available with 512MB or 1GB of RAM.

But now with the emergence of new games using textures more impressive, the need arises in memory, so if you play in resolutions at or above 1650x1050, you might consider purchasing this card with a little more memory. 1GB becomes the norm today, 2GB is useful only in rare cases. Note that dual-gpu cards (GTX 295 or 4870X2 for example) show a double quantity has what they can really exploit because of their design. The analysis of the structure of the GPU can go further in predicting the performance of cards, with less precision than practical tests.

As many of you know, the drivers of GPU regularly updated each time bringing their own set of bug fixes and improved performance theory. In practice, the impact on performance is zero or almost. The 50% improvement regularly described by the manufacturers, are in fact very special cases which are not representative at all gains you get. The difference this is especially so on the bug fix, but beware, some appear in the updates. So if all the games you play does not present problems, there is no reason to change.