A Free Trial That Lets You Build Big!
Start building with 50+ products and up to 12 months usage for Elastic Compute Service
How to identify the main performance parameters of a video card
Gamers who first know the video card may be confused about the performance parameters of the video card when purchasing the video card, I don't know who has the biggest impact on the performance of the video card, which parameters are not the bigger the better, and the same price of the video card, but in some individual items, the card or one of the N cards is more powerful than the opponent. These problems must be the information that every friend who has just been in touch with a video card wants to know most, it can be said that sales staff in many stores use these users to deceive and blind consumers by failing to understand the basic performance parameters of the video card. Today, the graphics card emperor is here to explain in detail the significance of the Main Performance Parameters for entry-level graphics card users.
For graphics card performance parameters, there are a lot of hardware detection software can be detailed detection of graphics card hardware information, such as: Everest, GPU-Z, GPU-shark and so on. Here we take the player's most commonly used GPU-Z software as the example software to parse the graphics card performance parameters.
The GPU-Z of gtx590
First of all, we have a rough partition of the interface of the GPU-Z software, from top to bottom a total of 8 partitions, the specific meaning of each partition is:
①. Video Card Name section:
Name/Name: the name of the video card, that is, the model of the video card.
②. Display the chip model:
Core code/GPU: the GPU chip code is displayed here, as shown in gf110 and Antilles.
Revision/Revision: displays the number of the GPU Chip's step-by-step process.
Manufacturing Process/technology: displays the GPU chip process, such as 55nm and 40nm.
Core area/Die Size:The core size of the GPU chip is displayed here.
③. The hardware information section of the video card:
BIOS version/BIOS version: the version of the video card BIOS is displayed here.
Device ID/device ID: the ID of the device.
Manufacturer/subvendor: the name of the OEM for this video card is displayed here.
④ Display chip parameters:
Grating operation unit/ROPS: displays the number of GPU-owned ROP grating operation units.
Bus Interface/Bus Interface: displays the bus interface type and interface speed between the video card and the main board's North Bridge Chip.
Color unit/shaders: displays the number of GPUs that have the color palette.
DirectX version/DirectX support: displays the DirectX version supported by the GPU.
Pixel fill rate/pixel fillrate: The pixel fill rate of the GPU is displayed here.
Texture filling rate/texture fillrate: displays the texture filling rate of the GPU.
⑤. Video storage Information Section:
Memory type/memory type: displays the memory type used by the video card, such as gddr3 and gddr5.
Memory width/bus width: displays the bandwidth between the GPU and memory.
Memory size/memory size: displays the physical memory size of the card.
Memory bandwidth/bandwidth: displays the data transfer speed between the GPU-Z and memory.
Driver version/driver version: the version number of the currently used video card driver in the system.
7. Video Card frequency section:
Core Frequency/GPU clock: displays the current GPU running frequency.
Memory/memory: displays the current operating frequency of the video memory.
Shader/shader: displays the current running frequency of the coloring unit.
Default core frequency/default clock: displays the default GPU running frequency.
(Default) memory/memory: displays the default operating frequency of the video memory.
(Default) shader/shader: display the default running frequency of the coloring unit.
Battle and computing capabilities:
Nvidia sli or ATI crossfire: whether to enable the SLI or crossfire multi-card crossfire.
Computing Capability: shows whether opencl, Cuda, physx, and directcompute 5.0 are supported.
Next, we will give a detailed explanation of the performance parameters of the graphics card, and finally end the performance of the graphics card that are most affected by the performance parameters, in addition, what aspects should our consumers consider when purchasing a video card, and how to make a rough judgment and Comparison on the video card based on the performance parameters of the video card. We hope that after reading this article, cainiao will have a rough "outline" of the video card. After reading this article, laruence will have a deeper understanding of the core performance parameters of the video card.Video Card Name, chip model, and hardware information
Video Card Name, chip model, and hardware information
When buying a video card, the consumer should first be clear that I should buy the video card model (name) What is, that is, the GPU-Z name shown in the parameter information, for example, "gtx590" in this example ".
Graphics card GPU-Z for graphics card names, chip models, and hardware information
Through the interpretation of the graphics card chip model, we can further understand the information of the graphics card core GPU. From the GPU, we can learn the gpu r & D code of the video card core. Generally, the GPU Code corresponds to the video card model name. For example:
NVIDIA geforce gtx590
Gf100: NVIDIA geforce
Antilles: radeon HD 6990
Rv870: radeon HD 5970
Based on the gf104 core, the video card also has three products: NVIDIA geforce GTX 460 (768 MB), NVIDIA geforce GTX 460 (1024 MB), and NVIDIA geforce GTX 460se, if the consumer is not familiar with these graphics cards, the capacity will be misled by the seller and the video card products will be "converted" will be purchased.
Differences between the three gtx460 models of the same model
From the above comparison, we can clearly see that the gtx460 of the 768mb and 1024mb Versions differ greatly in the memory capacity and memory bandwidth, the difference between gtx460se and gtx460se is that Cuda processors are reduced to 288.
The same GPU Code corresponds to multiple graphics card models. Naturally, the same graphics card model corresponds to multiple GPU codes, for example: radeon
Two versions of hd5670
We can see that the main difference between the two lies in the number of core GPUs, stream processors and the core area. Although the same as hd5670, the performance of the 640sp hd5670 is almost close to that of the hd5750.
Based on the above summary, we learned that when purchasing a video card, our players must understand the graphics card model they want to purchase and what is the GPU core code of the video card, when buying it, it is best to conduct a simple computer test on the scene of the video card, with GPU-Z and other related testing software to see if there is an exception in the hardware information of the video card, in this way, players can minimize their chances of being cheated.Analysis of graphics card chip parameters: Understanding ROPS
Analysis of graphics card chip parameters: Understanding ROPS
This part is what we want to focus on, because many friends or old players who are playing a video card are not very clear about these performance parameters. Let's take a detailed explanation.
Video card chip Parameters
The first important concept is ROPS (raster operations units), which is a raster processing unit that shows the number of OPS processing units owned by the GPU.Generally speaking, 3D image processing can be divided into four main steps: geometric processing, setting, texture and grating processing. ROPS is the processing of grating units.. So what are the effects of the number of raster processing units on the performance of the video card?
ROPS (raster processing unit) is mainly responsible for the light and reflection operations in the game, taking into account AA, high resolution, smoke, flame and other effects.The more powerful the AA and light effects are in the game, the higher the performance requirement for ROPS (raster Processing Unit ).Otherwise, the number of game frames may drop sharply. for example, for a game with the highest image quality, a video card with eight grating units may only run at 25 frames. the graphics card with 16 grating units can be stable at more than 35 frames. For example, gtx550ti and hd6790 have 24 ROPS units and 16 ROPS units. Although hd6790 is ahead of gtx550ti in most test projects, however, in the case of high AA (Anti-sawtooth) load, the weakness of hd6790 is immediately exposed, and 16 ROPS units seem a little powerless. From farcry
2 also confirmed this point: in the game, the lag of HD 6790 in 4xaa is about 4%, and the lag of performance after 8 xAA is enabled is expanded to 15-17%.
Note that,AMD and NVIDIA graphics cards differ in ROPS design.The ROPS units of the N-card and the stream processor are "Bundled", that is, placed within the SIMD. Therefore, if the number of stream processing units of the N-card is reduced, The ROPS units of the N-card are also reduced; A card is different, and its ROPS unit and stream processor unit are not associated.
Traditional pipeline architecture
The second important concept is shaders. Traditional pipeline architecture: In the past, a video card was composed of a vertex rendering pipeline and a pixel rendering pipeline. The image generation process was first composed of vertex shader in the vertex rendering pipeline) generate the basic geometric skeleton (composed of triangles), fill the color with the pixel shader in the pixel rendering pipeline, and map the texture units in the pixel rendering pipeline. WhileAfter the new unified rendering architecture is proposed, the vertex shader and the pixel shader are combined into one to become a stream processor (shaders)It will be responsible for vertex coloring and pixel coloring at the same time to avoid load imbalance. Microsoft DirectX was the first to propose a unified rendering architecture.
In the dx10 era, the number of shader units has become one of the important parameters for measuring the video card level.
It should be noted that the core architecture of N card and a card is different, and N card adopts the MIMD architecture. Multipleinstructionstreammultipledatastream (MIMD), which uses multiple controllers to asynchronously control multiple processors to achieve spatial parallelism. Therefore, N card is a transmitter; card a adopts the SIMD architecture design, that is, Single Instruction Multiple Data (single-instruction stream multi-data stream). card a packs four simple commands + one complex command and sends it with a single transmitter. Therefore, a/n cannot compare the number of stream processors.
Finally, we need to parse the pixel fill rate (pixel fillrate) and texture fill rate (texture fillrate ).
Pixel filling rate refers to the number of pixels rendered by the graphic processing unit per second.The Unit is gpixel/s (Billions of textures per second)
Pixel fill rate = core frequency X number of grating units/1000
Texture filling rate refers to the color texture filled on each polygon surface.The Unit is gpixel/s (Billions of pixels per second)
Texture filling rate = core frequency X number of texture units/1000
The value of these two parameters is naturally larger in the GPU-Z, the more powerful the graphics card can handle. In addition, the core frequency is the computing factor of pixel fillrate and texture fillrate. Obviously, the higher the core frequency of the video card, the larger the two values. The number of grating units is the value of ROPS. The larger the value of ROPS, the larger the pixel filling rate.Do not forget the memory parameter "Bit Width brother"
Do not forget the memory parameter "Bit Width brother"
The role of memory is similar to the memory used in our machine. Sales staff in the store often perform "hypes" on the "selling points" of the display. Of course, many "little white" are also cheated. The following graphic card Emperor will be used for parsing for junior gamers.
GPU-Z about memory
Memory type (video memory type). Currently, the latest mainstream high-end graphics cards use gddr5 video memory particles. The mainstream gddr3 video memory particles have also been retired to the second-line, the gddr4 memory is only a transitional product, and there are not many commercial graphics cards.The core advantage of gddr5 over gddr3 is its significant increase in memory bandwidth..
Memory bandwidth = (memory width × memory frequency)/8
From the above computing companies, we can clearly see that because the gddr5 memory has two data buses, although the same 8-bit prefetch mechanism as gddr3 is used, however, the operating frequency of the video memory can be twice that of gddr3. The most typical example is that gt240 graphics cards with gddr5 Display memory are about 16% ahead of the performance of gt240 graphics cards with gddr3 Display memory. Therefore, with its powerful bandwidth advantages, gddr5 can surpass gddr3 in the case of same-bit width.
Bus width is often the most common concept for gamers.The video memory width is the number of digits that can be transferred by the video memory within a clock period. The larger the number of digits, the larger the amount of data that can be transmitted in an instant.. It can be said that the effect of the video memory width on the performance of the video card is much greater than that of the video memory capacity.
Graphics card View
How much is the weight of the video card memory width? Let's look for the answer from the above picture. Based on the test results of most media sets, we can sort the above graphics cards in a simple order.
We focus on the two graphics cards hd6790 and gtx550ti. Although gtx550ti is higher than hd6790 in the core frequency and number of ROPS, why did it fail in most test projects? First, the core architecture of hd6850 used by hd6790 is naturally reduced. Second, on the memory width, gtx550ti is only 98.4 Bit Memory width, so the memory bandwidth processing capability is only Gb/s, the hd6790 adopts the 134.4-Bit Memory width, and the memory bandwidth processing capability reaches 36.58% Gb/s, which is higher than that of gtx550ti. Why can gtx550ti fully suppress hd5770 compared to hd5770? The truth is the same.Bandwidth = (bandwidth)
× Memory working frequency)/8 we can also see that when gddr5 memory particles are used, the memory width becomes a key bottleneck affecting performance..
Finally, we need to remind players that the most classic "scam" of memory size is the use of hyper memory (HM) on card a or turbo cache on card N (TC) on card) the dynamic sharing system memory technology to lie about the video memory capacity of the video card, presumably such a trick in the past few years after the "baptism, nowadays, many consumers have gradually been able to quickly and accurately identify such skills.Video Card frequency: core frequency> video memory frequency
Video Card frequency: core frequency> video memory frequency
The frequency of the video card. We mainly focus on the core frequency and memory frequency. In comparison, the core frequency has a greater impact on the performance of the video card. Therefore, the core frequency for our players to increase before the memory frequency. Why is the core frequency more important? For example,The core frequency is equivalent to the individual's own ability, while the memory frequency is like an external condition. The success of a person is often dependent on the individual's ability, and the external condition only affects it to a certain extent. In short: one is internal, and the other is external..
For graphics card frequency in GPU-Z
It should be noted that due to the different design of the core architecture, the core frequency of the N-card GPU and the shader frequency are two times the relationship, and the GPU core frequency of the card is the same as that of the shader.
The memory frequency is the default operating frequency of the video card. The unit is MHz (Z.
Gddr5 memory Particles
With regard to the gddr5 Display memory frequency, since gddr1/2/3/4 and ddr1/2/3 data buses used the DDR technology in the past (data is transmitted each time on the rising and falling edges through the differential clock ), the official nominal frequency X2 is the data transfer rate, which is usually referred to as the equivalent frequency.Gddr5 is different. It has two data buses, which is equivalent to Rambus's QDR technology. Therefore, the official nominal frequency X4 is the data transfer rate.. For example, the gtx590 official memory frequency is 854 MHz, which is often called 3416 MHz.
Now we know the effect of the video card frequency on the performance of the video card. We need to discuss the question: is the higher the video card frequency, the better?
From the frequency settings of the flagship video card, their core frequency settings are not particularly high, and some time ago, it was also reported that AMD will not ensure the quality of the hd6990 graphics card due to overclock damage to the video card users, at the same time, some foreign media also burned out the gtx590 video card. It can be seen that the high-end graphics card with high frequency settings is indeed not appropriate because of the extremely high GPU core temperature, resulting in the burning of such a card playing method. After all, players can still operate stably with their own video cards. Of course, in the middle-end graphics card, we can also see high-frequency graphics cards set in air-cooled situations, such as the recently launched gtx550ti, 1G core frequency set graphics card products also have some. What is the significance of a high-frequency video card? analyze it from another perspective: as a method to detect the quality of the video card, because graphics cards that can run at high frequencies require good workmanship and strong graphics card radiators for support. In fact, in most cases, our players apply the default frequency settings of the video card, so we are also worried about the length of service life of the high-frequency video card. Therefore, we do not recommend players to pursue the ultimate high frequency unless you are an avid overclocking player.Drive, crossfire, and other computing capabilities
Drive, crossfire, and other computing capabilities
To some extent, the performance of the graphics card has a certain relationship with the graphics card driver, because GPU manufacturers will optimize the graphics card. Therefore, we recommend that you use the latest WHQL driver to experience your video card.
Driving, crossfire, and computing capabilities
SLI and crossfire technologies provide technical solutions for Multi-card interconnection. What kind of video card is more cost-effective to build a multi-card interconnected system? It is more cost-effective to use an interrupted video card to build a system. Of course, if you have enough funds, you can use a high-end video card to build a multi-card interconnected system. If you use a low-end video card to build a crossfire, this is not very cost-effective, because the performance advantage of the platform established by the low-end graphics card is only equivalent to the capability of the Middle-end card, but the price has already exceeded the price of a single middle-end video card.
SLI fast force technical platform
Crossfire crossfire Technology Platform
Finally, let's take a look at the computing capability. opencl (Open Computing language, Open Computing language), Cuda (general parallel computing architecture), and physx (physical acceleration) and directcompute 5.0 (an application interface for general GPU computing ).
AMD hd6990 Problems
Compared with a card, N card is superior in the computing performance. N card supports all four computing capabilities, while a card only supports one directcompute 5.0. If players require physx physical acceleration to support HD transcoding or games that use Cuda technology, they can consider purchasing n cards, because these are the places where n cards are strong.Summary: "video card performance parameter scheduling"
Summary: "video card performance parameter scheduling"
After a detailed introduction and analysis, we have a comprehensive understanding of the main performance parameters of the video card. When a player purchases a video card, it may be a bit dizzy with so many performance parameters. Therefore, we need to sort the weights of these video card performance parameters in a descending order. The following is a simple sort of video card performance summarized by the author:
①Graphics core and Process
The core of the video card is the key. If the core is not good, the cloud is the best. If the core is advanced, the performance of the video card will naturally increase to a great level. The more advanced the process is, the lower the heat and power consumption of the video card.
②Stream processor and ROPS
The increase or decrease in the number of stream processors has an immediate impact on the performance of the video card. Therefore, GPU manufacturers often use this method to segment the video card product market. The majority of ROPS affects the AA (anti-aliasing) and light and shadow effects in the game screen.
③Core frequency and memory frequency
The core frequency affects the pixel filling rate and texture filling rate, while the video memory frequency affects the video memory bandwidth. Both are used as the influencing factors, so the larger the parameter value, the more powerful the natural graphics card performance. However, excessively high frequency settings have a certain impact on the video card itself. A reasonable frequency setting is the video card we want to choose.
④Pixel filling rate and texture filling rate
Pixel fill rate = core frequency X number of grating units/1000
Texture filling rate = core frequency X number of texture units/1000
⑤Memory width and memory bandwidth
Memory bandwidth = operating frequency × memory width/8 (memory bandwidth = memory width × memory Frequency/8/1024)
The larger the memory width, the larger the data size that can be transmitted instantly. The function of the memory bandwidth is like a bridge. It provides a data exchange channel for the display core and memory.
⑥Memory size and other parameters
If the video memory is too small, there will be unstable display of frames during the game.
Start building with 50+ products and up to 12 months usage for Elastic Compute Service