At CES 2022, NVIDIA officially released the GeForce RTX 3050 desktop GPU. From the earliest news in late 2020 to the release in early 2022, there were many rumors during the time when NVIDIA launched the GPU, and finally crossed the entire 2021 We finally saw this graphics card. Compared with the previous GeForce RTX 3060/RTX 3060 Ti, the GeForce RTX 3050 on the desktop platform is closer to mainstream users, and for the first time, it brings the latest ray tracing technology to the "50" series of desktop GPUs.
NVIDIA said that 75% of gamers are still using GeForce GTX graphics cards, and a significant portion of them will be the target group of GeForce RTX 3050, which will replace the GeForce GTX 1650 series in the future. The core codename geforce RTX 3050 is GA106-150 and is still based on the GA106 core, which is closer to the GeForce RTX 3060.
GA106-150 core introduction
GeForce RTX 3050 uses the new GA106-150 core, in fact, the complete GA106 core with 3 sets of GPC shielded 1 group of GPC and obtained, so a total of 10 groups of TPC is 20 groups of SM units, the production process of silicon wafers is still Samsung's 8nm process customized for NVIDIA.
Here's a table of specifications for the GeForce RTX 3050 versus the GeForce GTX 1650 and GeForce RTX 3060:
Compared with the previous generation of the same positioning of gtX 1650, the biggest change in the macro of the RTX 3050 is the addition of Tensor Core and RT Core, which is a qualitative change from 0 to 1, not only that, this big step change is directly to the use of the third generation of Tensor Core and the second generation of RT Core, it is such a leap forward, so that the RTX 3050 may also be able to position the entry card but can also be competent for the mission of running ray tracing games smoothly.
In addition to the addition of Tensor Core and RT Core from 0 to 1, the RTX 3050 is also a huge improvement over the GTX 1650 in other aspects, the process process has been increased from 12nm to 8nm, and the number of CUDA units has increased from 896 to 2560, which is nearly 3 times that of gtX 1650! Texture units increased from 56 to 80, while raster units remained unchanged at 32.
The video memory part of the RTX 3050 compared to the GTX 1650 is a huge improvement, the memory type evolved from GDDR5 to GDDR6, the memory capacity increased from 4GB to 8GB, in terms of the memory bandwidth that best reflects the video memory performance, from 128GB/s to 224GB/s, which is 1.75 times that of the GTX 1650.
The number of CUDA units has increased to nearly 3 times, the memory bandwidth has increased to 1.75 times, and the third generation of Tensor Core and the second generation of RT Core have also been added, and the clock frequency has also increased by a certain range, and subsequently, its price and power consumption have also increased by a certain range, the new RTX 3050 is sold at a suggested retail price of 1899 yuan, the entry-level graphics card has come to nearly two thousand yuan, and the days of buying an entry-level game graphics card for hundreds of thousands of dollars have passed. However, counting the improvement of its performance, in fact, the cost performance has increased a lot.
In terms of power consumption, the power consumption of the previous GTX 1650 public version scheme is 75W, theoretically can make a graphics card without external power supply, but it may be affected by market preferences, the actual DIY market GTX 1650 graphics card almost did not see the game graphics card without external power supply, and the power consumption of this generation of RTX 3050 public version of the scheme was increased to 130W, theoretically, the power supply interface of 6pin is enough, but the actual estimate is also difficult to see, Most graphics cards should be equipped with an 8pin power supply interface.
Finally, compared with the RTX 3060, in addition to the frequency, most of the specifications of the RTX 3050 are about 70% of the RTX 3060, as the entry of the 50 series and the dessert positioning of the 60 series, compared with the traditional impression, the gap is actually small, indicating that the performance of the entry RTX 3050 is still very good, but this gap is actually similar to the previous generation of GTX 1650 and GTX 1660.
The GeForce RTX 3050 graphics card based on the new generation ampere architecture also has some common features of this generation of Ampere graphics cards, and is a great improvement over the previous generation of Turing architecture.
The NVIDIA Ampere Architecture introduces the evolution of SM units
The biggest change from the SM unit on the RTX 30 Series graphics card compared to the RTX 20 Series is the doubling of the FP32 unit for traditional computing, the introduction of the second generation RT Core, and the third generation Tensor Core.
As we all know, in the Turing architecture, NVIDIA integer (INT32) and single-precision floating-point (FP32) two different data types are handed over to two different ALU for computation. However, the most common in modern gaming applications is FP32, so in order to improve computing efficiency, NVIDIA has introduced a new ALU that can support both INT32 or FP32 data types on the NVIDIA Ampere architecture. That is, there are now two different data paths, one that can handle integers or single-precision floating-point at the same time, and the other that simply handles single-precision floating-point calculations.
The dedicated hardware unit RT Core, which performs real-time ray tracing operations, has also been updated to the second generation on the NVIDIA Ampere architecture, most notably with the addition of dynamic blur accelerated computing support. NVIDIA's new interpolation algorithm can improve the efficiency of real-time ray tracing while ensuring the accuracy of motion blur, and the official said that it can reach up to 8 times the speed of the previous generation. In addition, in the basic BVH calculation, the second generation of RT Core can also be 2 times faster than the first generation.
Tensor Core, the hardware unit responsible for running AI computing, has also been upgraded to the third generation on the NVIDIA Ampere architecture. In fact, the previously released A100 computing card has used the new third-generation Tensor Core, which can provide 4 times more performance than the second-generation Tensor Core, but the Tensor Core on the game card has been streamlined to a certain extent, and the throughput of its FP16 FMA calculation is only half of the Tensor Core in the GA100 core.
And what is the use of the more powerful AI operations brought by the third generation Tensor Core? The answer is DLSS. The full name of DLSS is Deep Learning Super Sampling, which translates to Chinese is deep learning super sampling, which is to improve the frame rate by reducing the rendering resolution in the game and improving the display resolution with stretching, such as 1080P rendering resolution and 4K (2160P) display resolution.
In addition, the idea of handing different types of computations to different units was adopted from the NVIDIA Volta architecture, when tensor Core diverted a lot of AI-related operations, and later RT Core shunted real-time ray-tracing-related calculations. However, on the Turing architecture GPU, not all operations can be performed in parallel, to the Ampere architecture, NVIDIA has improved the parallelism between various units inside the GPU, and now the three major units of traditional computing units, RT Core and Tensor Core can work at the same time, and continue to shorten the frame rendering time on the original basis.
Then there's HDMI 2.1, the highly anticipated new output port. Under HDMI 2.1, graphics cards can achieve an output of 8K60Hz or 4K120Hz with a single cable. It's a good boon for players who want to play games on a large-size TV.
NVIDIA Reflex
Released along with the RTX 30 series graphics cards, there's something new that's important for esports games, or more importantly for esports players, and that's NVIDIA Reflex. So what exactly is this NVIDIA Reflex? In fact, it is divided into two parts, one part is hardware and the other is software.
The hardware part is the Reflex Latency Analyzer, which can actually be seen as an advanced version of the LDAT, which is pre-installed directly in the monitor and can be used to measure the time difference between the player's mouse click and the change in the picture, that is, all the delays of the entire system.
The software part is the NVIDIA Reflex SDK. The NVIDIA Reflex SDK reduces and measures rendering latency, and developers can integrate it directly into the game. When its low-latency mode is turned on, the CPU can be synchronized with the graphics card, which greatly reduces the rendering sequence, thereby reducing rendering latency.
NVIDIA Broadcast
NVIDIA Broadcast is for live streamers, which uses the AI capabilities of RTX GPUs to remove or replace the background of live streamers, as well as camera reconstructions, and even helps microphones to remove background noise.
After installing NVIDIA Broadcast software, it will establish an intermediary role between the camera, headset and live broadcast software, so that external devices can use the AI capabilities of the RTX GPU to do some AI enhancement effects, headphones and microphones now support noise reduction, AI will analyze which is the main audio, which is the background noise for noise reduction, to the live broadcaster and the audience to present a clear, useful sound.
The camera now has automatic reconstruction of the map and background processing capabilities, from the camera acquisition to the picture, can be set through the Broadcast processing, and then transmitted to OBS these live broadcast software, which can make the background of the live broadcast owner more vivid and flexible, but also can reduce the construction cost of the live broadcast scene.
Asus ROG STRIX RTX 3050 O8G GAMING graphics card exterior
RTX 3050 does not have a founder version, all are non-public shipments, this time to do the first review is an Asus ROG STRIX RTX 3050 O8G GAMING, as an entry-level model graphics card, Asus ROG STRIX RTX RTX 3050 O8G GAMING looks actually very powerful. This is related to its high-end series of ROG STRIX.
Asus ROG STRIX RTX 3050 O8G GAMING overall is very large, with a distinct over-the-shoulder design, the front is three large size axial fans, all 90mm specifications, all equipped with 9 fan blades, with its barrier ring design, you can blow directly to the graphics card heatsink air volume greatly increased.
Look closely at the fan blades, you can see that the fan blade of the central fan is opposite to the direction of the fans on both sides, and this design of the adjacent fan is forward and reverse can reduce the air turbulence during the operation of the graphics card heatsink to improve the heat dissipation efficiency. In addition to enhancing heat dissipation, it's also important to reduce unwanted noise, the ASUS ROG STRIX RTX 3050 O8G GAMING has a standby shutdown feature where the cooling fan remains stationary for absolute silence when the core temperature is low.
Turning to the back of the graphics card, the ASUS ROG STRIX RTX 3050 O8G GAMING has a longer metal backplane than its PCB, which completely covers the PCB, and also forms a penetrating ventilation design at the tail of the graphics card, so that the hot air through the heat dissipation fin can be blown to the air duct of the chassis fan, making the cooling air duct of the entire chassis more reasonable.
And asus ROG STRIX RTX 3050 O8G GAMING is equipped with two sets of fan tuning schemes, one performance mode, one is silent mode, players can adjust through the small paddle here on the back panel, without having to set it through software, which is very convenient.
I / O bezel on this side, ASUS ROG STRIX RTX 3050 O8G GAMING uses a 304 stainless steel bracket, and adopts a more luxurious video output interface design than the public version, a total of 2 HDMI interfaces and 3 DP interfaces, HDMI interface is version 2.1, can support 8K 60FPS video output.
Compared to this generation of ROG STRIX series graphics cards with other higher-end GPU models, the ASUS ROG STRIX RTX 3050 O8G GAMING has a slight reduction in fan diameter and number of fans, but compared to other graphics cards that also wear RTX 3050 GPUs, the ASUS ROG STRIX RTX 3050 O8G GAMING is a very high-end configuration.
Asus ROG STRIX RTX 3050 O8G GAMING graphics card disassembly
After disassembly, it can be clearly seen that the part of the heat sink in contact with the GPU of the ASUS ROG STRIX RTX 3050 O8G GAMING adopts a mirror polishing treatment to maximize the flatness to make its contact with the GPU core present the tightest state, increase the thermal conductivity efficiency, and improve the heat dissipation effect.
The heat pipe connecting the pure copper base with the large-area heat dissipation fin has a total of 5, all of which are 6mm specifications, and are nickel-plated, which can prevent oxidation very well. Many high-heat electronic components are covered with high thermal conductivity silicone pastes to increase the thermal conductivity effect.
Asus graphics card uses a fully automatic production line, so the PCB is very clean, the most conspicuous on the PCB is definitely the RTX 3050 GA106-150 core, it is surrounded by 4 memory particles, all of which are Micron's D9ZPM memory, which is GDDR6 memory, equivalent data rate 14Gbps, single 2GB capacity, 32bit bit width, a total of 4 pieces of 8GB capacity and 128 bit width.
In terms of power supply, the card adopts 8 phases of power supply, of which 6 phases are powered to GA106-150 core, 2 phases are powered to video memory, and the MosFET powered by the core and video memory is SiC654A from Viser Semiconductor, and the PWM master can see a uP9512R.
Test platform with description
As can be seen from the GPU-Z, the base frequency of the ASUS ROG STRIX RTX 3050 O8G GAMING graphics card is 1552MHz, the Boost frequency is 1860MHz, the Boost frequency has a large pre-overclocking compared to the public version of the scheme, the video memory equivalent data rate is 14Gbps, and the FAN of the ASUS ROG STRIX RTX 3050 O8G GAMING has a standby shutdown function. From the GPU-Z sensor monitoring page, you can see that the fan speed is zero in standby.
In addition, the default temperature limit of the ASUS ROG STRIX RTX 3050 O8G GAMING is 83 degrees Celsius, players can choose to unlock up to 90 degrees Celsius, because the Boost frequency has a large pre-overclocking compared to the public version, the card power consumption limit is 150W by default, higher than the public version of 130W, and players can also choose to unlock up to a maximum of 165W.
Test platform as shown in the table above, the motherboard uses the MSI PRO Z690-A DDR4 motherboard, the memory with two XPG Longyao D50 DDR4-3600 16GB to form a dual channel, of which BIOS settings, in addition to turning on XMP to let the memory run at 3600MHz frequency, while opening the support of Resibility BAR in the motherboard BIOS, the others use default settings.
The graphics cards representing each GPU model are: XFX Vision RX 5500 XT Black Wolf Edition, 4GB video memory version, Boost frequency 1845MHz, XFX Vision RX 6500 XT Black Wolf Edition, XFX Vision RX 6600 Black Wolf Edition, Boost frequency 2491MHz, EVGA GTX 1650 XC ULTRA, 4GB GDDR5 video memory, Boost frequency 1860MHz, ASUS ROG STRIX RTX 3050 O8G GAMING, Boost frequency 1860MHz; iGame RTX 3060 Mini OC 12G L, using public BIOS file, Boost frequency 1777MHz.
3DMark theoretical performance testing
We used 3DMark as a graphics card benchmark, including Fire Strike, Fire Strike Extreme, Fire Strike Ultra, Time Spy, Time Spy Extreme, and Port Royal. Among them, Fire Strike, Fire Strike Extreme, and Fire Strike Ultra test the performance index of the graphics card at 1080p resolution, 2K resolution and 4K resolution in DX11 games, respectively, and the two projects of Time Spy and Time Spy Extreme are the performance index of graphics cards in DX12 games at 2K resolution and 4K resolution, Port Royal is the performance index of real-time ray tracing of the tested graphics card, the specific results are shown in the following table, and the scores listed in the table are the scores of the 3DMark graphics card individual items.
Due to the 4GB of graphics memory in the graphics card participating in the test, and the performance is weak, the score of the project with relatively high rendering resolution may fluctuate greatly, and the score of the FireStrike project is the most referenced. The results of theoretical tests can be considered that the RTX 3050 has at least a 70% performance improvement over the GTX 1650, and the former is relatively even greater in the face of insufficient 4GB of video memory.
Compared to the RTX 3060, the RTX 3050 has about 72% of the performance of the RTX 3060, and since their 8GB and 12GB of video memory capacity are sufficient, the gap in each test project is about this number.
Compared with the new card RX 6500 XT just released by competitor AMD, or pay attention to the FireStrike project, the RTX 3050 seems to be only 5.6% higher than the RX 6500 XT, but we know that the theoretical test score of this generation of A card is generally higher than the actual score of its game, I believe that the actual game score of the RTX 3050 will still be higher than the RX 6500 XT by more than 5.6%, not only that, RTX The 3050 has 8GB of video memory, which is much more abundant than the RX 6500 XT, which will also make it perform better in games with larger memory needs.
However, since the graphics card representing RTX 3050 is Asus ROG STRIX RTX 3050 O8G GAMING, this card is pre-overclocked compared to the frequency scheme of the public RTX 3050, so the performance results are slightly improved compared to the frequency scheme of the public RTX 3050, because the RTX 3050 does not have a public version, the specific improvement is not easy to estimate, the author may increase by about 2% according to experience, which is also to be considered.
1080P resolution real game test traditional rasterized game picture quality comparison test
Considering the purpose of RTX 3050, this part of the test will set the game quality to the preset in the high quality, do not turn on ray tracing, DLSS and FSR, the default is not full-screen manual change to full screen, the default open vertical synchronization of the manual off vertical synchronization, in addition to the other options are default settings, all use the game's own Benchmark output results.
According to the actual measurement results of many games, the RTX 3050 has increased the average frame rate of the game by up to 77.9% compared to the GTX 1650, up to 34.6% compared to amD's previous generation RX 5500 XT, and 41% higher than the RX 6500 XT just released by AMD. That's about 75% of the RTX 3060 and RX 6600.
Real-time ray-traced image quality comparison test
This part of the test will set the game quality to the high quality in the preset, and turn on ray tracing to "high", the default is not full-screen manual to full screen, the default is turned on the vertical synchronization of the manual off the vertical synchronization, in addition to the other options are default settings, all use the game's own Benchmark output results.
If the game turns on ray tracing, the average frame count of RTX 3050 is 73.3% of RTX 3060, which is similar to the gap under the traditional rasterization, and compared with the RX 6500 XT, it is more complicated, which involves the RX 6500 XT itself has weak light tracking performance and only 4GB of video memory The more the 6500 XT, even with the RX 6600 there are back and forth.
In general, it can be considered that the RX 6500 XT cannot play AAA masterpieces without the help of FSR and other technologies under the ray tracing effect, while the RTX 3050 has a certain playability, and in the case of high performance requirements, the light chasing game performance of the RTX 3050 is even stronger than that of the RX 6600, and in the case of low performance requirements, the light chasing game performance is weaker than the RX 6600.
In terms of game measurement, due to the large gap with the comparison graphics card, although the author believes that the frequency scheme of the ASUS ROG STRIX RTX RTX 3050 O8G GAMING may be increased by about 2% compared to the public RTX 3050, and it has no impact on the conclusion under such a large gap.
RTX 3050 delightful gaming capabilities
In the above game test, on the one hand, there are 4GB of video memory in the contrasting graphics cards, which will not have enough video memory under the preset maximum image quality of today's AAA masterpieces, and some will cause the test results to fluctuate greatly, so considering that all the traditional rasterized game screens are set to high for testing, but it can be seen that the RTX 3050 can be said to be more than enough under the preset high quality, or can continue to challenge higher image quality.
On the other hand, in the game testing part of real-time ray tracing, although the N card has a DLSS that is not supported by the A card, but for the sake of fairness (rendering the same picture quality), the N card does not turn on DLSS, and does not fully reflect the value of the N card. In summary, the author has conducted a series of tests on the RTX 3050.
The traditional rasterized game screen tested below is set to the preset maximum image quality, and the ray tracing "high" setting is turned on for real-time ray tracing, DLSS is turned on for DLSS technology, and DLSS for DLSS 2.1 and above is turned on to "performance mode" (in the games under test, except for Dust 5 and Far Cry 6, several other games support DLSS technology).
In addition, the default is not full-screen manual change to full-screen, the default open vertical synchronization manually turn off vertical synchronization, in addition to the above other options are default settings, and all use the game's own Benchmark output results are as follows:
Through the test, we can see that under the traditional rasterized game screen, even if it is a very new AAA masterpiece today, most of the RTX 3050 can still run smoothly with the highest image quality, and after using DLSS, the ray tracing can also reach a fluency of about 60 frames, which is very good.
Temperature performance: full load up to 56.7 degrees Celsius
The temperature test is carried out in the environment of the bare metal platform, the ambient temperature is about 19.7 °C, using the 3DMark Time Spy pressure test as a full load project, the data is recorded by the GPU-Z Log to File function, the following is the test curve of the full load temperature:
ASUS ROG STRIX RTX 3050 O8G GAMING has a fan standby shutdown function, so the standby temperature tested above is the temperature under the fan shutdown, although the fan is stopped, the card's standby temperature is also very low, the maximum standby temperature of the ten minutes of standby is only 35.9 degrees Celsius, and the hot spot temperature is also up to 48.4 degrees Celsius, which is very cool and does not trigger the fan to actively dissipate heat.
Under full load, the maximum temperature of the card is only 56.7 ° C, and this maximum temperature is before the active heat dissipation of the fan is triggered, and the temperature after the fan is rotated is lower than this. If you pay attention to the hottest temperature, it is only 68.4 degrees Celsius, which is also before triggering the fan to actively dissipate heat.
Fully loaded GPU core actual frequency: Maintained at 2010MHz
During the 3DMark Time Spy stress test for temperature testing, we also recorded the actual frequency of the GPU core of the graphics card in the fully loaded state as shown in the following figure:
After stabilizing at full load, the actual GPU core frequency of the ASUS ROG STRIX RTX 3050 O8G GAMING has been stable at 2010MHz, which is much higher than its Boost frequency, and it is rare that the core frequency graph is so stable.
Power consumption test: The average power consumption at full load is 134W
Through the PCAT kit in our hands, we can accurately measure the number of graphics card PCI-E and external power interface watts, the maximum power consumption of the graphics card is obtained in the 3DMark Time Spy stress test, and the standby power consumption is recorded for 1 minute after entering the system to take the average.
The results of the statistical power test calculate that the average standby power consumption of the asus ROG STRIX RTX 3050 O8G GAMING is 12.7W, the average power consumption of full load is 134W, and the peak power consumption is over 146. For such a level of graphics card power consumption, a 550W power supply is enough.
The official nominal power consumption of the RTX 3050 is 130W, and the Asus ROG STRIX RTX RTX 3050 O8G GAMING is slightly more than 130W due to the Boost frequency compared to the pre-overclocking of the public version, considering that the ASUS ROG STRIX RTX RTX 3050 O8G GAMING has a relatively large overclocking compared to the public frequency scheme, and here is no longer a separate overclocking test of this non-public version of the RTX 3050.
Summary: Flawless, entry-level graphics cards can also be smoothly chased
For the GeForce RTX 3050, I'm willing to praise it for its flawlessness, especially compared to the Radeon RX 6500 XT card that AMD just released. First of all, the standard memory configuration of geForce RTX 3050 is 8GB, which can open the highest image quality at 1080p resolution without the embarrassment of insufficient video memory, and the extra margin can also allow players to cope with the game environment in the next few years.
The layout idea of N card large-capacity video memory can be seen at least from the GeForce RTX 3060, when the release of the GeForce RTX 3060 said that equipped with 12GB of video memory actually feels a little too much for gamers (relative to its performance can be competent in the game environment), and to the GeForce RTX 3050, the video memory naturally becomes 8GB, this video memory capacity with the performance of GeForce RTX 3050, Then it feels like a golden partner, very balanced.
Graphics Mini Ladder List (Full Graphics Ladder List)
Secondly, the game performance of GeForce RTX 3050 is also particularly satisfactory, first of all, the most surprising light chasing game performance performance, as a "50" level entry-level graphics card, in the 1080p high quality with high light chasing settings, GeForce RTX 3050 can actually be combined with its own DLSS technology in those extremely high performance requirements can also achieve more than frame fluency, such as "Bright Memory" to reach 62 frames, "Metro: Gone (Enhanced)" Up to 72.94 frames. For games like Dust 5 and Far Cry 6 that don't have a harsh light chase effect, it can be stable beyond 60 frames without the help of DLSS. And in the traditional rasterized game test, the GeForce RTX 3050 is a 41% improvement over the RX 6500 XT, which is far from the point of being left behind.
In addition, Radeon RX 6500 XT also comes standard with only 2 video outputs on the interface and cut the short board of video encoding, in this regard, GeForce RTX 3050 has shown the advantage of crushing, so Radeon RX 6500 XT can only have advantages in price and power consumption, however, relative to the experience it can give, the gap between 1599 yuan and 1899 yuan is not so large, and it is difficult to feel itchy in terms of power consumption of 107W and 130W.
In the current market, it can be expected that GeForce RTX 3050 will be a big seller, and it seems inevitable that the actual market price after the initial launch will be higher than the suggested retail price, and the estimated premium will be quite a lot, depending on the supply situation.