Categories
科技報導

The number of game frames skyrocketed & no need to read pictures?Talk about the black technology of PS5 and RTX3000



For gamers, it has been a series of heavy news recently. The new generation of game consoles, whether PS5 or Xbox SX, have announced their prices and will go on sale at the end of the year. In the PC field, Huang also offered a new nuclear bomb RTX3000 series, which is surprisingly cost-effective. These brand-new game hardware, in addition to surprises in terms of performance and price, the brand-new technology they bring will also completely change the experience of future games.

For gamers, it has been a series of heavy news recently. The new generation of game consoles, whether PS5 or Xbox SX, have announced their prices and will go on sale at the end of the year. In the PC field, Huang also offered a new nuclear bomb RTX3000 series, which is surprisingly cost-effective. These brand-new game hardware, in addition to surprises in terms of performance and price, the brand-new technology they bring will also completely change the experience of future games.

The number of game frames skyrocketed & no need to read pictures?Talk about the black technology of PS5 and RTX3000 1

Xbox SX and PS5 have already appeared, but do you know how they will change the game?

Regarding the gaming technology brought by the new hardware, the most mentioned may be ray tracing, which can bring more realistic pictures. However, for the overall experience of the game, it may be another technology that will have an even more earth-shaking impact. And this aspect of technology has been ignored by many players. Today, let’s talk about the new generation of game hardware, the innovation in game data transmission I/O mechanism, and what amazing changes will be brought to the game.

The big problem of today’s games: data I/O bottleneck

If you are a game player with a certain amount of experience, it should not be difficult to find that there has always been a big problem in the game for many years-you need to read the picture. In the past, this may not have been a big problem, but in the past ten years, more and more games have embraced the open world, maps have become bigger and bigger, and the time for reading images has become increasingly exaggerated. For some games, even in the SSD environment, the time for reading images is still enough for you to make a cup of tea, and the experience is quite crazy.

The number of game frames skyrocketed & no need to read pictures?Talk about the black technology of PS5 and RTX3000 2

Even if you switch to SSD, you still need to wait for a while to read the picture

This actually reflects the bottleneck of the game in terms of I/O. In the traditional process, the data is stored on the hard disk, and the game data must be read from the hard disk to the memory first, then distributed by the CPU, then transferred to the video memory, and finally processed by the GPU. It can be seen that game data needs to go through the lengthy transmission link of hard disk-memory-video memory, and in these links it also needs to rely on CPU scheduling, which is not efficient.

The number of game frames skyrocketed & no need to read pictures?Talk about the black technology of PS5 and RTX3000 3

The transfer process of game data is first read from the hard disk to the memory via the CPU, and then the CPU loads the data into the video memory and hands it to the GPU for rendering

As the game’s map becomes larger and more detailed, the bottleneck in game data I/O becomes more prominent. This mainly restricts the performance of the game in the following aspects.

·Read the picture.Naturally, the first thing that affects is the image reading mentioned earlier. As long as the map is large enough, even if SSDs are used, there will still be significant waiting time, so I won’t go into details.

The number of game frames skyrocketed & no need to read pictures?Talk about the black technology of PS5 and RTX3000 4

For games like “The Curse of Blood”, resurrection, clearance, and map change all require lengthy picture reading

As an ancient game mechanism, picture reading has been with players for a long time. But why does the game have to read pictures? Why should the game be made into different levels? Today, when the game data I/O bottleneck becomes more and more prominent, more and more people are beginning to think about this issue.

· Scene production.The I/O efficiency of game data is not high, which makes it impossible for game creators to call too much data at one time to create an ideal huge scene.

When the game is running, the data processed by the GPU comes from the video memory, and the capacity of the video memory and even the memory is not the same as the hard disk. The game maker must build a dojo in the snail shell and use several G data to construct all the game scenes in the current screen; otherwise, the data needs to be read from the hard disk again to replace the current video memory and internal memory.

In this way, the game scene must either sacrifice finesse or reduce the scale, or it will read pictures frequently-in fact, many games do need to read pictures frequently. In order to conceal the picture reading and avoid interruption of the game experience, game makers have adopted many clever methods. For example, in “Uncharted Seas”, when the player climbs a mountain for a period of time, there will be a cliff collapse behind the scenes, which is exciting and exciting at the same time. With the collapse, the old map was also deleted, the data in the video memory was emptied, and the data of the next scene was read. The process of reading the map was concealed by CG animation broadcasts and other means.

The number of game frames skyrocketed & no need to read pictures?Talk about the black technology of PS5 and RTX3000 5

The “Uncharted Seas” series of games will use methods such as collapse and explosion to delete old maps to free up memory capacity

But in any case, this is only a stopgap. Limited by the limited video memory and low I/O efficiency, it is always difficult for game scenes to expand as desired. Nowadays, with more and more detailed pictures and high-precision textures occupying more and more capacity, the production of game scenes is constrained by the bottleneck of game data I/O, which becomes more and more obvious. Many games have done a lot of scenes such as drilling caves, which is a last resort-drilling caves does not require much data to pile up the scene, and you can also use this opportunity to load the data of the next big scene. Want to fly over several cities in seconds like Superman? Sorry, the current game cannot do it.

·Maximum number of frames.In recent years, many high-refresh rate displays have appeared on the market, and the best have even piled up the refresh rate to 240Hz. However, it is not difficult to find that even if the game quality is reduced, the frame rate of many games still cannot reach this level, and the GPU occupancy rate at this time is actually not high, which is actually caused by the game data I/O bottleneck.

The number of game frames skyrocketed & no need to read pictures?Talk about the black technology of PS5 and RTX3000 6

The CPU frequency is at its maximum, but the GPU is far from being fully loaded, and the number of frames still cannot exceed 240. Part of the reason is that the CPU has become an I/O scheduling bottleneck.

As mentioned earlier, the transfer of game data from the hard disk to the memory and finally to the video memory requires the CPU to participate in the scheduling of these processes. When scheduling data at high frequencies, the CPU becomes a bottleneck, and the number of game frames cannot be further increased.

The game hardware of the next era often shoulders the mission of leading the game reform. This time PS5 and RTX3000 “ampere” GPUs are just like that.

PS5: Unique architecture breaks the I/O bottleneck

PS5: Unique architecture breaks the I/O bottleneck

Let me talk about PS5 first. Whether it is Xbox SX or PS5, SSD is used as a storage medium. However, the two sides are different. The PS5 architecture is more designed to break the bottleneck of game I/O.

According to current information, the SSD transfer rate of PS5 has reached 5.5 GB/s, but this is not critical. The key is that PS5’s game data reading is different from the traditional process. In the PS5 architecture, to a certain extent, the GPU can directly request the required game data from the SSD, which reduces the data transfer link and greatly improves the efficiency.

The number of game frames skyrocketed & no need to read pictures?Talk about the black technology of PS5 and RTX3000 7

The SSD of PS5 is very fast, but the real innovation lies in the architecture

How does PS5 do this? This needs to start with the current game data storage architecture.

The current game data is stored in the hard disk in a packaged form. A file of textures and various textures is often packaged together. When the game needs the corresponding data, it will be unpackaged and sent to the video memory, and then handed over to the GPU for rendering.

The number of game frames skyrocketed & no need to read pictures?Talk about the black technology of PS5 and RTX3000 8

The texture file in a PC game is actually a texture package, which needs to be transferred to the video memory and decompressed

As a result, the video memory has to store a large amount of texture data that may not be used, and each time a new scene is switched, it must go through a process of reading and unpacking. This is reflected in the image reading during the game.

Why should game data be packaged? In traditional HDD hard disks, unifying the scattered files can help reduce head seeks. This is the performance bottleneck of HDD hard disks. Packing game data into large files allows HDD to continuously transmit data and improve performance.

The number of game frames skyrocketed & no need to read pictures?Talk about the black technology of PS5 and RTX3000 9

PS4 and other machines still use HDD. If textures are not packaged and distributed for storage, the lengthy seek time of HDD will take up more time for image reading

Now that we have entered the SSD era, SSD does not have the concept of seeking. The reading speed of scattered files is more than a hundred times that of HDD. A more reasonable mechanism is that the GPU needs to read the textures directly. However, since the current game consoles PS4 and Xbox One still use HDD, and games can only be adapted to the mainstream HDD, the game’s approach is to pack the data together, which makes the SSD need to read a large amount of extra data, and does not play Should have expertise. When you switch from HDD to SSD, the change you feel is that image reading is faster, but image reading is always a barrier that cannot be overcome.

In PS5, the situation has finally changed. According to the information disclosed so far, for games developed for PS5, the data such as textures are no longer packaged centrally, but distributed (or packaged into smaller packages). The game can only read the required data to the video memory, which is very efficient. Promote.

The number of game frames skyrocketed & no need to read pictures?Talk about the black technology of PS5 and RTX3000 10

With the storage architecture of PS5, the GPU can read the data of the SSD more directly, and the master control of the SSD also provides the function of decompressing the texture package.

The unique design of PS5 has two advantages.

· Eliminate the process of reading pictures.Game data no longer needs to be packaged and read uniformly, and can be read in real time on demand, which is more flexible. There is no unified loading, which is the process of reading images;

·Remove the scene scale restriction.As mentioned earlier, the limited scale of game scenes is due to the difficulty of video memory to carry the massive data required for large scenes. In the new storage architecture of PS5, the data required by the scene is stored in the SSD and read in real time when it is actually needed. There is no need to load the data into the video memory all at once, which greatly reduces the burden of the video memory and makes it bigger. The production of the scene becomes possible.

The number of game frames skyrocketed & no need to read pictures?Talk about the black technology of PS5 and RTX3000 11

In PS5, the role of SSD is in a sense more like video memory

This is true in theory, but also true in practice. Sony previously demonstrated a new work of “Spider-Man” in the running PS5. In the demonstration, Spider-Man flies freely over tall buildings on the map. Although the map is very broad, PS5 basically does not need to read the map.

Can’t believe in Sony’s own demo? Let’s take a look at a third-party demo. Epic previously released a demo of the next-generation engine Unreal 5. This demo was run on the PS5 real machine. From it, you can see that the scene is very vast, and the characters can move in the scene at high speed, in one go, without any trace of reading the picture, very refreshing.

The official demo of the Unreal 5 engine, running on the PS5 real machine, you can jump directly to 7:50 to experience the high-speed movement of the seamless big map

The official demo of the Unreal 5 engine, running on the PS5 real machine, you can jump directly to 7:50 to experience the high-speed movement of the seamless big map

RTX3000 “Ampere” GPU: Revolutionary RTX IO

RTX3000 “Ampere” GPU: Groundbreaking RTX IO

Not only Sony has thought of improving the game data reading process, but Lao Huang also realized it. In the recently released RTX3000 series of “Amp” GPUs, DirectX Storage features are brought, which is expected to break the I/O bottleneck of PC games.

DirectX Storage is actually Microsoft’s new DirectX API, and NV has supported it in Ampere GPUs and introduced RTX IO technology. In simple terms, RTX IO allows PC game data transmission to bypass the CPU and directly transmit to the graphics memory for GPU use. This is actually similar to the design of PS5.

The number of game frames skyrocketed & no need to read pictures?Talk about the black technology of PS5 and RTX3000 12

DirectX Storage is optimized for SSD, whether it is a PC or a new generation of Xbox, will use this API

There is no doubt that RTX IO can greatly increase the efficiency of PC game data transmission. According to the data given by NV, the performance of decompressing game data through RTX IO and GPU even exceeds the transmission speed of SSD under PCIe Gen4. On the one hand, it can greatly reduce the time for image reading, and on the other hand, it can break through the CPU bottleneck and make the frame rate of PC games more “e-sports grade.”

As mentioned earlier, in many PC gaming games, the frame rate cannot be increased, but it is not a GPU problem. Observed through the software, the GPU occupancy rate is actually not high at all, but the CPU is already running at full frequency, and the bottleneck of the frame number lies in the CPU. The data scheduled by the CPU “cannot feed” the GPU? Now RTX IO can allow game data I/O to bypass the CPU, fundamentally solving this problem. In this way, the number of game frames has a higher upper limit. Today, as high refresh rate screens are gradually popularized, this technology is of great significance to both professional e-sports players and ordinary players.

The number of game frames skyrocketed & no need to read pictures?Talk about the black technology of PS5 and RTX3000 13

A bottleneck of the current PC game frame rate lies in the I/O system. Game data transmission needs to go through the CPU, which is already overwhelmed.

The number of game frames skyrocketed & no need to read pictures?Talk about the black technology of PS5 and RTX3000 14

Schematic diagram of RTX IO, data can be transferred directly from SSD to video memory

In addition, RTX IO can also bring the benefits of lifting the scene size limit, which is similar to the description in the PS5 paragraph above, so I won’t repeat it here.

to sum up

Many old players feel that in recent years, the game screen has not improved in quality. Is there a qualitative improvement in Crysis, which compares the screens of most games now? “Crysis” was born in 2007. Compare it with the game of Crysis five years ago, which is also in 2002, and compare it with the current game. It is not difficult to find the last ten years. The improvement in game picture quality is much slower than before.

The number of game frames skyrocketed & no need to read pictures?Talk about the black technology of PS5 and RTX3000 15

Most of the game screens compared to 2007 “Crysis”, there is no qualitative improvement, the popularization of ray tracing can change all this

The reason for this is that the graphics technology of the game has not been fundamentally innovative for a long time. The PS4 and Xbox One generation of game consoles have brought graphics technologies such as rasterization to a new level, but the rendering process of the game has not changed qualitatively. The normal map, ambient light occlusion, dynamic lighting, volumetric light, etc. that began to appear in the PS3/Xbox360 era are still the magic weapon to improve the quality of the game. Similarly, the rendering process of the game still remains in the tradition of reading data packets from the hard disk, transferring them to the memory and decompressing them, and then handing them over to the GPU for rendering. From this point of view, PS4 and Xbox One are actually a generation of game consoles that are rather conservative in design and lack groundbreaking.

In this generation, the turning point is finally here. GPUs that support ray tracing are installed in next-generation game consoles such as PS5, Xbox SX, and PC graphics cards, and SSDs have also become standard for game hardware. The game has undergone qualitative changes in all aspects, and finally has the preconditions.

Ray tracing has greatly increased the upper limit of the game screen, and changes in game data I/O can turn the game process upside down. I believe that with the popularization of the new generation of game hardware, more and more players will lament “the original game can still do this”, let’s look forward to it together.