AMD has decided to reveal some details about the architecture of VEGA, the existence of which we knew with the schedule of the work published in March last year. At the moment, the company has made to lick the candy through the glass. As that before the official release of VEGA remained for several months – AMD promises in the first half of this year, we can expect exciting details we will be slowly metered through the corporate marketers. The purpose which puts before Vegą manufacturer, not gonienie competition. VEGA has a fit in the trends prevailing on the market of graphics processors and become the benchmark for other.
high-performance PROCESSORS have long ceased to be fanaberią rich players. Graphics processors are a key element of machine learning, and nobody is without them myself processing huge amounts of data. And the needs of the modern market of gaming or video processing is so high that it is not enough to have a large processing unit, it is necessary to select the appropriate memory architecture that will handle the transfer and storage of large amounts of information. Here there is room for VEGA.
Along with Vegą AMD introduces GPU its greatest change since the debut of the GCN architecture. New bone will be in the new computing unit with NCU name (Next-Generation Compute Unit). This is four 16-bit vector SIMD units, but is optimized to work with higher frequencies and greater number of instructions per cycle (IPC). One 32-bit ALU to handle two instructions szesnastobitowe. If such instructions are compatible type can be done simultaneously in the same clock cycle of the timer. Big surprise here support via VEGA format INT8, fast path processing ośmiobitowych integers are useful in applications related to neural networks. So, as we see, AMD decided to seriously challenge Nvidia and its Author. AMD does not reveal much about the organization of the NCU. We learn only that the NCU consists of 64 ALU that perform 128 operations with 32-bits per cycle hours.
r e k l a m a
it Seems that managed to get rid of the main limitations GCN: VEGA, to be able to distribute the load on working on more than four engines of the shaders simultaneously. This is followed by a new thread, geometry, promising twice the performance over the cycle from: know nothing except the assertion that the VEGA processing for 11 of the polygons in the cycle of hours in the four engines of the geometry.
improved engine geometry also mentions about the new shaderze primitives. His work will consist in the rapid resolution of all invisible or unnecessary primitives.
the Greatest power of VEGA, however, the new memory architecture. But rather a new incarnation of the already known High Bandwidth Memory (HBM), which debuted along with the Fiji GPU card Radeon R9. HBM2, because it is a speech, provides higher capacity and throughput than its predecessor our blessed. The new version provides a throughput of up to 2 GB/s, while 1024-bit bus width data given to us bandwidth up to 256GB/s to the bone. In addition, the specification HBM2 allows for the construction of bones, up to 8 GB. So we have not only twice the bandwidth, but eight times more capacity in comparison with the first version of HBM. This opens the way for the manufacture of graphic cards up to 32 GB of fast RAM. It is, however, not all changes in the memory subsystem.
the VEGA will use a separate memory controller called High-Bandwidth Cache Controller (HBCC). It is to mediate between the cache memory of the GPU and video memory. By optimizing the architecture on the availability of best service to transfer large amounts of data, graphics using VEGA will be able to operate with different memory types capacity (up to 512 TB. AMD says here about the best scalable architecture GPU memory in the world. You should pay attention to this aspect of VEGA, since until now, graphics processors are better able to cope in situations where memory was not enough. What would you say the SSD works as a cache of the video card?
the above Mentioned 512 TB of memory or revolution on the field of memory management assume that AMD is preparing the architecture for the industry, with its huge external data, not for casual players. However, it should be noted here that the memory usage of the game called engineers from AMD to improve this aspect of graphics cards. During the tests with Fallout 4 and III games closes the door. it was noticed because both games alokowały two times more memory than actually used. And it was on the principle of because. Games have always worked, therefore it was not necessary to change anything. AMD decided, however, to change this and put great emphasis on the correct use of memory. In the near future will take advantage of this industry, but in the future a new philosophy can imprint their mark on the entertainment market. And don’t forget about virtualization. AMD only makes its first steps, but the same assumption that VEGA can use up to 512 TB of memory clearl y indicates that the creators of the new architecture are also interested using it in intense team work, where virtualization and the demand for space, these go hand in hand.
the Fact that AMD still showed looks very promising. The company is certainly interested in the market of processing vast amounts of data, machine learning, and networks neuronowymi. Don’t forget about virtualization and efficient graphics processing and consumer market. However, only the disclosure of more details about VEGA and its independent tests will tell us whether and how the new architecture will affect the entertainment market for consumers.
No comments:
Post a Comment