AMD Official Answers Bulldozer Top Ten Hot Issues

Although many public relations people have been cut off, AMD’s connection with the player community has not been interrupted. Recently, media enthusiasts were asked by the media to solicit a series of questions about the bulldozer processor, and selected the most in more than 100. The ten concerned have responded. Although there is a lot of official rhetoric, it is still worth a look.

In general, AMD has repeatedly stressed that bulldozer architecture is a forward-looking design for future multi-threaded applications, but also more high-throughput server environment, rather than the ordinary desktop environment, so sometimes it seems to be "too grievances." Now.

Interestingly, AMD confirmed that the Windows 8 thread scheduler will be specifically optimized for bulldozers to increase its execution efficiency. Even Windows 7 will have an update in this area.

The next question and answer begins -

1. Why is the level 1 cache capacity of a bulldozer less than Phenom II?

AMD Academician and Chief Engineer of Bulldozer Mike Butler: For many years, AMD's processor architects have been analyzing and simulating different client and server workload data to efficiently control the cache capacity of all levels and achieve the highest performance. optimization.

On bulldozers, AMD uses a write-thru level one data cache, and has focused on improving prefetching algorithms and increasing the bandwidth of commonly used small-sized blocks. For those less commonly used, larger-sized block transfers, we rely on a larger 16-way associated 2MB L2 cache.

Based on today's general workload, we believe that this new design approach is even more efficient when processing data, even though the L1 cache is smaller.

2. Why is the performance of integer performance test much lower than AMD's previous quad-core?

Mike Butler: The bulldozer is a new architecture that is different in many ways. Bulldozers use both exclusive and shared resources to maximize efficiency, improve unit power consumption, and maintain IPC, including high-speed acceleration for unlocked desktops and low-voltage, low-power server versions. These are all previous generations. The architecture is incomparable.

Because of this, some individual tests will show different performance levels than previous core designs, some higher and some lower. The performance of multi-threaded applications has made significant progress, but some older test programs cannot benefit from the bulldozer architecture.

In addition, it needs to be pointed out that the bulldozer architecture is optimized for server throughput configuration. The two integer execution cores inside the bulldozer (per module) are designed to provide more efficient multi-threaded output per unit area and power consumption.

3, modular and core sharing design looks very exciting, but at the same time to improve the frequency of IPC is lower but like a step backward. how so?

Mike Butler: It is clear that the IPC is an important parameter for processor performance, and the first-generation bulldozers did have a slight drop. Bulldozers are a forward-looking architectural design. There are many parameters that reflect their performance, and some of them have compromises.

The new processor core frequency is higher, and basically maintains IPC. There are also multithreaded parallel computing performance, unit power consumption instructions, advanced acceleration capabilities, a new x86 instruction set, and powerful overclocking capabilities that are not available in previous generation architectures.

We believe that as operating systems and application software gradually take advantage of the inherent advantages of bulldozers, these enhancements will bring positive results to end users. Looking into the future, as the processing technology continues to mature, the potential for raising the frequency of the (bulldozer) core is also very large.

4, according to different evaluations, bulldozers seem lower than the Intel Sandy Bridge. How to explain the relative positioning with Intel?

Adam Kozak, Product Marketing Manager, AMD: AMD designs (bulldozers) based on the applications and environments customers use, and they will continue to use them in the future. The architecture emphasizes high frequency and resource sharing to achieve optimized throughput and speed in next-generation applications, high-resolution games.

This is a forward-looking, innovative processor design concept. Although it is difficult to measure with the old single-threaded tools, we believe that the FX processor can bring outstanding PC experience to our high-performance customers.

5. Has AMD promoted the development of more real multithreaded code? Is there a guide for developers to help them get the best possible multithreaded code performance?

Gabe Gravning, Senior Product Marketing Manager, AMD: We are working with developers and ISVs (Independent Software Developers) to encourage the development of multi-threaded code and applications. This is a major issue that AMD shares with other industry partners. It is also the main driving force for us to host the 2011 Developer Summit. It gives us the opportunity to sit down with developers and talk about parallel computing and heterogeneous computing.

Yes, we are promoting new (development) tools to make it easier for developers to make full use of the ever-increasing computing cores of CPUs and GPUs.

6. It is said that the Windows 8 scheduler will give bulldozers performance improvements. Can you explain this?

Gabe Gravning: We have been working with Microsoft to improve the thread scheduling of the bulldozer architecture in the Windows 8 system. Windows 7 inside, the workload is simply executed sequentially on each core, Windows 8 scheduler will be optimized for the bulldozer architecture, first the workload is assigned to each core pair (that is, each module), and then handed over to each Core, so there will be better thread performance.

For example, AMD's internal testing of the FX-8150 found that the Windows 8 Developer Preview was able to deliver up to a 10% performance improvement over Windows 7 in numerous games. Of course, different test results are certainly not the same.

We are also working with Microsoft on a scheduler upgrade for Windows 7, which is expected to be released shortly.

The $170 U.S. Phenom II X6 1090T is no worse than the $275 bulldozer in all the benchmarks and games I have seen, and sometimes even better. Why do I spend so much money?

Adam Kozak: We understand that users buy processors based on PC usage. In many cases Phenom II is really good. For those who need a more advanced architecture, need high-resolution PC games, and need to run new multi-threaded applications, AMD FX is a great upgrade.

8. What is the special or universal calculation that bulldozers are best at? For example, virtualization, Windows 8, card games?

Adam Kozak: High-definition video authoring, file processing, graphics processing, and high-resolution gaming performance at the default frequency (bulldozer) are very good. Most of them can use multiple cores, and some even begin to use the new instruction set introduced to further enhance performance.

9. Why increase the length of pipelines to achieve higher frequencies instead of improving efficiency?

Mike Butter: AMD and Friends of the latest architecture design have deeper pipeline. The pipeline inside the bulldozer increased by approximately 25% from the previous generation. Deeper pipelines are a key technological advance that can bring record frequency and performance improvements.

In addition, bulldozers can operate at a higher frequency at a specific voltage, which is not possible with other architectures, and therefore has a higher energy consumption ratio. We expect that performance will continue to increase over time, especially after the manufacturing process becomes more mature.

For example, in the parallelized DX11 games such as Civilization 5 and Metro 2033, the performance of the FX-8150 surpassed that of the Core i7-2600, leading them up to 18% and 8%, respectively.

It is also based on this design concept, AMD FX-8150's maximum acceleration frequency of 4.2GHz, compared to Phenom II X6 1100T 3.7GHz increased by 15%.

10. We have seen some tests of multi-threaded development of processors, and also learned that "Battlefield 3" and future games can use multi-threaded DX11. Given that bulldozers are extremely multi-threaded chips, is there a performance advantage over competing products in these areas? Have you collaborated with DICE or other game developers to make similar optimizations for future games and multi-threaded works?

Gabe Gravning: The problem is not that it will turn to multi-threaded applications. It is only a matter of time. AMD is driving the industry's turn with 64-bit computing and APU platforms.

Obviously, multi-threaded applications are the trend of the future, including the multi-threading technology inside DX11, and the bulldozer architecture is the bridge to this future. AMD has a great opportunity to lead the industry toward a multithreaded future, as we have done in the past. This of course also absolutely requires the cooperation of the software development community.