If some analysts predict AMD’s quick death, the Sunnyvale company itself is not going to bury itself clearly. Yes, the architecture of Zen, developed almost from scratch, is not yet ready, as well as the transition to a 14-nm process technology, but within the available technologies AMD is capable of much. At the time of publication of this material at the Computex 2015 conference, the AMD conference should be held, at which the data on the new Fiji schedule and the new APU Carrizo will be revealed. And if only rumors about Fiji reached us before, Carrizo AMD decided to devote a full conference, called Carrzio Tech Day and held in Munich, Germany, shortly before the official announcement of these chips. In turn, we would like to share details about Carrizo with you – which we will do in this material.
1. AMD Carrizo – is the the sixth generation mobile APU AMD . The sixth, because Brazos representatives of AMD for some reason do not want to take into account – they say, it’s different. Llano, Trinity, Kabini, Kaveri, Beema – and now Carrizo. On the pictograms, which are used to glue laptops from the inside, it will be indicated that the chips installed inside belong to the sixth generation – so AMD hopes not to mislead customers. In this case, the designations A8 and A10 new processors will inherit from their predecessors.
We turn to the interesting facts about new products: Carrizo are the world’s first APUs of the production class, fully placed on a single chip, whereas previously in chips of this class the graphics chip or the south bridge, if located on a single substrate with the processor, form of a separate crystal. Here, the north bridge, the Fusion Controller Hub (South Bridge), graphics and processor cores fit on a single crystal grown within the 28-nm process technology Global Foundries.
2. The cores of the general destination in the Carrizo have architecture Excavator . It represents the fourth iteration of the logical development of the ideas of Bulldozer architecture: first there was actually Bulldozer, then Piledriver, Steamroller and finally Excavator. According to AMD, this is also the last update of this architecture – then the dark horse Zen will not replace the most successful design. In terms of performance per megahertz, Excavator can not offer any supernatural improvements – according to AMD itself, the IPC (Instructions per Clock) in Excavator grew by 4-15% compared to the Steamroller.
Growth was achieved due to several factors. First, the first-level cache memory is doubled, while maintaining the same level of delays, which made pre-election more efficient. Secondly, the branch prediction unit has been greatly upgraded – the buffer size has grown by 50%, the number of records for tracked jump addresses has accordingly increased from 512 to 768. Also, the flush-to-zero reset of the floating point unit is accelerated. Third, the Excavator kernels support new instruction sets – AVX2, MOVBE, SMEP, BMI versions 1 and 2, which AMD processors could not previously use. Finally, in the fourth place, support for modern idle states is added, which saves energy when the load is low. But the most important thing is that the Excavator cores have become much smaller than Steamroller cores within the same process technology, which significantly reduced their power consumption and – not significantly – increased frequencies. About how it turned out, we’ll tell you just below. Well, here we note that in the Carrizo chips will be used four cores Excavator – two blocks of two – with 2 MB of total for each block cache L2.
3. The graphic part of in Carrizo is used graphics which itself AMD calls [GC] the third of the third generation . In the third generation, the architecture has undergone some changes, with most of which attentive readers are already familiar – in fact, this generation of GCN was used in the GPU Tonga (Radeon R9 285). Also, the integrated graphics core received 512 KB of its own second-level cache memory. Among other things, support for DirectX 12 (Level 12), improved performance with tessellation, color lossless compression, an updated set of ISA instructions, connectivity of CPU- and GPU-caches and a high-quality scaler are claimed. In Carrizo, the Radeon R7 graphics controller has 8 compute clusters, while the mobile versions of Kaveri had only six such blocks, that is, the Carrizo graphics core has 512 stream processors and is capable of delivering peak performance to 819 GFLOPS. Carrizo has three built-in display controllers and supports image output with resolutions up to and including 4K.
4. As for 4K, in Carrizo embedded video decoder UVD-6 allowing hardware ] decode 4K-video in format H.264 or H.265 (HEVC) with a frequency of 60 frames per second. According to AMD, this is the world’s first laptop chip with fully hardware H.265 decoding, which the Californians tried to back up with a practical example – in the demozone there were laptops on the chips of Carrizo and Broadwell, reproducing 4K-video HEVC. The laptop on Carrizo did not load the CPU even by a third and played the video quite smoothly, whereas the Broadwell device showed 80-100% CPU utilization and showed a slideshow.
In addition to H.264 and H.265, Universal Video Decoder 6 is capable of decoding 4K MJPEG, VC-1 / WMV Profile D, MPEG-2, MVC, AVC / H.264 and MPEG4 / DivX. Thanks to the increased four times bandwidth, he can spend as much less time to process the HD-frame, and all the rest of the time to stay in a state of low power consumption. According to AMD, only this ability allows to increase the battery life when watching a video for half an hour.
5. Carrizo – is the first APU, completely satisfying the specifications HSA 1.0 , established by the HSA Foundation. This means that the processor and graphics cores are equal and have absolutely the same access to all RAM (AMD calls this hUMA – heterogeneous uniform memory). Accordingly, AMD prefers to talk not about 4 processor and 8 graphics cores, but about 12 computational ones. Tasks can be quickly transferred between processors, and the processors themselves can quickly save their states, switch to different loads and also quickly return to the saved states later, which allows to execute in parallel more diverse tasks. In addition, Carrizo supports hardware acceleration of code processing in Java, Python and C ++ AMP.
Unfortunately, at this stage, very few real applications allow the use of HSA, so to demonstrate the capabilities of this technology, AMD has to write with varying success the software itself. However, Carrizo also works well with OpenCL 2.0, which allows you to accelerate the work in such popular applications as Adobe Photoshop CC, Adobe Premiere and Libre Office.
6. AMD states that “Carrizo is created to speed up Windows 10″ . In fact, this means that Windows Meida Player can use HEVC decoder, which can also be used to stream video through Skype for Business or streaming images from Xbox One to PC. Also Carrizo supports such features of Windows 10 as SECURE boot, InstantGo, TPM, bitlocker and ARM Trustzone applications.
7. In the chips Carrizo it is established one the nucleus ARM Cortex-A5 which, in fact, answers for the work of ARM Trustzone. He also allocated some amount of isolated permanent memory and static memory with random access. The Cortex-A5 core is complemented by a cryptographic coprocessor that allows to work with algorithms RSA, Sha and AES, use Zlib and RDRAND.
8. The design of the crystal Carrizo was designed with the using the technological libraries with the high density of transistors . This approach to the design of the semiconductor chip processor unit took over from ATI – it took a lot of time to implement it, but the result is obvious. The Carrizo crystal is based on the patterns used in the creation of graphic chips. It has 3.1 billion transistors against 2.4 billion from Kaveri (29% more), while the new chip by 23% is smaller by the area of its predecessor . Individual blocks managed to be reduced even more efficiently: for example, the scheduler of the floating point unit shrank by 38%, the FMAC operation block by 35%, the instruction cache controller also “lost weight” by 35%.
Let’s remind once again: all of the above happened within the same 28-nm process technology. This – in combination with architectural changes – allowed to reduce power consumption when idle from 4.5 W from reference Kaveri-based notebooks to 2.7 W from devices based on Carrizo (although here AMD does not take into account the 100% efficiency of the power supply, the loss energy in which are proportional to the total energy consumption of the system). In the ideal case, the power consumption of systems based on Carrizo is almost 2 times lower than that of the analogs on the Kaveri, and the performance is 1.5 times higher, which gives a 2.4-fold increase in performance per watt.
9. TDP Carrizo is 15W . It is at this level of energy consumption that the high density design reveals itself best. Moreover, the reduced transistors are characterized by lower leakage currents, which allows an additional increase in the frequency. According to AMD research, high-density design wins at the classical high-performance (this design underlies the nuclei of the previous generation of Steamroller) with TDP less than 20 W. However, it should be understood that high-density design does not allow the production of classic desktop processors with a relatively high frequency, not driven into the framework of low TDP. That’s why we do not see the Carrizo desktop variants.
10. Notebooks on based on The APU Carrizo is being prepared all major manufacturers ] mobile computers – Acer, ASUS, Dell, HP, Lenovo and Toshiba. On the Tech Day, you could only watch reference laptops made to order AMD itself. Unfortunately, they were not provided for testing.
⇡ # Conclusion
The above 10 facts will provide an overview of the APU Carrizo. We will take a closer look at the architecture of these chips when we turn to their practical study. According to AMD’s plans, this should happen around August-September this year. In the meantime, we can conclude that Carrizo is the evolutionary development of Kaveri, from which one should not expect significant performance improvements (according to our sources, the average growth is 10-15%). However, at the same time, the transition from Kaveri to Carrizo promises a very noticeable increase in the battery life of devices based on these chips (according to the same sources, you can expect to increase the battery life by about one and a half times). The main purpose of Carrizo – universal multimedia laptops, where you can work, and play online games, and watch movies. Let’s hope that AMD partners will release really interesting models without trying to save on what they do not need to save on.