GCN 3.0 | KitGuru https://www.kitguru.net KitGuru.net - Tech News | Hardware News | Hardware Reviews | IOS | Mobile | Gaming | Graphics Cards Wed, 08 Oct 2014 01:42:48 +0000 en-US hourly 1 https://wordpress.org/?v=6.4.3 https://www.kitguru.net/wp-content/uploads/2021/06/cropped-KITGURU-Light-Background-SQUARE2-32x32.png GCN 3.0 | KitGuru https://www.kitguru.net 32 32 AMD rumoured to unveil next-gen ‘Carrizo-L’ APUs this December https://www.kitguru.net/components/cpu/anton-shilov/amd-rumoured-to-unveil-next-gen-carrizo-l-apus-this-december/ https://www.kitguru.net/components/cpu/anton-shilov/amd-rumoured-to-unveil-next-gen-carrizo-l-apus-this-december/#comments Tue, 07 Oct 2014 22:35:16 +0000 http://www.kitguru.net/?p=215423 Advanced Micro Devices reportedly plans to release its next-generation code-named Carrizo-L accelerated processing units (APUs) already in December, according to a media report. If the information is correct, AMD’s competitive positions against Intel Corp. will get significantly better early in 2015. Traditionally AMD unveils new APUs early in calendar year and gradually increases their production …

The post AMD rumoured to unveil next-gen ‘Carrizo-L’ APUs this December first appeared on KitGuru.]]>
Advanced Micro Devices reportedly plans to release its next-generation code-named Carrizo-L accelerated processing units (APUs) already in December, according to a media report. If the information is correct, AMD’s competitive positions against Intel Corp. will get significantly better early in 2015.

Traditionally AMD unveils new APUs early in calendar year and gradually increases their production to address different segments of the market. With the code-named Carrizo family of APUs AMD wants to change its approach a bit. DigiTimes claims that AMD will release its Carrizo-L APUs for the entry-level notebook segment in December 2014, which will help the company to better compete against Intel’s new Celeron and Pentium microprocessors. Unfortunately, “fully-fledged” Carrizo APUs will only be launched next March.

AMD’s Carrizo APUs are based on the next-generation Excavator x86 micro-architecture and improved AMD Radeon graphics. The new chips will offer higher performance compared to existing offerings. Unfortunately, since Carrizo and Carrizo-L are expected to be produced using 28nm process technology, they will not be as energy-efficient as Intel’s code-named Broadwell microprocessors made using 14nm technology. Still, any new chip will improve AMD’s position on the market.

amd_carrizo_excavator_fusion_1

Carrizo-L APUs are projected to replace AMD's existing Beema and Mullins APUs for entry-level notebooks and tablets, according to the report. Keeping in mind that Carrizo-L is based on high-performance micro-architecture, they should provide a significant performance increase when compared to existing offerings.

It is not completely clear why AMD wants to release new entry-level APUs already in December, ahead of the Consumer Electronics Show in early January. Traditionally all hardware designers roll-out their novelties at CES.

Given the fact that the information comes from unofficial sources, it is possible that AMD will begin revenue shipments of the Carrizo-L this December, but will formally introduce the new chips at the CES trade show in early 2015.

AMD did not comment on the news-story.

Discuss on our Facebook page, HERE.

KitGuru Says: It is interesting that for some reason AMD decided not to proceed with APUs based on the Puma+ low-power/low-cost microarchitecture. Perhaps, in the light of the fact that Intel’s Broadwell chips offer both high-performance and very high energy efficiency, AMD simply has to use high-performance architecture for its low-power offerings to stay competitive.

The post AMD rumoured to unveil next-gen ‘Carrizo-L’ APUs this December first appeared on KitGuru.]]>
https://www.kitguru.net/components/cpu/anton-shilov/amd-rumoured-to-unveil-next-gen-carrizo-l-apus-this-december/feed/ 24
AMD to keep FM2+ socket till 2016, no DDR4 for AMD’s chips next year https://www.kitguru.net/components/apu-components/anton-shilov/amd-to-keep-fm2-socket-till-2016-no-ddr4-for-amds-chips-next-year/ https://www.kitguru.net/components/apu-components/anton-shilov/amd-to-keep-fm2-socket-till-2016-no-ddr4-for-amds-chips-next-year/#comments Fri, 25 Jul 2014 22:53:56 +0000 http://www.kitguru.net/?p=204832 Advanced Micro Devices has reportedly chosen to take a rather conservative approach to development of its desktop platforms. Instead of adding support of the DDR4 memory to its next-generation code-named “Carrizo” accelerated processing unit, the company decided to stick to the DDR3 and therefore the FM2+ socket. While this is a good news for owners …

The post AMD to keep FM2+ socket till 2016, no DDR4 for AMD’s chips next year first appeared on KitGuru.]]>
Advanced Micro Devices has reportedly chosen to take a rather conservative approach to development of its desktop platforms. Instead of adding support of the DDR4 memory to its next-generation code-named “Carrizo” accelerated processing unit, the company decided to stick to the DDR3 and therefore the FM2+ socket. While this is a good news for owners of current-generation platforms and PC makers, it also means that the evolution of AMD’s platforms will slowdown.

Earlier this year it transpired from a document aimed at developers that AMD’s upcoming “Carrizo” accelerated processing units (with Excavator x86 cores and Radeon “Volcanic Islands” GCN 3.0 graphics engine) supports both DDR3 and DDR4 types of memory. While it was clearly said that the “Carrizo” was compatible with the FM2+ platform, the support of DDR4 implied that there will be other platforms with the new memory support as well as other innovations.

amd_carrizo_excavator_fusion_1

Apparently, something has changed in AMD’s plans. According to Bitsandchips.it, the desktop Carrizo APUs from AMD will exclusively support the DDR3 memory and will thus fit only into the FM2+ sockets. As a result, the FM2+ will remain on the market till 2016, whereas AMD desktop platforms will not support the DDR4 memory for quite a while. Given the fact that AMD officially supports DDR3 at 2133MHz transfer-rate, the support for DDR4 memory would not provide it a lot of advantages in case of “Carrizo”, which is pretty similar to “Kaveri”.

From AMD’s presentation released in November, 2013, it is obvious that the “Carrizo”/”Toronto” silicon (“Toronto” is a server version of “Carrizo” SoC) does support DDR4 memory, hence, dropping its support from the desktop version of the APU was something AMD did intentionally. The main reason why AMD could do this is pretty clear: the company does not want to support two different desktop APU platforms. It is possible that server-class “Toronto” APUs due next year will support the new memory type.

amd_opteron_server_roadmap_2015

It looks like DDR4 memory will be supported only by AMD’s platforms due in 2016. Unfortunately, we know almost nothing about AMD’s plans for 2016, except the fact that by that time the company is expected to roll-out an all-new high-performance microprocessor architecture developed under supervision of the legendary Jim Keller.

AMD did not comment on the news-story.

Discuss on our Facebook page, HERE.

KitGuru Says: AMD is traditionally very conservative when it comes to support of new types of memory. Since the DDR3 will most likely be more affordable that the DDR4 next year, such approach is justified. However, this indirectly proves that AMD’s next-generation APUs will simply not need a new type of memory and their performance will not be limited by memory bandwidth simply because that performance will not be considerably higher compared to today’s APUs.

The post AMD to keep FM2+ socket till 2016, no DDR4 for AMD’s chips next year first appeared on KitGuru.]]>
https://www.kitguru.net/components/apu-components/anton-shilov/amd-to-keep-fm2-socket-till-2016-no-ddr4-for-amds-chips-next-year/feed/ 16
AMD ‘Carrizo’ specifications, block-diagram get published https://www.kitguru.net/components/cpu/anton-shilov/amd-carrizo-specifications-block-diagram-gets-published/ https://www.kitguru.net/components/cpu/anton-shilov/amd-carrizo-specifications-block-diagram-gets-published/#comments Thu, 17 Jul 2014 22:59:38 +0000 http://www.kitguru.net/?p=203513 The first specifications as well as a block-diagram of AMD’s next-generation code-named “Carrizo” accelerated processing unit (APU) have been published by a web-site. Despite expectations, the “Carrizo” will not support high-bandwidth memory (HBM) technologies, at least, in its mainstream implementation. Nonetheless, the new chip will carry a fair amount of innovations compared to the current-gen …

The post AMD ‘Carrizo’ specifications, block-diagram get published first appeared on KitGuru.]]>
The first specifications as well as a block-diagram of AMD’s next-generation code-named “Carrizo” accelerated processing unit (APU) have been published by a web-site. Despite expectations, the “Carrizo” will not support high-bandwidth memory (HBM) technologies, at least, in its mainstream implementation. Nonetheless, the new chip will carry a fair amount of innovations compared to the current-gen Kaveri APU. Unfortunately, there are some cut-downs too.

The structure

Based on an excerpt from an alleged AMD’s document published by Chinese VR-Zone web-site, AMD’s “Carrizo” chip will be structurally similar to its predecessors, which means that all performance improvements will be a result of architectural enhancements, not from increases of execution units. As expected, starting from the “Carrizo” accelerated processing unit AMD will offer system-on-chips (SoC) that will place both APU and core-logic set (FCH, Fusion controller hub) onto the same piece of substrate, which will help to shrink mobile and all-in-one PC form-factors.

amd_carrizo_excavator_fusion

AMD Fusion “Carrizo” accelerated processing unit will feature two dual-core AMD “Excavator” (XV) modules with 2MB L2 cache (1MB of cache per module, 512KB of L2 cache per core), AMD Radeon graphics based on the third-generation graphics core next architecture (GCN 3.0) architecture with eight compute units (512 stream processors in total), an integrated dual-channel DDR3 memory controller, a new special high-performance bus to connect x86 cores to graphics cores and DRAM, further HSA [heterogeneous system architecture] enhancements, PCI Express 3.0 controller as well as all-new multimedia and input/output engines.

AMD’s “Carrizo” SoCs will come with integrated I/O controller that will support PCI Express, USB 3.0, Serial ATA-6Gb/s, SD, GPIO, I2S, I2C, UART, an integrated clock-generator and so on. Such system-on-chips will have thermal-design power between 12W and 35W.

The Excavator and cut-down L2 cache

According to the published slide, AMD expects its “Excavator” cores to provide 30 per cent performance increase presumably compared to the current-gen “Steamroller” cores at 15W thermal design power. Typically such predictions have little to do with reality since they are based on virtual simulations, not real tests. However, keeping in mind significant architectural enhancements of the “Excavator” compared to the “Steamroller”, it is possible that the upcoming x86 cores from AMD will perform noticeably better compared to the current ones.

For some reason AMD decided to cut-down L2 cache of the new APUs to 2MB from 4MB on the contemporary accelerated processing units (even 15W quad-core A10 “Kaveri” chips feature 4MB of L2). Cutting-down L2 cache without adding large L3 cache will almost inevitably result in performance degradation in single-threaded applications.

amd_fusion_kaveri_steamroller_piledriver_excavator

The exact reason why AMD decided to decrease L2 cache sizes is unknown. Perhaps, this was done because the “Excavator” cores are considerably larger than the “Steamroller” cores and AMD needed to keep the die size of Carrizo small enough to ensure the same manufacturing costs, which is why it decided to reduce the level-two cache size. Maybe, AMD’s new high-speed interconnections and improved memory controller somehow reduce the importance of large caches.

Hopefully, AMD knows what it is doing and the “Carrizo” APU will not be slower than the “Kaveri” chip in certain applications that require big caches.

AMD Radeon and GCN 3

The forthcoming APU will also feature a new AMD Radeon graphics engine based on the third-generation GCN architecture. The new architecture will feature a number of enhancements, including fine-grain pre-emption for context switches (should increase efficiency of heterogeneous computing by making context switches more flexible) as well as delta-color compression (an image compression technique that encodes a row of pixels by replacing their colour information with values that represent difference between subsequent pixels [e.g. if one pixel in RGB format is 255:0:0, the other one is 255:1:1, the third one is 255:2:0, then with delta compression the sequence can be represented as 255:0:0, 0:1:1, 0:1:-1].

The GCN 3 is also projected to support DirectX 12 application programming interface, but it is unclear whether it will actually support all the new capabilities of the API, or will be just compatible with the API while supporting DirectX 11.2 feature level.

Multimedia, I/O and power management

The “Carrizo” APU will also feature enhanced multimedia, security and I/O capabilities.

The new chip will come with universal video decoder 6.0 (UVD6), video encoder engine 3.1 (VCE3.1) and audio co-processor 2 (ACP2). At present there is no word about support for ultra-high-definition (UVD) video codecs, such as H.265/HEVC or VP9. Looks like decoding of 4K video will not be a strong side of the new APU from AMD.

Despite of the fact that there will be no hardware support of UHD decoding, the Carrizo will support HDMI 2.0 output (which means it will be able to display 3840*2160 video with 120Hz refresh rate, stereo-3D UHD video and so on). The APU will support up to three displays.

In addition to TrustZone technology, AMD’s new chip will also support TPM 2.0, crypto acceleration and secure boot, the first time for AMD.

As expected, the “Carrizo” will also support advanced power management capabilities, including connected standby, STAPM, PPT/TDC/EDC tracking and so on.

Some thoughts

Since AMD’s “Carrizo” accelerated processing unit will be made using the same 28nm process technology as the “Kaveri”, it was illogical to expect AMD to significantly enhance its performance by increasing the number of x86 cores or stream processors because of the manufacturing costs reason. Unfortunately, improvements on the micro-architecture levels no longer bring breakthrough performance boosts. As a result, the “Carrizo” will not change the situation on the market of microprocessors any significantly. It will be somewhat faster than the predecessor (in applications that don't take advantage of large caches), but Intel’s “Broadwell” and “Skylake” processors will also be somewhat faster than the existing “Haswell” offerings.

Discuss on our Facebook page, HERE.

KitGuru Says: For several years now AMD is trying to change the rules of the game on the market of microprocessors. Instead of rapidly advancing its x86 micro-architecture, the company is trying to persuade the industry to use GPU compute technologies everywhere. Since x86 performance remains a very important aspect, AMD’s chips are gradually losing popularity among those, who need truly high performance in all applications.

The post AMD ‘Carrizo’ specifications, block-diagram get published first appeared on KitGuru.]]>
https://www.kitguru.net/components/cpu/anton-shilov/amd-carrizo-specifications-block-diagram-gets-published/feed/ 2