← 返回全部时间线
TheElec(韩国电子产业媒体)原文发布 09-24 09:27Radar 收录 09-24 10:16查看原文 ↗

[骁龙峰会] 高通以双轨策略推出骁龙8 Elite Extreme与标准版芯片

[Snapdragon Summit] Qualcomm Takes Two-Track Approach With Snapdragon 8 Elite Extreme, Standard Chips
中文摘要

高通在骁龙峰会上推出骁龙8 Elite Extreme与标准版两款芯片,采用两套独立设计,面向不同用户群体,并为智能手机厂商提供更大灵活性。原文未披露具体型号、制程、时间地点及供应链影响等细节。

原文深度解读

本文报道高通在骁龙峰会上发布骁龙8 Elite Extreme与标准版两款芯片,采用两套独立物理设计,面向不同用户群体,并为智能手机厂商提供更大灵活性。文章涉及高通多位高管在2026年9月22日于毛伊岛骁龙峰会上的发言,关键数字包括Extreme版多AI场景响应时间低于1秒、标准版约1.2秒、差距约20%,以及现有Adreno GPU具备3个矩阵加速器和18MB HPM。与电子元器件供应链的关系在于:两款芯片采用不同物理裸片、不同封装(Extreme版首次采用偏移PoP封装),并共享Oryon CPU IP,这些设计差异可能影响AP、DRAM封装、散热方案及代工与封测环节的供应链安排。

  • 高通移动手机业务高级副总裁兼总经理Chris Patrick在2026年9月22日毛伊岛骁龙峰会上表示,骁龙8 Elite Gen 6的Extreme版与非Extreme版具有非常不同的物理裸片,并非简单衍生SKU,而是从设计之初就分开设计的不同产品。
  • Extreme版在GPU扩展能力、AI加速器数量和封装方面具有优势,包括更多内部矩阵计算加速器和更大Adreno HPM容量;高通未披露具体增幅,而官方最新公布的Adreno GPU有3个矩阵加速器和18MB HPM,原文称Extreme版据信提升了这些规格。
  • Extreme版还包含用于Advanced Professional Video(APV)专业视频编解码器的专用硬件,以及Versatile Video Coding(VVC)解码器,以增强相机和多媒体处理性能。
  • 高通增加了Hexagon NPU中标量和向量加速器线程数量;线程是处理器接收和执行指令的最小单位,增加线程可同时处理更复杂工作负载。
  • Extreme版在处理多个AI场景同时运行时,将请求到完成的响应时间降至1秒以内;标准Elite版约需1.2秒,差距约20%。高通AI产品管理副总裁Vinesh Sukumar称,近期代理式AI提示超出简单命令,需要从单一请求执行多个动作,因此增加标量和向量线程以并行执行任务。
  • Extreme版首次引入偏移封装上封装(offset PoP)技术,将内存封装置于AP上表面侧边,使部分AP暴露,相比传统PoP垂直堆叠DRAM封装在AP封装上方、限制AP向上散热的方式,可显著改善散热并帮助维持芯片组性能;两款产品共享基于高通自研Oryon核心微架构的同一CPU IP,最新微架构应用于CPU及其子系统。

供应链影响

  • 由于Extreme版与非Extreme版采用不同物理裸片,可能意味着晶圆制造与封测环节需要分别安排产能与流程,具体影响取决于两款产品的实际出货比例。
  • Extreme版首次采用偏移PoP封装,可能改变DRAM封装与AP封装的堆叠及组装方式,并可能对散热材料或封装基板提出不同要求,具体取决于封装方案的量产成熟度。
  • Extreme版增加GPU矩阵加速器、HPM容量及NPU线程数量,可能提高对先进制程与封装产能的需求,但原文未披露具体制程与增量,实际影响取决于最终规格与良率。
  • 两款产品共享Oryon CPU IP,可能在一定程度上共用CPU相关设计资源与验证流程,但物理裸片不同仍可能使后端设计与制造环节保持独立。
  • Extreme版面向旗舰设备并强调AI与多媒体性能,标准版面向不同用户群体,可能使手机厂商在AP采购与产品分层上拥有更大灵活性,进而影响不同档位机型的元器件配置,具体取决于厂商采用策略。

系统解读,仅作为判断线索,不构成备货、出货、涨价或投资建议。

正文内容 · AI翻译

高通AI产品管理副总裁Vinesh Sukumar在毛伊岛骁龙峰会2026上发言。(照片:联合采访团)

高通已为其骁龙8 Elite Extreme和标准芯片组采用独立设计,面向不同用户群体,同时为智能手机制造商提供更大灵活性。

“骁龙8 Elite Gen 6的Extreme版和非Extreme版具有非常不同的物理裸片,”高通移动手机业务高级副总裁兼总经理Chris Patrick在9月22日于毛伊岛举行的骁龙峰会2026上表示。“它们不是简单的衍生SKU,而是从设计之初就分开设计的不同产品。”

这两款产品是作为智能手机大脑的应用处理器(AP),用于主要智能手机制造商的旗舰设备。Extreme型号在图形处理单元(GPU)扩展能力、人工智能(AI)加速器数量和封装方面具有优势。

额外的GPU能力包括更多内部矩阵计算加速器和更大的Adreno高性能内存(HPM)容量。高通未披露具体增幅。高通官方公布的最新Adreno GPU具有三个矩阵加速器和18MB HPM。据信Extreme型号提升了这些规格。

它还包括用于Advanced Professional Video(APV)专业视频编解码器的专用硬件,以及Versatile Video Coding(VVC)解码器,以增强相机和多媒体处理性能。

高通还增加了Hexagon神经处理单元(NPU)中标量和向量加速器线程的数量。线程是处理器接收和执行指令的最小单位。增加线程数量允许同时处理更复杂的工作负载。

因此,Extreme型号在同时处理多个AI场景时,将请求到完成的响应时间降至1秒以内。标准Elite型号约需1.2秒,差距约为20%。

“近期的代理式AI提示超出了简单命令,需要从单一请求执行多个动作,”高通AI产品管理副总裁Vinesh Sukumar表示。“我们增加了更多标量和向量线程,以并行同时执行这些任务。”

Extreme型号中使用的偏移PoP结构。(照片:联合采访团)

Extreme型号还首次引入偏移封装上封装(PoP)封装。与传统PoP相比,该技术改善了散热。

传统PoP将DRAM封装垂直堆叠在智能手机AP封装之上,限制了AP向上散热的能力。偏移PoP则将内存封装置于AP上表面的侧边。这使部分AP暴露,显著改善散热并帮助维持芯片组性能。

这两款产品也共享关键要素。两者使用相同的中央处理单元(CPU)知识产权(IP),基于高通内部设计的Oryon核心微架构。最新微架构应用于CPU及其子系统。

“标准版骁龙8 Elite Gen 6和Extreme版都使用具有最新微架构的下一代Oryon CPU,”高通产品管理高级总监Manju Varma表示。“我们将产品名称的重点从CPU代际数转向骁龙Elite平台,因此没有强调核心代际数字。”

Vinesh Sukumar, vice president of AI product management at Qualcomm, speaks at Snapdragon Summit 2026 in Maui. (Photo: Joint Press Corps)

Qualcomm has adopted separate designs for its Snapdragon 8 Elite Extreme and standard chipsets, targeting different user segments while giving smartphone makers greater flexibility.

“The Extreme and non-Extreme versions of Snapdragon 8 Elite Gen 6 have very different physical dies,” Chris Patrick, senior vice president and general manager of mobile handsets at Qualcomm, said at Snapdragon Summit 2026 in Maui on Sept. 22. “They are not simply derivative SKUs, but different products designed separately from the outset.”

The two products are application processors (APs) that serve as the brains of smartphones and are used in flagship devices from major smartphone makers. The Extreme model offers advantages in graphics processing unit (GPU) expansion capabilities, the number of artificial intelligence (AI) accelerators and packaging.

The additional GPU capabilities include more internal matrix compute accelerators and greater Adreno High Performance Memory (HPM) capacity. Qualcomm did not disclose the specific increases. The latest Adreno GPU officially announced by Qualcomm has three matrix accelerators and 18MB of HPM. The Extreme model is believed to have increased those specifications.

It also includes dedicated hardware for Advanced Professional Video (APV), a professional video codec, and a Versatile Video Coding (VVC) decoder to enhance camera and multimedia processing performance.

Qualcomm has also increased the number of scalar and vector accelerator threads in the Hexagon neural processing unit (NPU). A thread is the smallest unit through which a processor receives and executes instructions. Increasing the number of threads allows more complex workloads to be processed simultaneously.

As a result, the Extreme model reduces response time from request to completion to less than one second when handling multiple AI scenarios simultaneously. The standard Elite model takes about 1.2 seconds, a difference of roughly 20%.

“Recent agentic AI prompts go beyond simple commands and require multiple actions to be performed from a single request,” said Vinesh Sukumar, vice president of AI product management at Qualcomm. “We added more scalar and vector threads to execute those tasks simultaneously in parallel.”

The offset PoP structure used in the Extreme model. (Photo: Joint Press Corps)

The Extreme model also introduces offset package-on-package (PoP) packaging for the first time. The technology improves heat dissipation compared with conventional PoP.

Conventional PoP vertically stacks a DRAM package on top of a smartphone AP package, limiting the AP’s ability to dissipate heat upward. Offset PoP instead positions the memory package to the side of the AP’s upper surface. This leaves part of the AP exposed, significantly improving heat dissipation and helping sustain chipset performance.

The two products also share key elements. Both use the same central processing unit (CPU) intellectual property (IP), based on Qualcomm’s internally designed Oryon core microarchitecture. The latest microarchitecture is applied across the CPU and its subsystem.

“Both the standard Snapdragon 8 Elite Gen 6 and Extreme use the next-generation Oryon CPU with the latest microarchitecture,” said Manju Varma, senior director of product management at Qualcomm. “We shifted the focus of the product name from the CPU generation count to the Snapdragon Elite platform, so we did not emphasize the core generation number.”

↑