Intel Xeon 6 is the general name for the chip giant’s new generation of processors for servers, data centers, cloud computing, edge and AI, and these represent an important advance compared to the previous generation not only in terms of performance and efficiency, but also in terms of specialization.
This commitment in favor of specialization is very easy to understand. The Intel Xeon 6 are divided into two great models, Sierra Forest and Granite Rapids. This division is Intel’s response to the transformation of the professional sector and the new needs that are emerging within it.
Just think, for example, of the different needs derived from different workloads that occur with cloud computing or edge computing, and also those generated through consumer digital services, DevOps, network microservices, storage, virtualization, modeling and simulation, Big Data and other more recent ones. , such as AI and inference.
Tackling each of those workloads with the appropriate software is important, and Doing it with the most appropriate hardware is essential to achieve not only an optimal level of performance, but also to maximize the value of performance per euro invested and optimize efficiency values, keys that, as you may have imagined, will help us reduce the TCO (total cost of ownership). ).
The chip giant has been able to see this reality, and that is why it has divided the Intel Xeon 6 into two clearly differentiated models. Each of them is specialized in various types of workloads, and they perform at their best when used to address these workloads. Sierra Forest focuses more on multithread loads that require greater parallelization capacity, and Granite Rapids focuses more on raw power and performance per core.
Intel Xeon 6, a look at Sierra Forest
Intel Xeon 6 Sierra Forest processors use E cores, also known as high efficiency. These nuclei are based on the crestmont architecture and manufactured in Intel node 3, which is one of the most advanced that currently exists. This node represents an important advance in terms of performance and efficiency, and has been key to achieving a high core density.
Sierra Forest does not use HyperThreading technology from Intel, which means that each core can only work with one process or thread. This is characteristic of E nuclei, which also have a much smaller size than P nuclei, which means that They take up less space at the silicon level.
This smaller footprint, together with the use of the Intel node 3, which significantly reduces the size of the transistors, has allowed Intel to configure the Sierra Forest processors with up to 288 cores. A version with 144 cores will hit the market first, and later the version with 288 cores. Both can be mounted in one or two socket configurations, which means they will allow us to scale up to 576 cores.
The model with 144 cores uses a single CPU block, while the second model will have two interconnected CPU blocks through a low-latency, high-bandwidth system to maximize performance and avoid bottlenecks. Intel’s modular packaging architecture combines monolithic networking to create shortcuts with flexible modular routing and multi-line interconnection, adopting EmiB technology to extend high-speed interconnection across the entire package within a single package.
Key specifications, performance and key sectors
- Up to 288 E cores.
- Crestmont architecture, Intel node 3.
- I/O module manufactured in Intel node 7.
- 4 MB of shared L2 cache per block of 4 cores.
- Improved AVX2 instructions for AI and vector operations (2 x 128).
- 64 KB of instruction cache per core.
- 32KB ECC data cache.
- 256-instruction out-of-order execution engine.
- 16 FP32 FLOPs per clock cycle.
- VNNI, INT8, BF16 and FP16 support.
- CMPccXADD, LAM and LASS ISA instructions.
- Support AES 256-bit and 2,048-bit encryption.
- Compatible with the CXL 2.0 standard.
- Configurations of up to 2 sockets.
Intel Sierra Forest processors are designed to offer high value in performance per watt consumed, especially in tasks that require a high degree of parallelization. They can multiply performance by four compared to Intel Xeon 2 (second generation), and as we can see in the attached graphs, they are also capable of greatly surpassing the performance of Intel Xeon 5 (fifth generation) processors, thanks to the improvements they bring at the level of architecture, memory, connectivity, interface and storage. .
With Intel Xeon 6 Sierra Forest processors it is possible reduce consumption by up to 280 watts compared to the Intel Xeon 5, and with the default configuration. This translates into a lower cost of ownership, and significant savings that our company will notice month after month. These processors are aimed, above all, at:
- Cloud Computing.
- Services related to data and networks.
- Multimedia services and microservices.
- Unstructured databases and off-scale analysis.
- Computing at the edge.
- DevOps applications.
- Storage services and tasks.
The first batch of Intel Xeon 6 Sierra Forest processors will hit the market starting today, and will be made up of the models that we can see in the attached image. The base configuration will have 64 cores and will support DDR5 memory at 5,600 MT/s. The top of the range model will have 144 cores and will support DDR5 memory at 6,400 MT/s.
Intel Xeon 6 Granite Rapids, power
These processors use P cores based on the Redwood Cove architecture, which means that they are high-performance solutions and have the latest technologies in the sector. They are also manufactured on the Intel node 3, which has allowed us to improve efficiency and performance values, as well as the density of cores per socket.
This range is configured with up to 128 cores per socket. In this case we do have support for HyperThreading technology, which means that each core is capable of working with one process and one thread. Therefore, that 128-core configuration is capable of working with up to 256 threads, thanks to HT technology. This translates into an enormous parallelization capacity and a very great advance compared to the previous generation.
Intel has improved parallelization capacity thanks to an increase in the number of cores and threads, but also increased per-core performance and instruction-level supportwhich is key for demanding tasks, such as high-performance computing and AI.
These processors also adopt a modular design with up to three CPU blocks, that are interconnected through a high-performance, low-latency system. I will give you more information about it when I have it, since at the time of writing this article Intel has not gone into depth. All in all, I imagine that the chip giant will have started from the keys that we already know, and that I have told you about when talking about Sierra Forest.
Specifications, performance, target audience and release date
- Up to 128 P cores.
- Redwood Cove architecture, Intel node 3.
- I/O module manufactured in Intel node 7.
- 2 MB of exclusive L2 cache per core and two threads.
- AVX512 (2 x 512) and Intel AMX instructions for AI and vector operations.
- 64 KB of instruction cache per core.
- 48KB ECC data cache.
- 512-instruction out-of-order execution engine.
- 1,024 BF16/FP16 Flops and 2,048 INT8 Flops per clock cycle.
- VNNI, INT8, BF16 and FP16 support.
- CMPccXADD, LAM and LASS ISA instructions.
- Support AES 256-bit and 2,048-bit encryption.
- FP16 instructions supported on the Intel AMX engine.
- Supports RAM memory at a maximum of 6,400 MT/s and up to 12 memory channels.
- Compatible with the CXL 2.0 standard.
- Configurations of up to 8 sockets.
Intel ensures that the new Xeon 6 Granite Rapids processors are capable of double inference performance compared to Intel Xeon Gen 5, and that can multiply performance in HPC (high-performance computing) tasks by 2.3 times compared to that same generation. On average, general computing performance doubles compared to that generation.
The launch of the Intel Xeon 6 Granite Rapids It is scheduled for the third quarter of this year, which means they will arrive a little after the Sierra Forest. This generation of processors is mainly aimed at:
- Simulation and advanced modeling tasks.
- Big Data.
- Memory level analytics.
- Inference, generative AI and deep learning.
- CDN, multimedia and gaming networks.
- Edge computing focused on video and analytics.
- Virtualization.
I still do not have information about all the models that Intel will launch within the Xeon 6 Granite Rapids, but at the moment two versions are confirmed, one with 128 cores and 256 threads and another with 86 cores and 172 threads. These two versions will have, respectively, three and two CPU computing blocks, and will serve to shape lower models with a smaller number of active cores and threads.
