Introduction to the Intel Xeon 6 6900P
We already had the opportunity to analyze in detail the architecture of the new Intel Xeon 6, now with two fundamental variants divided between efficiency and performance core processors. The E Series was launched a few weeks ago with processors with up to 144 cores per processor. The one we tested today is a processor that is close to these figures, with 128 cores, but with twice as many threads and a more powerful architecture with more accelerators and processing capacity per core.
He Intel® Xeon® 6980P Processor The one we analyze today has this configuration: 128 cores with 2 threads per core and, therefore, 256 process threads. A real beast that we can find in configurations of two processors per motherboard for a total of 512 process threads, 256 cores. Each processor supports up to 3 TB of RAM capacity in DIMM or RDIMM format with speeds of 6400 or 8800 MT/s respectively, mounted on 12 channels.
Each processor has all the accelerators of this generationat the rate of: Intel® QuickAssist Technology (QAT) 4 devices, Intel® Dynamic Load Balancer (DLB) 4 devices, Intel® Data Streaming Accelerator (DSA) 4 devices, Intel® In-memory Analytics Accelerator (IAA) 4 devices.
Regarding these processors and their architecture, we already had time to do a complete analysis with all the details. We have also been able to test a configuration with two efficient Intel® Xeon® 6780E Processorsforming a server with 288 process cores.
The P series has been oriented more towards more demanding applications, with also higher consumption, up to 500 W in the processor we tested today, in configurations, for now, of up to 128 cores. With constructions in this generation of up to three tiles for the P variants with up to 128 cores and up to two tiles for the efficient models that we will see with up to 288 cores. The frequencies, in the model we analyzed, are moderate, with turbo modes of up to 3.9 GHz and base of 2 GHz, and for its 128 cores, a stable 3.2 GHz mode under load.
All of these new processors are based on the Intel 3 manufacturing process.with communications tiles built using an Intel 7 process. The new processors are not only much faster, up to 2 times faster per watt in general computing than previous generations, with a completely renewed architecture and better oriented to specific applications, but They also present important technological improvements at a connective level.
Support more PCI Express 5.0 linesbut also CXL in its 2.0 variant, with up to 96 lines available in these processors. Some of them are also tremendously scalable, with configurations of up to 8 sockets per motherboard, not in the densest models, like this Intel Xeon 6980P, which is limited to configurations of two sockets per motherboard.
The practical objective of these models, which are now more specialized than in previous generations, is modeling, simulation, Big Data, AI, virtualization, video, Edge Analytics, etc.
The different SKUs available
In this new range, which was expected precisely for this time, of sixth generation Xeon processors in performance mode, Intel has put a total of 5 new processors on the market. They all share their mounting capacity in a maximum of two sockets per motherboard, a maximum of 12 memory channels, support for DDR5 6400 memory or 8000 MT/s MRDIMM memory. They also all have the four accelerators per category, 1024 Intel TDX keys per processor, 2 units AVX-5126 UPI connectors and up to 96 PCI Express 5.0/CXL 2.0 lanes.
The differences are found in the number of cores available, TDP consumption and, of course, maximum frequencies that the processor can access with all its cores under load. In turbo mode, for some cores, the maximum frequency is always 3.9 GHz. The new models range from 72 to 128 coresso they are not simple processors; Those are left for the 6700 ranges that are yet to arrive.
The data bus reaches 768 GBps bandwidth50% more than its brothers in the Xeon 6 6700 range, and also increase its connectivity with six links, 50% more, of type **Intel UPI** (Ultra Path Interconnect), all also with a 24 GT/s max link. All this makes them very advanced and much faster processors per watt, but also with outstanding connective potential for hybrid computing systems with, for example, AI accelerators.
The complete platform
This generation of processors enjoys a new platform called Venuecity. Unlike the models available for E series processors, this model is designed to cover additional features such as support for dual socket with 12 memory channels divided into 24 memory modules12 per socket.
The socket is also different, requiring an FCLGA7529 since these models are very different in their connective capabilities and memory support. More PCI Express lanes, more memory channels, more UPI interfaces, it all adds up and requires more pins on the socket.
These motherboards must also have sufficient phases and layers for these connective levels and to support up to 500 W of consumption per processor. Specifically, a 20-layer PCB, double what we find in a high-end desktop motherboard for very recent processors.
Among the little that it can provide us, we can find 20 PCI Express 3.0 lanes, 20 SATA 6 Gbps ports and some USB 2.0 and USB 3.0 ports. It also handles supporting Ethernet connectivity, typically a Gigabit port. The rest is provided by the processors, which have great connective capacity, as we saw before.
Normally we will see these processors supporting next-generation network interfaces, Ethernet or fiber optics, with up to 200 Gigabit of bandwidth. Its purpose, precisely, is to provide great support for this type of high-speed interfaces, both for the network and mass storage environments.
Our test server
The test server that we have enjoyed these days is based on a dream machine. Two processors have been installed on the Venuecity platform Intel Xeon 6980P with 128 cores per head256 threads per CPU, for a total of 512 process threads.
1.5 TB of RAM is added to this configurationin 24 DDR5 modules of 64 GB per unit, taking advantage of the 12 memory channels of both processors. Each of these processors costs a whopping $25,000 per 1,000 units.
Each of these processors consumes a peak of about 500W.although in our tests the general behavior is somewhat more efficient. The platform is also equipped with a simple storage system, based on a medium-performance M.2 disk with PCI Express 4.0 interface and 4 TB capacity. Access speed, about 7,000 MBps, nothing special really, but the objective is to test the processors.
The network connectivity of this BeechnutCity platform is top-notch, with a Mellanox MT2892 (now NVIDIA ConnectX-6 Dx) dual-interface 100 Gbps Ethernet chipset. This chipset allows the configuration of one 200 Gbps interface over Ethernet, which certainly impresses, but it can also be implemented in two 100 Gbps interfaces that should, properly configured, offer the same performance, or load balancing and fault tolerance , maintaining a truly impressive transfer capacity.
Performance testing
In our tests we have added its direct competition, also the latest processors Xeon 6 efficient cores which we also had the opportunity to try. In general, it consumes less than those 500 W, according to our efficiency tests, also producing excellent performance in all tests. Our platform is based on Linux, specifically Ubuntu 24.04 LTS.
Linux Kernel Compilation 6.8. Seconds, less is better.
Node.js Compilation 21.7.2. less is better
LLVM Compilation 16.0. less is better
OpenSSL 3.3 SHA256.
John The Ripper: WPA PSK
PostgreSQL 16.
RocksDB 9.
CoreMark 1
QuantLib 1.32
Blender 4.1. less is better
OSPRay 3.1
uvg266 0.4.1
StockFish 16.1
Average consumption (W).
Conclusion
The advances of these new processors, taking advantage of new tile architectures and new manufacturing processes, allow Intel to develop highly efficient processors with an incredible performance per watt ratio, competing head to head with some of the best solutions on the market. .
This processor that we have tested is absolutely massive, with 256 process threadshigh sustained frequencies and somewhat lower consumption than expected. Double AVX-512 units and dedicated accelerators make it a very versatile processor, adapted to any application, although undoubtedly very oriented to applications with a high need for calculation and in environments where storage connectivity and accelerator cards are key.
End of Article. Tell us something in the Comments!
Add Comment