The 5th Generation Intel® Xeon® Scalable processors, formerly codenamed Emerald Rapids, have been recently launched by Intel. These CPUs are designed with one key element in mind, more performance whilst maintaining or improving power efficiency. This means more performance per watt. The improvements mean Datacentre workloads will see increased performance in areas such as AI, HPC, Data Analytics, Networking/5G and Storage.
Intel have increased the core count with this new CPU stack, the top end SKU now reaching 64 cores. Memory bandwidth too is increased with up to 5600 MT/s modules now supported. Also, we see a boost in last level cache by up to 3x, with the Platinum series CPUs now equipped with 320MB L3 cache.
In this article we will be taking a closer look at these 5th Gen processors and benchmarking 3 of the SKU’s, across a range of core counts and comparing them to their last gen equivalents.
Looking at the specifications of the 5th Gen Intel® Xeon® Scalable processors, it’s fair to say this is evolution rather than revolution. The new CPUs are using the same LGA-4677 socket as Intel’s 4th generation (Sapphire Rapids) and the same ‘Intel 7’ which is their 10nm process. The benefit of using the same socket is drop-in support (with a BIOS update), so end users can benefit from the performance increase of these CPUs with existing servers/motherboards such as Boston X13 based systems. Or for customers considering 4th Gen Intel® Xeon® Scalable it makes sense now to look at the 5th gen processors instead.
Of course, if upgrading platform from an older server (e.g., a Boston X12 server using Intel 3rd gen Xeon Scalable) then the performance boost will be much larger.
Under the hood there is a difference in the CPU die, Intel have moved from a four-tile design to two.
The XCC (Extreme Core Count) SKU’s will use this architecture, with the MCC (Medium Core Count) sticking with the monolithic die approach as with Sapphire Rapids.
Additionally, as well as the performance improvements that Intel promise, we still see the features introduced in the 4th Generation Intel® Xeon® Scalable processors (Sapphire Rapids), some of which have been improved on:
PCIe 5.0 and CXL (Compute Express Link) 1.1 is supported. New here is support for Type 3 devices in addition to the previously supported Type 1 and 2. For a summary of what these can offer, see below:
Type 1 Devices: Accelerators such as smart NICs typically lack local memory. Via CXL, these devices can communicate with the host processor’s DDR memory.
Type 2 Devices: GPUs, ASICs, and FPGAs are all equipped with DDR or HBM memory and can use CXL to make the host processor’s memory locally available to the accelerator—and the accelerator’s memory locally available to the CPU. They are also co-located in the same cache coherent domain and help boost heterogeneous workloads.
Type 3 Devices: Memory devices can be attached via CXL to provide additional bandwidth and capacity to host processors. The type of memory is independent of the host’s main memory.
Intel® UPI 2.0 increases inter socket bandwidth up to 20GT/s, a 1.25x increase on the previous generation. So, communication between 2 CPUs in a dual socket server will be faster.
Built in accelerators to improve performance even further on specific workloads such as Advanced Matrix Extensions, Dynamic Load Balancer, Data Streaming Accelerator, In-Memory Analytics Accelerator and QuickAssist Technology.
There are fewer SKUs in the 5th generation so perhaps Intel is streamlining. This is a good thing in my view as there were a huge range of SKUs for 4th Gen, which potentially led to some confusion with the sheer number of options. A more refined approach is welcome here.
Most 5th Gen SKUs have a boost in base or turbo clock speeds when compared to their predecessor. However, it would have been nice to see the full 5600MHz memory support across all CPU’s. Instead, the lower end CPU SKUs will once again only support lower memory speeds.
The full lineup can be seen below.
For our benchmarking we looked at three of the 5th Gen Intel® Xeon® Scalable CPUs and compared them directly to their 4th Gen counterparts.
We have 16 and 32 core CPUs for a direct comparison. I also wanted to compare the top end SKU’s so completing the lineup we have the 56-core 4th gen part vs the 64-core 5th gen CPU. All CPUs were installed in a dual socket server.
As seen in the table below, the 16 and 56/64 core clock speeds are almost identical. Only the 32 core has a significant difference which is the single core boost of 4GHz vs 3.4Ghz. The TDP is the same in each comparison so any performance increases will be at the same power draw, confirming the increased performance per watt as Intel have stated.
The test server was as follows:
6444Y | 6544Y | 6430 | 6530 | 8480+ | 8592+ | |
---|---|---|---|---|---|---|
Platform | Sapphire Rapids | Emerald Rapids | Sapphire Rapids | Emerald Rapids | Sapphire Rapids | Emerald Rapids |
Cores/Threads | 16/32 | 16/32 | 32/64 | 32/64 | 56/112 | 64/128 |
Base/ All core/ single core boost (GHz) | 3.6/4/4.1 | 3.6/4.1/4.1 | 1.9/2.6/3.4 | 2.1/2.7/4 | 2/3/3.8 | 1.9/2.9/3.9 |
CPU TDP (Watts) | 270 | 270 | 270 | 270 | 350 | 350 |
In the rendering tests of Cinebench, we can see a small boost in single core performance except in the case of the 32 core 6530 which has a massive 50% increase. The single core turbo going from 3.4Ghz to 4Ghz vs the 6430 makes this possible.
In the multicore tests we see various levels of improvement. The 6544Y has a very similar spec to its predecessor so the increase is marginal but the 32 core 6530 shows bigger gains at around 5%. Of course, the top end SKU now has more 8 more cores so whilst it’s not a like for like comparison, the new 64 core 8592+ scored around 25% higher than last gens best CPU, the 56 core 8480+.
Similarly, we see reduced rendering times in Blender scenes. This is more pronounced in the 32 core 6530 and top end 8592+ SKU, with the 16 core 6544Y only showing marginal gains.
V-Ray runs a CPU test as well as a GPU rendering test where you can select the CPU to render the scene. The 6530 in its like for like test vs 6430 again looks like a good upgrade. At the top end it’s a big jump in performance with up to 23% better scores.
Passmark runs a series of CPU benchmarks to test overall performance. The 6444Y is showing small gains around 3% but the other CPUs tested have a much-improved performance. The 6530 again is showing impressive results when compared with the equivalent 4th gen 6430 SKU, increasing the score by 26%.
This benchmark performs a data compression and decompression and shows a rating in MIPS (million instructions per second). As the graphs show these CPU’s offer more performance in these workloads.
In the CPU-Z single threaded test the 6530’s higher turbo clock speed meant a whopping 53% increase. For the multi-threaded test, we can see better performance across the board.
With the 5th Gen Intel® Xeon® Scalable we see a refinement rather than a redesign, but as a CPU refresh there is plenty of things to be positive about. The performance is increased across the board and at the same power consumption, meaning Intel’s claim of more performance per watt rings true. The CPU stack is refined and now tops out at 64 cores so offers even more performance for high-end deployments. There is also faster memory support, in a future article we will take a look at 5600MT/s memory modules and what performance improvements can be seen.
These CPUs were tested in our Boston Labs facility. Boston Labs is all about enabling our customers to make informed decisions in selecting the right hardware, software and overall solution for their specific challenges. If you’d like to request a test drive of 5th Generation Intel® Xeon® Scalable Processors, please get in contact by emailing [email protected] or call us on 01727 876100 and one of our experienced sales engineers will gladly guide you through building the perfect solution just for you.
Sukhdip Mander, Field Application Engineer, Boston Limited
To help our clients make informed decisions about new technologies, we have opened up our research & development facilities and actively encourage customers to try the latest platforms using their own tools and if necessary together with their existing hardware. Remote access is also available
Boston are exhibiting at SC23 in Georgia, Atlanta this November!