Intel has finally taken the stage and launched its long-awaited Sapphire Rapids-SP lineup. Let us give you some context about these CPUs. Sapphire Rapids has been in development for over 5 years and has faced various setbacks and delays. Reportedly, team blue suffered these delays due to their new Intel 7 process, especially on the high-core count SKUs. By November 2022, Intel finally gave us the release date of Sapphire Rapids, which is today.
LGA 4677 Socket
Sapphire-Rapids is the successor to Ice Lake and Cooper Lake series. Featuring up to 60 cores, it uses the brand-new LGA 4677 socket codenamed ‘C741‘. These CPUs are packed with all the goods of modern-day Intel CPUs, making them a worthy successor to the previous generation of Intel Xeon Scalable Processors.
Since we’re talking about the new socket, let’s go over its specifications. The C741 platform comprises entirely of PCIe Gen 5.0 CPU lanes. In addition, the SPR-SP CPUs support up to 8-memory channels with 2 DIMMs per channel. The chipset on the other hand provides only PCIe Gen 3.0 lanes (up to 20) with a maximum of 20 SATA 3.0 (6Gb/s) ports.
- New Processor Core Architecture for State-of-the-Art Performance
- Up to 56 Cores & 105 MB Intel@ Smart Cache (L3) per Socket
- Intel@ Turbo Boost Technology 2.0
- Intel@ Hyper-Threading Technology
- 8-Channel RDIMM/3DS-RDlMM with up to DDR5-4800 per Socket
- Workstation Reliability, Serviceability, and Availability (RAS) Features including Error Correction Code (ECC) Memory Support
- Intel@ Virtual RAID on CPU (Intel@ VROC) 8.0 and Intel Volume Management Device (Intel@
Sapphire Rapids-SP or Intel’s next-gen server offerings come in two categories. One is the base Sapphire-Rapids-SP and the other is Sapphire Rapids-SP with HBM (Xeon Max CPUs). We discussed Xeon Max in an article before. In addition, Sapphire Rapids also extends to the workstation segment, but that is planned for release sometime in the near future. Fret not, for we have nearly everything covered related to the SPR-WS CPUs. Read more about there here.
Sapphire Rapids Platform
In terms of I/O, Sapphire Rapids supports PCIe Gen 5.0, DDR5 memory, and on-package HBM memory (Xeon Max) along with CXL 1.1 which is a huge feat for Intel. These CPUs will support 8-channel DDR5 memory having speeds of up to 4800MT/s. Sapphire Rapids-SP is based on an MCM design with each tile housing 15 Performance Golden Cove cores.
- Intel Golden Cove Cores
- AVX512-FP16 Instruction Set
- TSXLDTRK Instruction Set
- Intel AMX Instruction Set
- Intel Data Streaming Accelerator
The memory controllers for each tile offer 2-channel DDR5 memory, for a total of 8-channel memory across 4 tiles. Similarly, the PCIe Gen 5.0 per each tile stands at 28 lanes or 112 total lanes. We will first go over the Sapphire Rapids-SP CPUs (excluding Xeon Max), as they are the latest addition to Intel’s arsenal against team red. Intel has also added various security features and accelerators with these CPUs which are as follows:
- Intel Advanced Matrix Extensions (AMX)
- Intel QuickAssist Technology (QAT)
- Intel Data Streaming Accelerator (DSA)
- Intel Dynamic Load Balancer (DLB)
- Intel In-Memory Analytics Accelerators (DLB)
- Intel Advanced Vector Extensions 512 (AVX-512)
- Intel Advanced Vector Extensions 512 (AVX-512) for vRAN
- Intel Crypto Acceleration
- Intel Speed Select Technology (SST)
- Intel Data Direct I/O Technology (DDIO)
- Intel Software Guard Extensions (SGX)
- Intel TrustDomain Extension (TDX)
- Intel Control-Flow Enforcement Technology (CET)
Xeon Scalable Processors are divided into various categories depending on their power consumption and the market segment they target. For Sapphire Rapids, we have 6 different categories making Intel more accessible to a wide variety of consumers.
- Xeon Max (9400 Series)
- Xeon Platinum (8000 Series)
- Xeon Gold (6000 Series)
- Xeon Gold (5000 Series)
- Xeon Silver (4000 Series)
- Xeon Bronze (3000 Series)
Sapphire Rapids Lineup
So, this chart you see below lists all the Sapphire Rapids CPUs that Intel has unveiled today. In fact, Intel went forward and included even the Xeon Max CPUs in this chart. Akin to AMD, Intel has also divided its offerings across all price points and market segments.
Xeon Platinum 8490H
We present to you the fastest Intel CPU available in the market, the Xeon Platinum 8490H. This monstrosity packs 60 cores / 120 threads and a whopping 112.5MB of L3 Cache. The base clock speeds stand at 1.9GHz, but can and will boost to 2.9GHz. It is noteworthy that AMD’s Genoa-based Bergamo with Zen4C cores will offer up to 128 cores. However, Sapphire Rapids is not aimed at Genoa. Rather it is Intel’s late response to AMD’s last-gen Milan.
Alright, as for the pricing, Intel is asking $17,000 which is quite a hefty cost. In fact, against Genoa, it is almost $5000 more expensive, so that is something Intel will have to worry about. The 8490H can offer an 8S configuration, which amounts up to 480 cores, 960 threads, and 900MB of cache. EPYC Genoa on the other hand only offers a 2S configuration. The only caveat here is the price. The 8490H in an 8S configuration will cost you $136,000, or 0.136 Million USD.
Intel Max CPUs
Intel’s new Max CPUs are the world’s first x86 CPUs to ship with high-bandwidth memory. The Xeon Max CPU (9480) features 56 Performance cores. These cores span across 4 tiles which are connected through Intel’s embedded multi-die interconnect bridge in a 350W package. The Xeon Max CPU contains 64GB of high bandwidth memory along with support for PCI Express Gen 5 and CXL1.1 Input/Output. These CPUs can be used in two different modes, namely the HBM Flat mode and the HBM caching mode.
As for the package size, it stands at 5800 mm2, which is roughly 28% higher than the standard SPR-SP lineup. Intel Max CPUs support 80 PCIe Gen 5.0 and Gen 4.0 lanes along with 8-channel DDR5-4800 memory.
|CPU||Cores||Base Clocks||Boost Clocks||Price|
|Xeon Platinum 9480||56||1.9GHz||2.6GHz||$12,980|
|Xeon Platinum 9470||52||2.0GHz||2.7GHz||$11,590|
|Xeon Platinum 9468||48||2.1GHz||2.6GHz||$9,900|
|Xeon Platinum 9460||40||2.2GHz||2.7GHz||$8,750|
|Xeon Platinum 9462||32||2.7GHZ||3.1GHz||$7,995|
The large memory size is going to be extremely competitive against AMD’s Genoa in specific workloads. Though, not as fast as cache, its massive size will give it a slight edge in optimized tasks. All CPUs in this lineup have 64GB of high-bandwidth memory, so that’s something new. As HBM uses more power than DDR5 memory, Intel Max CPUs top out at just 56 cores, due to TDP limits.
Intel Max GPUs
On the GPU side of things, we have Intel’s new Data Center Max GPUs codenamed ‘Ponte Vecchio‘. Based on the Xe-HPC architecture, they offer up to 128 Xe Cores specifically designed for handling high-speed computing. In addition, we see 408MB of L2 cache which is termed the highest in the industry along with 64MB of L1 cache. What’s better is that these GPUs support ray tracing which can help accelerate scientific visualizations.
The Max GPU series features 3 SKUs equipped with blazing-fast HBM and support PCIe Gen 5.0, which is an industry-first.
- Max Series 1100 GPU: A 300-watt double-wide PCIe card with 56 Xe cores and 48GB of HBM2e memory. This GPU offers a base clock of 1GHz, which can boost up to 1.55GHz.
- Max Series 1350 GPU: A 450-watt OAM module with 112 Xe cores and 96GB of HBM.
- Max Series 1550 GPU: Intel’s maximum performance 600-watt OAM module with 128 Xe cores and 128GB of HBM.
When compared to last-gen, Sapphire Rapids offers the following increases:
- 53% higher general-purpose compute
- Up to 10x faster AI performance
- Up to 2x capacity for vRAN at the same power level
- Up to 2x higher data compression
- Up to 3x higher performance in Data Analytics
- Up to 3.7x (Xeon Max CPU) faster in memory-bound workloads
Against the 3rd generation of Xeon Scalable CPUs, Sapphire Rapids is 2.9x more power efficient. On using the Optimized Power Mode, Intel is promising power savings of up to 70W. Similarly, with Sapphire Rapids, expect to see 55% lower TCO with a 524,000 kg reduction in CO2 emissions.
Performance (Max Series)
Against the NVIDIA A100, the Max Series GPUs are 2x faster in both OpenMC and miniBUDE. That is a huge feat to achieve, considering Intel is dabbling in unfamiliar territory.
The story does not end there as Intel terms the GPU Max series 1.5x faster than the NVIDIA A100 in Virtual Reactor Simulation.
The blood-bath for NVIDIA continues because Intel claims that its new GPUs are almost 2.4x faster in financial workloads.
With Sapphire Rapids officially out to the market, AMD will aim to launch Bergamo very soon. These CPUs are promising on paper, but they are sadly nowhere near what Genoa packs. It’s like comparing Ryzen 7000 against Alder lake. If we could reverse time, we probably would and tell Intel to release these CPUs earlier. But since breaking one of the most fundamental laws of physics is not our forte, Intel will have to suffer in the server department until they get ‘Intel 4‘ or even ‘Intel 3‘ ready.