Intel just added the Cascade Lake-SP flagship to its latest server offerings without any marketing or fan-fare. The Xeon Platinum 8284 is the fastest SP by Intel in this bracket, and comes with 28 cores running at a base clock of 3GHz and a boost of 4GHz. The chip costs a staggering $15460. That’s 5 grand more than the 64 Core Epyc Rome chip, recently launched by AMD, based on the 7nm Zen 2 cores.
The Xeon Platinum 8284 is based on 14nm++ node and leverages the Cascade Lake CPU architecture which if we’re being honest isn’t all that different from Coffee Lake or Kaby Lake. We’re basically looking at the same design with increased core counts, higher clocks, more cache, but at the same time a higher TDP as well.
The new Cascade Lake-SP flagship comes with slightly higher clocks at the expense of a higher TDP (240W) and hotter thermals. The max memory is capped at 1TB with a speed of 2933MHz, and 6 channel memory support. The L3 cache tops out at 38.5MB and the PCIe lanes are limited to 48. An improvement over the last generation but still not quite up there with Epyc.
However, Intel has been taking cues from AMD especially when it comes to the packaging technologies. The upcoming Xeon Cascade Lake-AP (Advanced Performance) processors meant for Data Centers leverage the Multi-Chip Package (MCP) architecture. Multiple dies are used on the same substrate to increase the core count, the same way AMD’s Ryzen and Epyc chips do it.
We are looking at 24 cores per die, 48 in total for a processor, paired with 12 DDR4 memory channels. This results in a total of 96 Cores in a 2S configuration, along with 24 lane memory support. Not bad. Now, only if Intel charge a fortune for its CPUs…
- AMD Ryzen 7 3800X Overclocked to 5.9 GHz on LN2
- NVIDIA GeForce RTX 2080 Super 7-8% Faster than RTX 2080 in AoTS