The latest benchmarks of Intel’s next-generation Ice Lake-SP Axon CPU server family have leaked and show some interesting results compared to AMD’s current generation 3G EPYC ROM CPU.
Intel’s Next-Gen 10nm Ice Lake-SP CPU tested, two chips with 28 million and 56 threads against a single AMP EPYC ROM 64 core flagship
As part of the Whitley platform, the Intel Ice Lake-SP CPU lineup will be made up of multiple Exxon chips. We have already seen the 6 core and 24 core parts, but the latest is the 28 core part and viewed by TUM_APISAK In the geekbench database and Momomo_US Also in SiSoftware database.
The Intel Ice Lake-SP CPU was tested on a dual socket server and had two chips. Each chip has 28 cores and 56 threads totaling 56 cores and 112 threads. Since the chip is still an early engineering model, it features a low clock speed of 1.5 GHz base and up to 3.20 GHz boost clocks. The CPU has a total of 77 MB cache with 42 MB L3 and 35 MB L2 cache. The 2S Ice Lake-SP server was equipped with 512 GB of memory to be clocked at 3200 MHz and featured in an 8-channel configuration which is one of the highlights of the new Whitley platform.
The performance of the 2S Intel Ice Lake-SP server was evaluated within Geekbench 4 which benefits from the AVX-518 instruction set depicted on Intel’s current and upcoming Exxon CPU families. In single-core tests, the server scored up to 3443 points, and in multi-core tests, the chip scored 37317 points.
Before we compare this to the AMP EPYC 7742, it should be noted that both of these test results are based on preliminary engineering samples that have low clock speeds so the final performance is expected to be much better. However, at the same time, Intel CPUs benefit from their AVX-51 instruction set in this benchmark which is a shortcoming of AMD CPUs. The entries have also been shown in various operating system environments, although the EPYC 7742 CPU was actually tested on Windows 10 Server setup, Geekbench is not reporting it correctly. Despite this, let’s take a look at how the two 28 core Ice Lake-SP Exon CPUs stacked up against an AMD EPYC4242 CPU.
We used a single EPYC 7742 entry for comparison, so we are comparing 64 cores and a total of 128 threads against 56 cores from AML and 112 threads from Intel. The AMD EPYC 7742 CPU easily outperforms Intel chips in single-core tests due to higher base clock speeds of 3.4 GHz vs. 1.5 GHz on Intel parts. At the same time, the AMD platform saves around 35000 multi-core points which is slightly lower than the Intel Ice Lake-SP parts. With the ultimate clock speed, Ice Lake-SP CPUs can easily outperform AMP EPYC ROM parts, but the lead may not be as large as Intel expected.
It clearly looks like Intel’s Ice Lake-SP was an EPYC ROM competitor that missed its opening schedule due to 10nm yields and now has to compete against AMD’s EPYC Milan which is just around the corner. We’ll wait to see some more testing and performance results for the Ice Lake-SP CPU in non-AVX512 optimized workloads, but every benchmark leak makes it very clear that Ice Lake-SP is late and AMD is getting things done. Is not meant to improve. For Intel.
And it’s not just about performance metrics, we still don’t know the Ice Lake-SP prices and power efficiency, but we do know that the current EPYC ROM lineup has a lower price and TCO than Cascade Lake. The SP lineup and the Ice Lake-SP ships are also expected to remain intact.
Intel Xeon SP Family:
|Family branding||Skylake-SP||Cascade Lake-SP / AP||Cooper Lake-SP||Ice Lake-SP||Sapphire rapids||Granite rapids|
|Process node||14nm +||14nm ++||14nm ++||10nm +||10nm ++||7nm +?|
|Platform name||Intel purley||Intel purley||Intel Cedar Island||Intel whitley||Intel eagle stream||Intel eagle stream|
|MCP (Multi-Chip Package) SKUs||No||Yes||No||Yes||TBD||TBD|
|Socket||LGA 3647||LGA 3647
|LGA 4189||LGA 4189||LGA 4677||LGA 4677|
|Maximum core count||Until 28||Until 28
Up to 48
|Until 28||To 56?||TBD||TBD|
|Max thread count||Up to 56||Up to 56
Up to 96
|Up to 56||To 112?||TBD||TBD|
|Max L3 Cash||38.5 MB L3||38.5 MB L3
66 MB L3
|38.5 MB L3||TBA (1.5 MB per core)||TBD||TBD|
|Memory support||DDR4-2666 6-Channel||DDR4-2933 6-Channel
DDR4 2933 12-Channel
|Up to 6-channel DDR4-3200||Up to 8-channel DDR4-3200||8-channel DDR5||8-channel DDR5|
|PCIe General Support||PCIe 3.0 (48 lanes)||PCIe 3.0 (48 lanes)||PCIe 3.0 (48 lanes)||PCIe 4.0 (64 lanes)||PCIe 5.0||PCIe 5.0|
|TDP Range||140W-205W||165W-205W||150W-250W||~ 250W – ~ 300W||TBD||TBD|
|3D Xpoint Optane DIMM||N / A||Apache Pass||Barlow pass||Barlow pass||Crow pass||Donhue pass|
|Contest||AMD EPYC Naples 14nm||AMD EPYC ROM 7nm||AMD EPYC ROM 7nm||AMD EPYC Match 7nm +||AMD EPYC Genoa ~ 5nm||AMD Next-Gen EPYC (Post Genoa)|
Intel Xeon 10nm + Ice Lake-SP Family
The Intel Ice Lake-SP processor will be shipping later this year and will be based on the 10nm + process node. We have stated in earlier slides that the Ice Lake family will feature up to 28 cores, but one of ASUS’s presentation says that it will actually feature 38 cores and 76 threads per socket. There are also rumors pointing to 56 cores and 112 threads, so we can’t say for sure what the actual cores will count on the new chips.
The highlight of the Ice Lake-SP processor will be support for PCIe Gen 4 and 8-channel DDR4 memory. The Ice Lake Xeon family will offer up to 64 PCIe Gen 4 lanes and support for 8-channel DDR4 memory at 3200 MHz (second gen percent memory support with 16 DIMMs per socket). The Intel Ice Lake Xeon processor will be based on the all-new Sunny Cove core architecture that delivers approximately 18% IPC improvements vs. the Skylake core architecture since 2015.
One thing to note is that for 2020 Intel’s 10nm is an augmented node of the original 10nm node which will mark its debut with the Tiger CPU CPU. It is marked as 10nm + and will exclusively use the Ice Lake-SP exon line. Some of the major upgrades that 10nm will deliver include:
- 2.7x density scaling vs 14nm
- Self-aligning quad-patterning
- Contact at active gate
- Cobalt Interconnect (M0, M1)
- 1st gen favros 3d stacking
- Second General EMIB
The Intel Ice Lake-SP lineup will compete directly with AMD’s enhanced 7nm-based EPYC matching lineup that will feature the brand new 7nm Zen 3 core architecture to be one of AMD’s largest architectural upgrades since the original Zen core Confirms it. It is expected to see more Intel and NVIDIA based servers in the coming months.