Intel Xeon Cascade Lake-SP Refresh
Computer Reseller News (CRN) is reporting that Intel is going to release eighteen new Xeon second-generation Xeon Scalable server processors. These Intel Xeon Cascade Lake-SP Refresh processors will have an R suffix on the model name to show that it is a refresh SKU. This is still a rumor at this point, but there have been enough hints and leaks that it seems pretty credible.
Cascade Lake-SP Refresh Analysis
These new SKUs typically have more physical cores and/or higher base clock speeds than the corresponding original Cascade Lake-SP SKU. None of these refresh SKUs have more than 28 physical cores, but you will be able to get higher core counts in the Intel Gold, Silver, and Bronze families compared to the original release. None of these refreshed SKUs will be in the Intel Xeon Platinum family.
Best Choices for SQL Server
Having more physical cores in a processor gives you more total CPU capacity. This assumes that the performance of each core is the same as you increase the core count. Often, that is not the case, since higher core count processors will usually have lower base clock speeds.
If you are using software that uses core-based licensing, then having more physical cores also means increased licensing costs. These increased software license costs can add up very quickly, and often will dwarf your hardware costs.
SQL Server is a prime example of this. SQL Server 2019 Standard Edition is $1,793.00 per physical core (with a limit of 24 physical cores/instance). If you need or want Enterprise Edition, SQL Server 2019 Enterprise Edition is $6,874.00 per physical core. There is no difference in the cost based on the actual performance of each physical core. Because of this, you should purposely pick the highest performance SKU at a given core count.
The two lists below highlight the “best” SKUs for SQL Server usage from these two groups of refreshed and original Cascade Lake-SP processors at different core counts.
Preferred Intel 2nd Generation Xeon Scalable Processor Refresh SKUs
- Xeon Gold 6258R: 28 cores/56 threads, 2.7 GHz Base clock
- Xeon Gold 6248R: 24 cores/48 threads, 3.0 GHz Base clock
- Xeon Gold 6242R: 20 cores/40 threads, 3.1 GHz, Base clock
- Xeon Gold 6246R: 16 cores/32 threads, 3.4 GHz Base clock
- Xeon Silver 4215R: 8 cores/16 threads, 3.2 GHz, Base clock
Preferred Intel 2nd Generation Xeon Scalable Processor SKUs
- Xeon Platinum 8280: 28 cores/56 threads, 2.7 GHz Base clock
- Xeon Platinum 8268: 24 cores/48 threads, 2.9 GHz Base clock
- Xeon Gold 6254: 18 cores/36 threads, 3.1 GHz Base clock
- Xeon Gold 6246: 16 cores/32 threads, 3.3 GHz Base clock
- Xeon Gold 6244: 8 cores/16 threads, 3.6 GHz Base clock
- Xeon Gold 5222: 4 cores/8 threads, 3.8 GHz Base clock
These new “refresh” SKUs fill in some core count gaps, and have slightly higher base clock speeds at some core counts. Being able to use a Gold family processor instead of a Platinum family processor in some scenarios (at the same core count) is going to save a decent amount of money on the hardware costs. For example, the 28C/56T Xeon Gold 6258R is likely to be much less expensive than the 28C/56T Xeon Platinum 8280 processor.
- Intel Set To Unleash ‘Cascade Lake Refresh’ In AMD EPYC Attack
- Rumor: Intel Prepping Xeon Cascade Lake Refresh to Fight AMD Epyc
- Intel’s Rumored Cascade Lake Xeon Refresh To Counterattack AMD’s Dominant EPYC Rome Enterprise CPUs
- Intel Readies Cascade Lake Refresh Xeon Family To Tackle AMD’s 2nd Gen EPYC Rome CPUs – Feature More Cores, Higher Clocks & Higher TDPs
If this rumor turns out to be true, I will put together an updated list of my preferred Intel processor models for SQL Server at each core count. Of the eighteen rumored “refresh” SKUs, only four or five actually look interesting for SQL Server. Even so, they will fill in some gaps, and offer combinations of better performance and/or lower hardware costs at some core counts. This is a welcome development, which proves that viable competition from both AMD and ARM is good for the consumer. This is also another piece of evidence that Intel is under some pressure in the server CPU segment, as I wrote about here.
Please let me know what you think in the comments. Thanks for reading!