Who doesn’t like a company that keeps its word?
In February, when Intel refreshed the 2nd gen Intel Xeon Scalable processors, the company promised new CPUs that would offer significantly better performance and better price/performance.
Since then, Intel has tested these new CPUs. We’ve just seen the test results, and some are even better than promised — especially when the new CPUs are combined with the latest Intel solid state drive (SSD) storage and Ethernet network adapters.
A bit of background
In February, Intel estimated that the 2nd gen Intel Xeon Gold processors would deliver an average of 1.36x higher performance than a comparable 1st gen Intel Xeon Scalable Gold CPU. And 1.42x better performance per dollar.
To deliver these kinds of gains, Intel boosted the new server processors by adding cores, increasing cache sizes and raising frequencies.
Intel also said that the new 2nd gen Intel Xeon Scalable processors — labeled with an R, T or U suffix — were designed for both dual- and single-socket mainstream and entry-level server systems. Typical workloads include virtualized clouds, hyperconverged infrastructures (HCI) and network function virtualization (NFV).
More recently, Intel tested the processors. To do this, Intel set up a workload-generating client node connected to three application servers and several storage nodes configured with both SATA and NVMe SSDs. Then the company tested them with various configurations of compute, storage and network components.
The test workload simulated an online transactional purchasing system database. It ran Microsoft Windows Server 2019 Datacenter, Microsoft Hyper-V and Microsoft SQL Server 2019 Enterprise.
Items measured included orders per minute, number of virtual machines, processor utilization, host server power and network throughput. And for the purposes of these tests, Intel used only the new R processors — specifically, the 4214R and 6238R.
Gen vs. Gen
So much for the background. Let’s examine Intel’s test results. The following table, courtesy of Intel, lays it out:
Along the top in blue, you can see the various processors. The first (on the left) is the previous generation E5 CPU, and it’s presented as a baseline for comparison. The second row shows the network adapter used.
Below that, the performance measures are shown in the blue column on the left: virtual machine (VM) count, total operations, operations/VM, etc. To see how an attribute performs on different CPUs, read the appropriate row from left to right.
For example, the second column shows the performance for an Intel Xeon Silver 4214R CPU. Note that total operations and OPM per VM (orders per minute per virtual machine) both rise over the E5, even though both CPUs are on the same network and have the same VM count. Also note that on the new R processor, the CPU utilization rate actually drops.
For the biggest performance gain, jump over to the column farthest to the right, the Intel Xeon Gold 6238R processor. When fitted out with a fast 25GbE network and the latest Intel SSD storage, it delivers some really big gains. These include:
> Over 4x more VMs (26 vs. 6)
> A more than 5x improvement in total operations (34.6 million vs. 6.5 million)
> A modest improvement in OPM per VM (21.8K vs. 17.8K)
> And a greater than 17x improvement in total MB/sec. (1,866 vs. 108)
Performance per dollar
Okay, so how do the new R processors do on price/performance? Again, pretty darned good. Check out this chart, also courtesy of Intel:
Let’s break it down. The scale on the left is thousands of dollars. The scale on the right is the number of servers needed. Each column represents a different configuration.
The column on the far left shows the baseline of the previous generation processor, which is again the Intel Xeon Processor E5-2630. And the workload being tested is support for 200 million database transactions.
What do we get? Well, a workload that required 31 servers with the previous generation can now be done with as few as 6. And the total solution cost drops by 60%.
Here’s another way to look at the same math. Assume a customer comes to you saying, “I have a server budget of $250,000. What will that get me?” Here, again courtesy of Intel, is the answer:
The five blue bars represent different configurations, with the one on the far left showing the Intel Xeon Processor E5-2630 as a baseline. The left scale is millions of database operations per server. The right scale is the number of servers your customer’s $250K would get them.
So the chart shows that for the same $250K, the Intel Xeon Gold 6248R processor delivers 1.9x more transactions. At the same time, the number of servers that support this increase in performance goes down from 15 servers to 8.
The slightly lower performance on the final column to the right is derived from the current increased cost of 25GbE switches. While this server configuration yields higher database performance, the cost of the 25GbE switches drops this number slightly — though still better than baseline.
Here’s yet one more way to look at it. To support 125 virtual machines, how many servers would you need, and at what cost? Again, from Intel:
Each blue column represents a configuration. The scale on the left is the cost in thousands of dollars. The scale on the right is the number of servers.
As you can see, the Intel Xeon Gold 6238R can do with 5 servers what it would take the previous-gen Intel Xeon Processor E5-2630 a total of 21 servers. That’s a reduction in the number of servers of 76%.
Also, the total solution cost drops from about $300K to $150K, a savings of 50%. Replace the Ethernet adapter with a 10GbE setup, and the cost comes down to an even lower $125K.
The new math
What’s it all add up to? Some new math, and pretty simple at that.
The 2nd gen Intel Xeon Scalable processors, when combined with the latest Intel SSD and Ethernet technologies, promise to deliver some truly great value for you and your customers.
That’s a promise that Intel — and now you — can keep.
> Learn more: Explore the Intel Xeon Scalable processor family