Welcome Back Intel Xeon 6900P Reasserts Intel Server Leadership

Intel Xeon 6980P Performance

Full disclosure on this one. We had a pre-production server for days, not weeks. As a result, many of our benchmarks that take a day or more to run are just not possible.

Python Linux 4.4.2 Kernel Compile Benchmark

This is one of the most requested benchmarks for STH over the past few years. The task was simple: we had a standard configuration file, the Linux 4.4.2 kernel from kernel.org, and made the standard auto-generated configuration utilizing every thread in the system. We are expressing results in terms of compiles per hour to make the results easier to read.

This one is really neat. There are brief single-thread segments during this workload. What we found a few years ago is that as the core counts scale, we get more compiles per hour by splitting the workload up into multiple instances running in parallel. The new Xeon 6980P just has lots of cores, cache, and memory bandwidth, so it does well here. We might be able to get a bit more performance by splitting the workload to 42/43 core instances to align to the structure of the chip, but we ran out of time for this review.

c-ray 1.1 Performance

We have been using c-ray for our performance testing for years now. It is a ray tracing benchmark that is extremely popular for showing differences in processors under multi-threaded workloads. Here are the 8K results:

This is so much fun. We built this back in the day when we thought an 8K render size would be more than we would ever need. Systems would take tens of minutes to do a single run. Now, we need something much bigger since we are seeing differences, but we are losing fidelity.

When we talk about how the new chips are so fast that they are really designed for running multiple workloads at once, this is a great example.

STH nginx CDN Performance

On the nginx CDN test, we are using an old snapshot and access patterns from the STH website, with DRAM caching disabled, to show what the performance looks like fetching data from disks. This requires low latency nginx operation but an additional step of low-latency I/O access, which makes it interesting at a server level. Here is a quick look at the distribution:

Here, we are seeing a familiar pattern. There are a few items to note, though. The Intel CPUs are doing really well. We are not, however, doing Intel QAT offload for OpenSSL and nginx. If you do not want to use acceleration, then this is a good result. If you are willing to use accelerators, then this is not ideal since there is silicon accelerators not being used. It is like not using a GPU or NPU for AI acceleration.

See Intel Xeon D-2700 Onboard QuickAssist QAT Acceleration Deep-Dive if you want to see how much performance QAT can add (albeit that would only impact a portion of this test.)

Even without QuickAssist and looking at cores only, Intel is doing really well.

MariaDB Pricing Analytics

This is a very interesting one for me personally. The origin of this test is that we have a workload that runs deal management pricing analytics on a set of data that has been anonymized from a major data center OEM. The application effectively looks for pricing trends across product lines, regions, and channels to determine good deal/ bad deal guidance based on market trends to inform real-time BOM configurations. If this seems very specific, the big difference between this and something deployed at a major vendor is the data we are using. This is the kind of application that has moved to AI inference methodologies, but it is a great real-world example of something a business may run in the cloud.

What is interesting about these non-microbenchmark and ultra-scalable workloads is that we are starting to run into non-CPU bottlenecks. We might have to split this into multiple instances to keep it scaling in the future.

STH STFB KVM Virtualization Testing

One of the other workloads we wanted to share is from one of our DemoEval customers. We have permission to publish the results, but the application being tested is closed source. This is a KVM virtualization-based workload where our client is testing how many VMs it can have online at a given time while completing work under the target SLA. Each VM is a self-contained worker. This is very akin to a VMware VMark in terms of what it is doing, just using KVM to be more general.

Although Intel’s E-cores do well here, the P-cores with Hyper-Threading help a lot, which is why this new chip does so well. Intel is just getting enormous generational gains here.

Since we had limited time, and only one CPU configuration, let us talk a bit about what Intel showed for its charts.

Intel’s Performance View of Intel Xeon 6900P

By doubling the core count, Intel can double its performance, plus get a bit more from architectural improvements. Perhaps the more notable part here is that, although the TDP is increasing from 385W to 500W, by doubling the core count and performance, it is getting much higher efficiency.

Intel also showed it is extremely competitive with both of AMD’s top-end Genoa/ and Bergamo parts. Of course, AMD said its Turin processors are launching next quarter so this will change.

Intel had a really cool benchmark here. At first, you might look at this and think this is 64 core EPYC Genoa to 96 core Xeon Granite Rapids-AP so things are way off. What Intel is actually doing is using a 16 vCPU VM and measuring performance while all of the rest of the cores are being utilized around 50%. So this is 16 vCPU to 16 vCPU just using different parts. If Intel had used a 96-core EPYC Genoa vs. a 96-core Xeon, it might have optically looked better, but then it would be using a top bin versus a not-top bin part, which would skew TDP/ core. Perhaps the more exciting thing is that this type of benchmarking is happening at all.

Of course, since everything is AI, and Intel has not just VNNI but also AMX for AI acceleration, it does very well at AI inference.

Many HPC workloads are memory instead of compute bound. This is where having fast DDR5-6400 and also 8800MT/s MRDIMMs helps a lot.

Simply adding double the number of cores along with 12 channels of faster DDR5 memory really helps Intel here. It has a leadership part on a per-socket basis. If we had a lot more time, we probably could find a few workloads that simply run better on Zen 4c cores due to microarchitectural fit, but the story is simple: Intel is back in the performance leadership position for now.

Next, let us talk a bit about power.