Boosting Performance with the Latest Generations of Virtual Machines in Azure
Microsoft Azure recently announced the availability of the new generation of VMs (v6)—including the Dl/Dv6 (general purpose) and El/Ev6 (memory-optimized) series. These VMs are powered by the latest Intel Xeon processors and are engineered to deliver: Up to 30% higher per-core performance compared to previous generations. Greater scalability, with options of up to 128 vCPUs (Dv6) and 192 vCPUs (Ev6). Significant enhancements in CPU cache (up to 5× larger), memory bandwidth, and NVMe-enabled storage. Improved security with features like Intel® Total Memory Encryption (TME) and enhanced networking via the new Microsoft Azure Network Adaptor (MANA). By Microsoft Evaluated Virtual Machines and Geekbench Results The table below summarizes the configuration and Geekbench results for the two VMs we tested. VM1 represents a previous-generation machine with more vCPUs and memory, while VM2 is from the new Dld e6 series, showing superior performance despite having fewer vCPUs. VM1 features VM1 - D16S V5 (16 Vcpus - 64GB RAM) VM1 - D16S V5 (16 Vcpus - 64GB RAM) VM2 features VM2 - D16ls v6 (16 Vcpus - 32GB RAM) VM2 - D16ls v6 (16 Vcpus - 32GB RAM) Key Observations: Single-Core Performance: VM2 scores 2013 compared to VM1’s 1570, a 28.2% improvement. This demonstrates that even with half the vCPUs, the new Dld e6 series provides significantly better performance per core. Multi-Core Performance: Despite having fewer cores, VM2 achieves a multi-core score of 12,566 versus 9,454 for VM1, showing a 32.9% increase in performance. VM 1 VM 2 Enhanced Throughput in Specific Workloads: File Compression: 1909 MB/s (VM2) vs. 1654 MB/s (VM1) – a 15.4% improvement. Object Detection: 2851 images/s (VM2) vs. 1592 images/s (VM1) – a remarkable 79.2% improvement. Ray Tracing: 1798 Kpixels/s (VM2) vs. 1512 Kpixels/s (VM1) – an 18.9% boost. These results reflect the significant advancements enabled by the new generation of Intel processors. Score VM 1 VM 1 VM 1 Score VM 2 VM 2 VM 2 Evolution of Hardware in Azure: From Ice Lake-SP to Emerald Rapids Technical Specifications of the Processors Evaluated Understanding the dramatic performance improvements begins with a look at the processor specifications: Intel Xeon Platinum 8370C (Ice Lake-SP) Architecture: Ice Lake-SP Base Frequency: 2.79 GHz Max Frequency: 3.5 GHz L3 Cache: 48 MB Supported Instructions: AVX-512, VNNI, DL Boost VM 1 Intel Xeon Platinum 8573C (Emerald Rapids) Architecture: Emerald Rapids Base Frequency: 2.3 GHz Max Frequency: 4.2 GHz L3 Cache: 260 MB Supported Instructions: AVX-512, AMX, VNNI, DL Boost VM 2 Impact on Performance Cache Size Increase: The jump from 48 MB to 260 MB of L3 cache is a key factor. A larger cache reduces dependency on RAM accesses, thereby lowering latency and significantly boosting performance in memory-intensive workloads such as AI, big data, and scientific simulations. Enhanced Frequency Dynamics: While the base frequency of the Emerald Rapids processor is slightly lower, its higher maximum frequency (4.2 GHz vs. 3.5 GHz) means that under load, performance-critical tasks can benefit from this burst capability. Advanced Instruction Support: The introduction of AMX (Advanced Matrix Extensions) in Emerald Rapids, along with the robust AVX-512 support, optimizes the execution of complex mathematical and AI workloads. Efficiency Gains: These processors also offer improved energy efficiency, reducing the energy consumed per compute unit. This efficiency translates into lower operational costs and a more sustainable cloud environment. Beyond Our Tests: Overview of the New v6 Series While our tests focused on the Dld e6 series, Azure’s new v6 generation includes several families designed for different workloads: 1. Dlsv6 and Dldsv6-series Segment: General purpose with NVMe local storage (where applicable) vCPUs Range: 2 – 128 Memory: 4 – 256 GiB Local Disk: Up to 7,040 GiB (Dldsv6) Highlights: 5× increased CPU cache (up to 300 MB) and higher network bandwidth (up to 54 Gbps) 2. Dsv6 and Ddsv6-series Segment: General purpose vCPUs Range: 2 – 128 Memory: Up to 512 GiB Local Disk: Up to 7,040 GiB in Ddsv6 Highlights: Up to 30% improved performance over the previous Dv5 generation and Azure Boost for enhanced IOPS and network performance 3. Esv6 and Edsv6-series Segment: Memory-optimized vCPUs Range: 2 – 192* (with larger sizes available in Q2) Memory: Up to 1.8 TiB (1832 GiB) Local Disk: Up to 10,560 GiB in Edsv6 Highlights: Ideal for in-memory analytics, relational databases, and enterprise applications requiring vast amounts of RAM Note: Sizes with higher vCPUs and memory (e.g., E128/E192) will be generally available in Q2 of this year. Key Innovations in the v6 Generation Increased CPU Cache: Up to 5× more cache (from 60 MB to 300 MB) dramatically improves data access speeds. NVMe for Storage: Enhanced local and remote storage performance, with up to 3× more IOPS locally and the capability to reach 400k IOPS remotely via Azure Boost. Azure Boost: Delivers higher throughput (up to 12 GB/s remote disk throughput) and improved network bandwidth (up to 200 Gbps for larger sizes). Microsoft Azure Network Adaptor (MANA): Provides improved network stability and performance for both Windows and Linux environments. Intel® Total Memory Encryption (TME): Enhances data security by encrypting the system memory. Scalability: Options ranging from 128 vCPUs/512 GiB RAM in the Dv6 family to 192 vCPUs/1.8 TiB RAM in the Ev6 family. Performance Gains: Benchmarks and internal tests (such as SPEC CPU Integer) indicate improvements of 15%–30% across various workloads including web applications, databases, analytics, and generative AI tasks. My personal perspective and point of view The new Azure v6 VMs mark a significant advancement in cloud computing performance, scalability, and security. Our Geekbench tests clearly show that the Dld e6 series—powered by the latest Intel Xeon Platinum 8573C (Emerald Rapids)—delivers up to 30% better performance than previous-generation machines with more resources. Coupled with the hardware evolution from Ice Lake-SP to Emerald Rapids—which brings a dramatic increase in cache size, improved frequency dynamics, and advanced instruction support—the new v6 generation sets a new standard for high-performance workloads. Whether you’re running critical enterprise applications, data-intensive analytics, or next-generation AI models, the enhanced capabilities of these VMs offer significant benefits in performance, efficiency, and cost-effectiveness. References and Further Reading: Microsoft’s official announcement: Azure Dld e6 VMs Internal tests performed with Geekbench 6.4.0 (AVX2) in the Germany West Central Azure region.Linux Virtual Machine Agent Status "Not Ready"
We currently have a CEF server deployed in Azure which is a Linux VM. This morning I had no logs in sentinel and checked on the vm and noticed there was an error stating the Agent Status is "Not Ready". Having a hard time finding a solution to this problem, has anyone had this issue before? Thanks.
Network Monitoring
Hi, I recently applied Network Security Groups on Virtual Networks (NSG). Now my question is, is it possible to monitor / record the network traffic? For example, I've configured many rules on the NSG, now a application on a Server won't work and my first guess is the NSG is blocking the communication. How do I see now which port the application is using so I can set a new rule to the NSG? I know when you already know the port you can check it in Network Watcher "IP flow verify and NSG diagnostics" as a whatif state. Traffic Analytics isn't the right answer too or am I seeing it wrong? Vnet Flow Logs should be the right thing. I configured it, applied traffic analytics and a account storage. Applied it for testing on a nic but I don't see anything practical for my use? The only thing Iwish is to see live or logged the traffic if the NSG blocked anything and troubleshoot.VM DSC Extension - Repository doesn't match Reality
Hi, I am provisioning VMs as session hosts for AVD, using Entra ID for login. During the deployment process, one of the resources is Microsoft.Compute/virtualMachines/extensions/Microsoft.PowerShell.DSC. Based on the deployment information, the artifacts (scripts) that are being used come from this path: https://wvdportalstorageblob.blob.core.windows.net/galleryartifacts/Configuration_1.0.02893.601.zip The confusing part is that when I view those files, they are a bit different than the ones stored in the Azure GitHub repo here: https://github.com/Azure/RDS-Templates/tree/master/ARM-wvd-templates/DSC Is the GitHub repo just out of date or need review? I have questions on the implementation of the scripts, but it seems like figuring this out is the necessary first step.
Determining sizing requirements for GPU enabled Azure VM
Greetings, We are trying to determine the correct VM sizing requirement for our AI workload, which is used for NLP processing. This workload does not require any training, but will only be used for inference. We have the following software configuration: a C# application that is heavily multithreaded using a lot of socket I/O. The application has concentrated bursts where 10-20 threads are fired concurrently to perform tasks (mostly socket I/O). This app communicates via dedicated sockets to: a Python application which performs various NLP tasks. This app is also multithreaded to handle multiple incoming requests from the .NET app. This app sends queries to a local LLM (model size will vary based on query type). We estimate we will need to support sub-second performance (at the very least) on a 7B parameter model. Ultimately, we may need to go to larger model sizes if accuracy is insufficient. The amount of text passed to the LLM will range from 300-3000 tokens. In short, we need: a) a CPU with sufficient cores to handle multiple concurrent threads on the .NET side. The app will have 5 or 6 background threads running continuously, and sudden bursts of activity which will require a minimum of 10-20 threads to run shorter-lived tasks. b) a GPU with sufficient VRAM to handle at the very least, a 7B parameter model. Ultimately, we may need to support larger models to perform the same task due to insufficient accuracy. We need the ideal configuration of GPU/VRAM and CPU/RAM to handle these tasks, and also, potentially, larger LLM sizes of up to 14B or 70B parameters. We are looking at the NC-series VMs, with a budget of about $1,000/month (see https://azure.microsoft.com/en-us/pricing/details/virtual-machines/windows/#pricing). Any feedback on the optimal configuration in terms of CPU/GPU would be greatly appreciated. Thank you in advance.
Adding VM Instance View Details, e.g. osName, to the VM Resource Object JSON (for Custom Policy Use)
I'm requesting to add more details to the JSON of the VM resource object, particularly from the VM instance view data. This is to include operating system information, such as the name and version (osName and osVersion), for use in a custom Policy. Although these details are visible in the portal, they're not present in the VM's resource object, which is necessary for our custom policy.
