Technical Intelligence Report: 2026-02-18

Executive Summary

  • Linux 7.0 Kernel Updates for AMD: The upcoming Linux 7.0 kernel now includes KVM virtualization support for AMD’s Zen 5 ERAPS (Enhanced Return Address Predictor Security), doubling the Return Stack Buffer capacity for guests.
  • GPU Market Inflation: A global report indicates a 15% average price hike for GPUs since Nov 2025. While NVIDIA’s RTX 5090 surged 32%, AMD’s RX 9070 XT remained the most stable high-end card, seeing only a 7% average increase.
  • FreeBSD Desktop Improvements: The FreeBSD Foundation has readied a KDE Desktop installer option, simplifying the deployment of open-source graphics drivers (including AMD’s) for BSD users.
  • NVIDIA’s Strategic Dominance in India: NVIDIA announced massive infrastructure and software wins in India, including deployments of Blackwell Ultra and Grace Blackwell systems with local clouds (Yotta, E2E) and deep integration with major Systems Integrators (Infosys, Wipro, TCS) and industrial giants (Siemens, Reliance).

🤖 ROCm Updates & Software

[2026-02-18] KVM In Linux 7.0 Adds Support For Virtualizing AMD ERAPS

Source: Phoronix

Key takeaway relevant to AMD:

  • Enables new Zen 5 security/performance features in virtualized environments, ensuring AMD server chips remain competitive for cloud providers.
  • Improves SEV-SNP (Secure Encrypted Virtualization) workflow by allowing certificate fetching from user-space.

Summary:

  • Linux 7.0 kernel merges bring updates to KVM virtualization.
  • Specific focus on enabling AMD Zen 5 architectural features within Guest VMs.
  • Includes updates for other architectures (Intel, LoongArch, RISC-V, ARM).

Details:

  • AMD ERAPS (Enhanced Return Address Predictor Security):
    • Introduced on Zen 5 processors.
    • The Linux 7.0 KVM update allows virtualizing this feature.
    • Technical Impact: Allows guest VMs to utilize a larger Return Stack Buffer (RSB). The RSB context is increased to 64 entries (previously limited to 32 entries).
  • SEV-SNP Updates: The AMD SVM code now supports fetching SEV-SNP certificates directly from user-space, streamlining confidential computing workflows.
  • Other Arch Updates:
    • Intel: Guest ownership of PMU hardware.
    • RISC-V: Zalasr, Zilsd, and Zclsd extension support.
    • ARM: Preliminary guest GICv5 support.

[2026-02-18] FreeBSD’s KDE Desktop Install Option Ready For Testing

Source: Phoronix

Key takeaway relevant to AMD:

  • Simplifies the installation of AMD GPU drivers on FreeBSD, potentially increasing the user base for AMD hardware in non-Linux Unix-like environments.
  • Leverages open-source graphics driver code updated from Linux.

Summary:

  • FreeBSD is testing a new desktop installer option to easily deploy KDE Plasma.
  • The update aims to automate the setup of SDDM, X.Org, and GPU drivers.
  • Targeting release in FreeBSD 15.1 (June).

Details:

  • Integration: The installer option provides a text-based selection to install the full desktop stack (KDE Plasma) and the “supported GPU drivers.”
  • Driver Sync: The update relies on ongoing work to update the open-source graphics driver code from Linux, ensuring newer AMD GPU generations function correctly on FreeBSD.
  • Power Management: Includes merged improvements for modern s0ix standby and ongoing work for S4 hibernation, critical for AMD laptop users running FreeBSD.

🤼‍♂️ Market & Competitors

[2026-02-18] GPU Price Report: 15% Average Global Hike

Source: Tom’s Hardware

Key takeaway relevant to AMD:

  • Competitive Advantage: The AMD RX 9070 XT is identified as the most price-stable high-end GPU globally (+7% vs Industry Avg +15%), making it a stronger value proposition against the RTX 50-series.
  • AMD pricing in Brazil actually decreased (-9% for 9070 XT), offering a specific regional sales opportunity.

Summary:

  • TechSpot data compares Nov 2025 vs. Feb 2026 GPU prices globally.
  • High-end cards saw the steepest inflation.
  • The $1,000 buying power has eroded significantly (buying an RTX 5070 Ti today vs. an RTX 5080 three months ago).

Details:

  • AMD Metrics:
    • RX 9070 XT: +7.5% global average increase (lowest fluctuation).
    • RX 9070: +7.0% global average increase.
    • Regional variance: The 9070 XT rose 21% in the USA but dropped 9% in Brazil.
  • NVIDIA Metrics:
    • RTX 5090: +31.6% global average; currently sells ~65% above MSRP.
    • RTX 5080: +24.6% global average.
    • RTX 5070 Ti: +25.0% global average.
  • Intel Metrics:
    • Arc B580: +10.7% global average.
  • Economic Context: The RTX 5090 is now retailing over $3,500 in the US (vs $2,500 in Nov 2025).

[2026-02-18] India Fuels Its AI Mission With NVIDIA

Source: NVIDIA Blog

Key takeaway relevant to AMD:

  • NVIDIA has secured a near-monopoly on India’s Sovereign AI infrastructure, with massive deployments of Blackwell/Grace Blackwell hardware.
  • This locks Indian startups and government initiatives into the CUDA/NIM ecosystem, raising the barrier to entry for AMD ROCm in this region.

Summary:

  • NVIDIA highlights major partnerships at the AI Impact Summit in New Delhi.
  • Focus on the “IndiaAI Mission” ($1B+ investment).
  • Deployment of sovereign AI clouds and Hindi-language models.

Details:

  • Hardware Deployments:
    • Yotta: Deploying “Shakti Cloud” with over 20,000 NVIDIA Blackwell Ultra GPUs.
    • E2E Networks: Building clusters with NVIDIA HGX B200 systems.
    • Netweb: Launching “Tyrone Camarero” systems based on NVIDIA Grace Blackwell (GB200 NVL4), manufactured locally in India.
  • Software & Models:
    • Nemotron: Widespread adoption of NVIDIA Nemotron open models for Indic languages (22 languages).
    • Startups:
      • BharatGen: 17B parameter Mixture-of-Experts (MoE) model.
      • Sarvam.ai: Training Sarvam-3 (3B, 30B, 100B params) on H100s using NeMo framework.
      • Gnani.ai: 14B speech-to-speech model; claims 15x reduction in inference costs via Nemotron fine-tuning.

[2026-02-18] India’s Global Systems Integrators Build Next Wave of Enterprise Agents With NVIDIA AI

Source: NVIDIA Blog

Key takeaway relevant to AMD:

  • Major Global Systems Integrators (GSIs) like Infosys, Wipro, and Tech Mahindra are standardizing their “Agentic AI” platforms on NVIDIA AI Enterprise and NIM (NVIDIA Inference Microservices).
  • This standardization makes it difficult for AMD to displace NVIDIA in large enterprise contracts managed by these GSIs.

Summary:

  • Leading Indian IT firms are building “Agentic AI” platforms using the NVIDIA stack.
  • Use cases focus on call centers, telecom ops, and coding assistants.

Details:

  • Wipro:
    • Platform: WEGA, powered by NVIDIA AI Enterprise.
    • Metrics: 42% of inbound calls handled by AI; sub-200ms latency on 900 concurrent calls.
  • Tech Mahindra:
    • Use Case: Autonomous network operations for Telcos.
    • Tech: Uses NVIDIA Nemotron embedding models and NIM microservices for inference.
  • Infosys:
    • Tech: Developed a 2.5-billion-parameter small language model (SLM) for coding using NVIDIA NeMo.
    • Benchmarks: Matches frontier performance on MBPP and MBPP+.
  • Persistent Systems:
    • Tech: Using NVIDIA BioNeMo for molecular discovery (GenMoIVS solution).

[2026-02-18] NVIDIA and Global Industrial Software Leaders Partner With India’s Largest Manufacturers

Source: NVIDIA Blog

Key takeaway relevant to AMD:

  • NVIDIA is cementing its position in “Physical AI” and Digital Twins via Omniverse, partnering with industrial software leaders (Siemens, Synopsys) used by manufacturing giants.
  • AMD’s lack of a direct competitor to the Omniverse ecosystem limits its reach in the “Software-Defined Factory” market.

Summary:

  • Indian manufacturers are adopting NVIDIA Omniverse and CUDA-X libraries for digital twins and robotics.
  • Key partnerships involve Siemens, Cadence, and Synopsys.

Details:

  • Reliance New Energy: Using Siemens digital twin tech + NVIDIA Omniverse for Gigafactory simulation.
  • Addverb Technologies: Using NVIDIA Cosmos world foundation models to train robots in simulation.
  • Havells India: Using Synopsys Ansys Fluent accelerated by NVIDIA CUDA-X; achieved 6x faster fluid dynamic simulations.
  • Larsen & Toubro Semiconductor: Using Cadence Spectre X (CUDA-X accelerated) for chip design.
  • TCS (Tata Consultancy Services): Deploying NVIDIA Metropolis for video search and Omniverse for digital twins at Tata Motors.

📈 GitHub Stats

Category Repository Total Stars 1-Day 7-Day 30-Day
AMD Ecosystem AMD-AGI/GEAK-agent 63 0 +1  
AMD Ecosystem AMD-AGI/Primus 74 0 0 +8
AMD Ecosystem AMD-AGI/TraceLens 58 0 0 +4
AMD Ecosystem ROCm/MAD 31 0 0  
AMD Ecosystem ROCm/ROCm 6,173 +1 +13  
Compilers openxla/xla 3,995 +3 +17 +82
Compilers tile-ai/tilelang 5,208 +1 +50  
Compilers triton-lang/triton 18,445 +7 +43 +262
Google / JAX AI-Hypercomputer/JetStream 407 0 +1  
Google / JAX AI-Hypercomputer/maxtext 2,139 0 +1  
Google / JAX jax-ml/jax 34,897 +12 +57  
HuggingFace huggingface/transformers 156,602 +32 +242  
Inference Serving alibaba/rtp-llm 1,049 0 +2  
Inference Serving efeslab/Atom 336 0 0  
Inference Serving llm-d/llm-d 2,504 +5 +27  
Inference Serving sgl-project/sglang 23,583 +22 +75  
Inference Serving vllm-project/vllm 70,541 +85 +471 +2650
Inference Serving xdit-project/xDiT 2,542 0 +7  
NVIDIA NVIDIA/Megatron-LM 15,222 +3 +34  
NVIDIA NVIDIA/TransformerEngine 3,163 0 +4  
NVIDIA NVIDIA/apex 8,925 +2 +10  
Optimization deepseek-ai/DeepEP 8,991 +1 +16  
Optimization deepspeedai/DeepSpeed 41,631 0 +34  
Optimization facebookresearch/xformers 10,341 +2 +7  
PyTorch & Meta meta-pytorch/monarch 971 +2 +3  
PyTorch & Meta meta-pytorch/torchcomms 333 +1 +3  
PyTorch & Meta meta-pytorch/torchforge 621 0 +5  
PyTorch & Meta pytorch/FBGEMM 1,535 +1 +5  
PyTorch & Meta pytorch/ao 2,691 0 +19  
PyTorch & Meta pytorch/audio 2,831 +1 +4  
PyTorch & Meta pytorch/pytorch 97,484 +25 +147  
PyTorch & Meta pytorch/torchtitan 5,076 0 +14  
PyTorch & Meta pytorch/vision 17,517 +3 +10  
RL & Post-Training THUDM/slime 4,239 +23 +496  
RL & Post-Training radixark/miles 885 +3 +23  
RL & Post-Training volcengine/verl 19,256 +10 +105