Built For What You Do Next.
Embedded developers need determinism, performance, bandwidth, and industrial endurance in a single device that delivers it all. And they shouldn’t need to trade one requirement for another to get it.
With AMD Ryzen™ AI Embedded X100 Series CPUs, designers can consolidate compute, graphics, and AI on one embedded x86 platform while preserving predictable performance for real-time applications. Full-performance AMD “Zen 5” cores, AVX-512 support, unified memory, and a shared 32 MB MALL cache help keep demanding workloads moving without forcing trade-offs between performance, determinism, and integration complexity.
Ryzen AI Embedded X199 CPU Sets a New Embedded Performance Standard
vs. Intel® Core™ Ultra X7 358H
SPECrate®2017_int_base
Consolidate workloads on one APU and simplify scheduling with full-performance cores designed for predictable throughput.
STREAM
Keep sensor fusion, AI, and graphics workloads moving through one shared memory pool.
GFXBench®
Render real-time visualization smoothly on the same APU that's running compute and control.
llama-bench
Run local LLM inference faster on the same APU, helping edge systems respond sooner without added complexity.
Ryzen AI Embedded X100 vs. Intel Ultra X7 358H
- CPU Performance
- Graphics
- Memory Throughput
- AI Inference
CPU Multi-Threading Performance
45W TDP, Normalized to Intel = 1.0
AMD Ryzen AI Embedded X199
Intel Ultra X7 358H
Based on benchmark scores from AMD testing as of May 2026. See Footnotes REX-007, REX-010, REX-006 and REX-005 for details.
Graphics Performance
45W TDP, Normalized to Intel = 1.0
AMD Ryzen AI Embedded X199
Intel Ultra X7 358H
Based on benchmark scores from AMD testing as of May 2026. See Footnotes REX-008, REX-002, REX-003 and REX-004 for details.
Sustained Memory Throughput
Geomean, GCC15
AMD Ryzen AI Embedded X199
Intel Ultra X7 358H
Based on benchmark scores from AMD testing as of May 2026. See Footnote REX-009 for details.
Token Generation (Tokens/sec)
Llama-bench with Vulkan backend
AMD Ryzen AI Embedded X199
Intel Ultra X9 358H
Based on benchmark scores from AMD testing as of May 2026. See Footnote REX-017 for details.
AMD Ryzen AI Embedded X100 Series: Consolidating Compute, Graphics, and AI at the Edge
See how a single AMD x86 APU helps you build more capable embedded systems by consolidating compute, graphics, and AI workloads, reducing data movement, and giving teams one platform to scale.
One APU. Every Workload. Full Performance.
Ryzen AI Embedded X100 processors deliver what embedded designs demand: determinism, full-performance cores, and long lifecycle.
| AMD Ryzen AI Embedded X100 |
Intel Core Ultra Series 3 |
NVIDIA® Jetson Orin™ and Thor™ |
|
| One Architecture, Every Workload |
|
|
|
| Deploy Once, Run for a Decade |
|
|
|
| Every Core Pulls Full Weight | Full AVX-512 support on all cores, no double-pumping | AVX-256 | 128-bit vector width (NEON) |
| Twice the Threads | Simultaneous Multithreading (SMT) on all cores | No SMT | No SMT |
Don’t Re-Write. Migrate.
Your CUDA® Investment Comes with You
Preserve more of your existing CUDA investment with AMD HIPIFY and the open AMD ROCm™ software stack. HIPIFY converts CUDA code to portable HIP C++, so teams can move quickly to AMD platforms while reducing rewrite effort and avoiding proprietary lock-in.
CUDA
// CUDA Source
// Declare I/O vector pointers
float *a, *b, *c;
// Allocate managed memory
cudaMallocManaged(&a, bytes);
cudaMallocManaged(&b, bytes);
cudaMallocManaged(&c, bytes);
// CPU writes directly
for (int i = 0; i < n; ++i) { a[i] = ...; b[i] = ...; }
// GPU uses the same pointers
vecAdd<<<blocks, threads>>>(a, b, c, n);
cudaDeviceSynchronize();
// CPU reads results directly
use(c);
HIPIFY
HIP
// HIP Source
// Declare I/O vector pointers
float *a, *b, *c;
// Allocate managed memory
hipMallocManaged(&a, bytes);
hipMallocManaged(&b, bytes);
hipMallocManaged(&c, bytes);
// CPU writes directly
for (int i = 0; i < n; ++i) { a[i] = ...; b[i] = ...; }
// GPU uses the same pointers
vecAdd<<<blocks, threads>>>(a, b, c, n);
hipDeviceSynchronize();
// CPU reads results directly
use(c);
Ryzen AI X100 Series Embedded Processors Give You Headroom to Spare
When perception, inference, and control all run at once, GPU-heavy platforms can become CPU constrained. Learn how X100 delivers more headroom for robotics navigation.
Retire the Fixed-Function AV Rack
Run multi-channel 4K encode/decode, AV-over-IP routing, and on-device AI on a single x86 APU. Add codecs and features in software, not new hardware.
Agentic AI
AMD Ryzen AI Embedded X100 Series Processors Scale Agentic AI at the Edge
with the mimik Agentix Operating Engine (mimOE)
In new third-party technical papers sponsored by AMD, mimik models production-scale agentic AI across heterogeneous SoCs and show how Ryzen AI Embedded X100 Series processors with mimOE can support more concurrent agentic AI workflows on the same edge platform, helping developers scale local intelligence without overprovisioning hardware.
Robotics
AMD Ryzen AI Embedded X100 Series Processors Deliver Real-Time Results
for high-performance autonomous robotics
A new full-system benchmark developed by Open Navigation and sponsored by AMD demonstrates the need for balanced performance in modern autonomous robotics applications. The results show that that Ryzen AI Embedded X100 Series processors offer superior real-time performance and more CPU headroom for modern robotics workloads.
Frequently Asked Questions
AMD Ryzen AI Embedded X100 Series processors are high-performance embedded APUs that combine x86 CPU cores, integrated graphics, NPU acceleration, unified memory, and embedded lifecycle features for physical AI and real-time edge systems.
AMD Ryzen AI Embedded X100 Series processors pair full performance x86 CPU cores with integrated graphics, an NPU, and unified memory on one APU, giving robotics developers the CPU headroom to run perception, planning, and control alongside AI inference, without the CPU/GPU imbalance and intergenerational incompatibility of NVIDIA Jetson modules.6
Yes. Ryzen AI Embedded X100 Series processors consolidate CPU, GPU, and NPU onto one x86 APU with unified memory, giving you the CPU headroom Orin lacks to run the full navigation and control stack.6 Open ROCm software and HIP tooling lets you bring existing code across without proprietary lock-in.
Ryzen AI Embedded X100 Series processors integrate up to 16 uniform AMD “Zen 5” cores with no P-core/E-core scheduling guesswork, full AVX-512, SMT, unified memory, and a long industrial lifecycle, helping developers build real-time systems designed to reduce scheduling complexity versus Intel’s mobile-first design.
Peak TOPS is a theoretical marketing metric that shows a device’s peak potential capacity. It ignores I/O bandwidth, memory bandwidth, and memory and cache architecture, all of which drive GPU and system-level efficiency. Ryzen AI Embedded X100 Series is architected for real-world workloads that require balanced, deterministic compute.
Unlike Intel Core Ultra Series 3, which relies on two separate foundries (Intel and TSMC), three process geometries, and Intel’s complex proprietary Foveros-S packaging, Ryzen AI Embedded X100 Series processors are fabricated on a mature TSMC Arizona 4 nm process and use proven and mature chiplet packaging.
AMD Ryzen AI Embedded X100 Series processors will be available for up to 10 years. For industrial designs, X100 processors support up to 10 years of 24/7 continuous operation and up to -40°C to 105°C temperature ranges, with chip-down design options, Kria AI SOMs, a robust ODM ecosystem, and mature TSMC 4 nm technology for long-lifecycle embedded deployments
AMD HIPIFY converts CUDA to portable HIP C++ that runs on the AMD Ryzen AI Embedded X100 Series, and the open ROCm software stack lets you leverage your CUDA investment without proprietary lock-in.
x86 compatibility lets developers preserve existing software investments, reuse familiar development and validation flows, and scale applications from workstations to deployed edge fleets. Many legacy physical AI systems pair a discrete GPU with a separate x86 CPU; Ryzen AI Embedded X100 Series processors integrate these into one device for high determinism.
A unified memory pool lets the CPU, GPU, and NPU share data more efficiently, helping reduce data movement across sensor fusion, AI inference, graphics, planning, and real-time control workloads.
Ryzen AI Embedded X100 Series is designed for an open AI software ecosystem, including the ROCm software stack, ONNX Runtime, PyTorch, Ryzen AI software, and familiar x86 tools for embedded development.
Ryzen AI Embedded X100 Series processors are designed for robotics, industrial automation, machine vision, pro AV, casino gaming, healthcare imaging, aerospace and defense, radar, drones, and test and measurement systems.
Footnotes
- Estimated by AMD as of May 2026 on the AMD Ryzen AI Max+ 395 (OPN 100-000002199) at 45W SPL (sPPT=45W, fPPT=64W) and the Intel Core Ultra 7 358H at 30W SPL (PL1=45W, PL2=64W) using SPEC CPU2017_int_base (configuration: GCC-15, rate, harness cpu2017-1.1.9, int base sub score) on test systems comprising: Ryzen AI Max+ 395 (OPN 100-000002199): AMD Ryzen AI Embedded X100 Series Reference board, 64 GB LPDDR5X-8000 SK Hynix RAM, Kingston Technology KV3000 NVMe SSD 512 GB PCIe Gen4 512 GB storage, Linux® Ubuntu® 24.04 with Kernel 6.18.15-14-amd (x86_64), Insyde BIOS REM60070A, SMT Enabled, STAPM Disabled. Intel Core Ultra X7 358H: MSI Prestige 16 Flip AI+ laptop C3MTG MS-2622, 32 GB LPDDR5X-8533 SK Hynix RAM, Kingston Technology KV3000 NVMe SSD 512 GB PCIe Gen4 512 GB storage, Linux Ubuntu 24.04 with Kernel 6.17.0-23-generic (x86_64), MSI BIOS E2622IMS.10E. Ryzen AI Max+ 395 (OPN 100-000002199) device is configured to reflect Ryzen AI Embedded X199 specifications. Intel data measured by AMD at 30W. 45W estimates were derived from AMD 30W measurements based on publicly available benchmark data. OEM published scores will vary based on system configuration and determinism mode used (default performance profile) and other factors. (REX-010) Testing conducted by AMD as of May 2026 on an AMD Ryzen AI Max+ 395 (OPN 100-000002199) configured to reflect Ryzen AI Embedded X199 specifications (STAPM disabled; 45W SPL; sPPT 45W; fPPT 64W; LPDDR5X-8000), compared to an Intel Core Ultra X7 358H (30W; PL1 45W; PL2 64W; LPDDR5X-8533). Results reflect relative geometric mean performance across STREAM (GCC 15) on Ubuntu 24.04. System manufacturers may vary configurations, yielding different results. Results may vary based on configuration, settings, usage, and other factors. (REX-009)
- Measurements and projections conducted by AMD as of May 2026 on an AMD Ryzen AI Max+ 395 processor (OPN 100-000002199) configured to reflect Ryzen AI Embedded X199 specifications (AMD Maple CRB, 5.1 GHz CPU, 2.9 GHz GPU, with sustained operation at 45W TDP/sPPT/fPPT, STAPM disabled and 64 GB soldered LPDDR5X-8000), compared to an Intel Core Ultra X7 358H (MSI Prestige 16 Flip AI+ C3MTG MS-2622, 5.1 GHz CPU, 2.5 GHz GPU, PL1 = 45W, PL2 = 64W, MSI BIOS E2622IMS.10E, 32 GB LPDDR5X-8533, measured 30W OEM Sustained Power Limit). Results reflect the relative geomean of offscreen Vulkan benchmark frame rates from GFXBench 5.0.0 on Ubuntu 24.04, comparing measured AMD performance at 45W sustained power to 45W Intel performance projected from 30W sustained power measurements, using scaling factors derived from public benchmark data for the 358H. System manufacturers may vary configurations, yielding different results. Results may vary based on configuration, settings, usage, and other factors. (REX-002)
- Based on AMD internal testing as of July 2026, an AMD Ryzen AI Max+ 395 processor, configured to reflect Ryzen AI Embedded X199 specifications (AMD Maple CRB, 5.1 GHz CPU, 2.9 GHz GPU, with sustained operation at 45W TDP/sPPT/fPPT, and 64 GB soldered LPDDR5X-8000) was used to measure inference throughput. Results are compared against Intel Core Ultra X7 358H (MSI Prestige 16 Flip AI+ C3MTG MS-2622, 5.1 GHz CPU, 2.5 GHz GPU, PL1 = 45W, PL2 = 64W, MSI BIOS E2622IMS.10E, 32 GB LPDDR5X-8533, performance mode), and reflect the geometric mean of llama-bench (build: llama-b9453-vulkan) performance across gemma4 26B.A4B MXFP4 MoE, gemma4 26B.A4B Q4_K - Medium, Llama 3.1 8B Q4_K - Medium, qwen35 27B Q4_K - Medium, qwen35moe 35B.A3B MXFP4 MoE, qwen35moe 35B.A3B Q4_K - Medium models. All models fit in <24 GB RAM. System manufacturers may vary configurations, yielding different results. Results may vary based on configuration, settings, usage, and other factors. (REX-017)
- Based on the mimik whitepaper “Architectural Fit for Production-Scale Agentic AI on Heterogeneous SoCs” commissioned by AMD, published by mimik on July 23, 2026, based on a modeled sweep of 455 feasible agentic AI workflows across two device classes (X100, NVIDIA Jetson T5000), checking spare CPU, GPU, and memory. For more information see: https://www.mimik.com/agentix-compute-benchmarking (REX-019)
- Based on the OpenNav Robotics Workload Benchmark commissioned by AMD, published by Open Navigation LLC on July 23, 2026, as measured on the GMKtec EVO-X2 AI Mini PC AMD Ryzen AI Max+ 395 configured to reflect Ryzen AI Embedded X199 specifications (5.1 GHz CPU, 2.9 GHz GPU, 120W TDP, LPDDR5X-7500), vs. the NVIDIA Jetson AGX Thor Developer Kit: https://opennav.org/news/opennav-robotics-workload-benchmark (REX-016)
- Measurements and projections conducted by AMD as of May 2026 on an AMD Ryzen AI Max+ 395 processor (OPN 100-000002199) configured to reflect Ryzen AI Embedded X199 specifications (AMD Maple CRB, 5.1 GHz CPU, 2.9 GHz GPU, with sustained operation at 45W TDP/sPPT/fPPT, STAPM disabled and 64 GB soldered LPDDR5X-8000), compared to an Intel Core Ultra X7 358H (MSI Prestige 16 Flip AI+ C3MTG MS-2622, 5.1 GHz CPU, 2.5 GHz GPU, PL1 = 45W, PL2 = 64W, MSI BIOS E2622IMS.10E, 32 GB LPDDR5X-8533, measured 30W OEM Sustained Power Limit). Results reflect projected relative CoreMark v1.01 multi-thread performance on Ubuntu 24.04, comparing measured AMD performance at 45W sustained power to 45W Intel performance projected from 30W sustained power measurements, using scaling factors derived from public benchmark data for the 358H. System manufacturers may vary configurations, yielding different results. Results may vary based on configuration, settings, usage, and other factors. (REX-007)
- Measurements and projections conducted by AMD as of May 2026 on an AMD Ryzen AI Max+ 395 processor (OPN 100-000002199) configured to reflect Ryzen AI Embedded X199 specifications (AMD Maple CRB, 5.1 GHz CPU, 2.9 GHz GPU, with sustained operation at 45W TDP/sPPT/fPPT, STAPM disabled and 64 GB soldered LPDDR5X-8000), compared to an Intel Core Ultra X7 358H (MSI Prestige 16 Flip AI+ C3MTG MS-2622, 5.1 GHz CPU, 2.5 GHz GPU, PL1 = 45W, PL2 = 64W, MSI BIOS E2622IMS.10E, 32 GB LPDDR5X-8533, measured 30W OEM Sustained Power Limit). Results reflect projected relative PassMark 11.0.1002 multi-thread performance on Ubuntu 24.04, comparing measured AMD performance at 45W sustained power to 45W Intel performance projected from 30W sustained power measurements, using scaling factors derived from public benchmark data for the 358H. System manufacturers may vary configurations, yielding different results. Results may vary based on configurations, settings, usage, and other factors. (REX-006)
- Measurements and projections conducted by AMD as of May 2026 on an AMD Ryzen AI Max+ 395 processor (OPN 100-000002199) configured to reflect Ryzen AI Embedded X199 specifications (AMD Maple CRB, 5.1 GHz CPU, 2.9 GHz GPU, with sustained operation at 45W TDP/sPPT/fPPT, STAPM disabled and 64 GB soldered LPDDR5X-8000), compared to an Intel Core Ultra X7 358H (MSI Prestige 16 Flip AI+ C3MTG MS-2622, 5.1 GHz CPU, 2.5 GHz GPU, PL1 = 45W, PL2 = 64W, MSI BIOS E2622IMS.10E, 32 GB LPDDR5X-8533, measured 30W OEM Sustained Power Limit). Results reflect projected relative Geekbench 6.1.0 multi-thread performance on Ubuntu 24.04, comparing measured AMD performance at 45W sustained power to 45W Intel performance projected from 30W sustained power measurements, using scaling factors derived from public benchmark data for the 358H. System manufacturers may vary configurations, yielding different results. Results may vary based on configuration, settings, usage, and other factors. (REX-005)
- Measurements and projections conducted by AMD as of May 2026 on an AMD Ryzen AI Max+ 395 processor (OPN 100-000002199) configured to reflect Ryzen AI Embedded X199 specifications (AMD Maple CRB, 5.1 GHz CPU, 2.9 GHz GPU, with sustained operation at 45W TDP/sPPT/fPPT, STAPM disabled and 64 GB soldered LPDDR5X-8000), compared to an Intel Core Ultra X7 358H (MSI Prestige 16 Flip AI+ C3MTG MS-2622, 5.1 GHz CPU, 2.5 GHz GPU, PL1 = 45W, PL2 = 64W, MSI BIOS E2622IMS.10E, 32 GB LPDDR5X-8533, measured 30W OEM Sustained Power Limit). Results reflect the relative geomean across Phoronix Test Suite 10.8.6 benchmarks (GravityMark, Unigine Heaven/Valley, Unvanquished, GLmark2, ParaView, and GpuTest) on Ubuntu 24.04, comparing measured AMD performance at 45W sustained power to 45W Intel performance projected from 30W sustained power measurements, using scaling factors derived from public benchmark data for the 358H. System manufacturers may vary configurations, yielding different results. Results may vary based on configuration, settings, usage, and other factors. (REX-008)
- Measurements and projections conducted by AMD as of May 2026 on an AMD Ryzen AI Max+ 395 processor (OPN 100-000002199) configured to reflect Ryzen AI Embedded X199 specifications (AMD Maple CRB, 5.1 GHz CPU, 2.9 GHz GPU, with sustained operation at 45W TDP/sPPT/fPPT, STAPM disabled and 64 GB soldered LPDDR5X-8000), compared to an Intel Core Ultra X7 358H (MSI Prestige 16 Flip AI+ C3MTG MS-2622, 5.1 GHz CPU, 2.5 GHz GPU, PL1 = 45W, PL2 = 64W, MSI BIOS E2622IMS.10E, 32 GB LPDDR5X-8533, measured 30W OEM Sustained Power Limit). Results reflect the relative geomean of offscreen OpenGL frame rates from GFXBench 5.0.0 on Ubuntu 24.04, comparing measured AMD performance at 45W sustained power to 45W Intel performance projected from 30W sustained power measurements, using scaling factors derived from public benchmark data for the 358H. System manufacturers may vary configurations, yielding different results. Results may vary based on configuration, settings, usage, and other factors. (REX-003)
- Measurements and projections conducted by AMD as of May 2026 on an AMD Ryzen AI Max+ 395 processor (OPN 100-000002199) configured to reflect Ryzen AI Embedded X199 specifications (AMD Maple CRB, 5.1 GHz CPU, 2.9 GHz GPU, with sustained operation at 45W TDP/sPPT/fPPT, STAPM disabled and 64 GB soldered LPDDR5X-8000), compared to an Intel Core Ultra X7 358H (MSI Prestige 16 Flip AI+ C3MTG MS-2622, 5.1 GHz CPU, 2.5 GHz GPU, PL1 = 45W, PL2 = 64W, MSI BIOS E2622IMS.10E, 32 GB LPDDR5X-8533, measured 30W OEM Sustained Power Limit). Results reflect projected relative performance running the Unigine Heaven 4.0 Pro “Extreme” preset on Ubuntu 24.04, comparing measured AMD performance at 45W sustained power to 45W Intel performance projected from 30W sustained power measurements, using scaling factors derived from public benchmark data for the 358H. System manufacturers may vary configuration, yielding different results. Results may vary based on configuration, settings, usage, and other factors. (REX-004)
- Based on the OpenNav Robotics Workload Benchmark commissioned by AMD, published by Open Navigation LLC on July 23, 2026, as measured on the GMKtec EVO-X2 AI Mini PC AMD Ryzen AI Max+ 395 configured to reflect Ryzen AI Embedded X199 specifications (5.1 GHz CPU, 2.9 GHz GPU, 120W TDP, LPDDR5X-7500), vs. the NVIDIA Jetson AGX Thor Developer Kit: https://opennav.org/news/opennav-robotics-workload-benchmark (REX-018)
© 2026 Advanced Micro Devices, Inc. All rights reserved. AMD, the AMD Arrow logo, EPYC, Ryzen, and combinations thereof are trademarks of Advanced Micro Devices, Inc. CUDA, NVIDIA, Orin and Thor are trademarks of NVIDIA Corporation. Core and Intel are trademarks of Intel Corporation. CoreMark is a registered trademark of the Embedded Microprocessor Benchmark Consortium (EEMBC). Linux is the registered trademark of Linus Torvalds in the U.S. and other countries. OpenGL® and the oval logo are trademarks or registered trademarks of Hewlett Packard Enterprise in the United States and/or other countries worldwide. PassMark is a registered trademark of PassMark Software Pty Ltd. SPEC CPU and SPECrate are trademarks or registered trademarks of Standard Performance Evaluation Corporation (SPEC). Ubuntu and the Ubuntu logo are registered trademarks of Canonical Ltd. Vulkan is a trademark of the Khronos Group. Other product names used herein are for identification purposes only and may be trademarks of their respective owners.
- Estimated by AMD as of May 2026 on the AMD Ryzen AI Max+ 395 (OPN 100-000002199) at 45W SPL (sPPT=45W, fPPT=64W) and the Intel Core Ultra 7 358H at 30W SPL (PL1=45W, PL2=64W) using SPEC CPU2017_int_base (configuration: GCC-15, rate, harness cpu2017-1.1.9, int base sub score) on test systems comprising: Ryzen AI Max+ 395 (OPN 100-000002199): AMD Ryzen AI Embedded X100 Series Reference board, 64 GB LPDDR5X-8000 SK Hynix RAM, Kingston Technology KV3000 NVMe SSD 512 GB PCIe Gen4 512 GB storage, Linux® Ubuntu® 24.04 with Kernel 6.18.15-14-amd (x86_64), Insyde BIOS REM60070A, SMT Enabled, STAPM Disabled. Intel Core Ultra X7 358H: MSI Prestige 16 Flip AI+ laptop C3MTG MS-2622, 32 GB LPDDR5X-8533 SK Hynix RAM, Kingston Technology KV3000 NVMe SSD 512 GB PCIe Gen4 512 GB storage, Linux Ubuntu 24.04 with Kernel 6.17.0-23-generic (x86_64), MSI BIOS E2622IMS.10E. Ryzen AI Max+ 395 (OPN 100-000002199) device is configured to reflect Ryzen AI Embedded X199 specifications. Intel data measured by AMD at 30W. 45W estimates were derived from AMD 30W measurements based on publicly available benchmark data. OEM published scores will vary based on system configuration and determinism mode used (default performance profile) and other factors. (REX-010) Testing conducted by AMD as of May 2026 on an AMD Ryzen AI Max+ 395 (OPN 100-000002199) configured to reflect Ryzen AI Embedded X199 specifications (STAPM disabled; 45W SPL; sPPT 45W; fPPT 64W; LPDDR5X-8000), compared to an Intel Core Ultra X7 358H (30W; PL1 45W; PL2 64W; LPDDR5X-8533). Results reflect relative geometric mean performance across STREAM (GCC 15) on Ubuntu 24.04. System manufacturers may vary configurations, yielding different results. Results may vary based on configuration, settings, usage, and other factors. (REX-009)
- Measurements and projections conducted by AMD as of May 2026 on an AMD Ryzen AI Max+ 395 processor (OPN 100-000002199) configured to reflect Ryzen AI Embedded X199 specifications (AMD Maple CRB, 5.1 GHz CPU, 2.9 GHz GPU, with sustained operation at 45W TDP/sPPT/fPPT, STAPM disabled and 64 GB soldered LPDDR5X-8000), compared to an Intel Core Ultra X7 358H (MSI Prestige 16 Flip AI+ C3MTG MS-2622, 5.1 GHz CPU, 2.5 GHz GPU, PL1 = 45W, PL2 = 64W, MSI BIOS E2622IMS.10E, 32 GB LPDDR5X-8533, measured 30W OEM Sustained Power Limit). Results reflect the relative geomean of offscreen Vulkan benchmark frame rates from GFXBench 5.0.0 on Ubuntu 24.04, comparing measured AMD performance at 45W sustained power to 45W Intel performance projected from 30W sustained power measurements, using scaling factors derived from public benchmark data for the 358H. System manufacturers may vary configurations, yielding different results. Results may vary based on configuration, settings, usage, and other factors. (REX-002)
- Based on AMD internal testing as of July 2026, an AMD Ryzen AI Max+ 395 processor, configured to reflect Ryzen AI Embedded X199 specifications (AMD Maple CRB, 5.1 GHz CPU, 2.9 GHz GPU, with sustained operation at 45W TDP/sPPT/fPPT, and 64 GB soldered LPDDR5X-8000) was used to measure inference throughput. Results are compared against Intel Core Ultra X7 358H (MSI Prestige 16 Flip AI+ C3MTG MS-2622, 5.1 GHz CPU, 2.5 GHz GPU, PL1 = 45W, PL2 = 64W, MSI BIOS E2622IMS.10E, 32 GB LPDDR5X-8533, performance mode), and reflect the geometric mean of llama-bench (build: llama-b9453-vulkan) performance across gemma4 26B.A4B MXFP4 MoE, gemma4 26B.A4B Q4_K - Medium, Llama 3.1 8B Q4_K - Medium, qwen35 27B Q4_K - Medium, qwen35moe 35B.A3B MXFP4 MoE, qwen35moe 35B.A3B Q4_K - Medium models. All models fit in <24 GB RAM. System manufacturers may vary configurations, yielding different results. Results may vary based on configuration, settings, usage, and other factors. (REX-017)
- Based on the mimik whitepaper “Architectural Fit for Production-Scale Agentic AI on Heterogeneous SoCs” commissioned by AMD, published by mimik on July 23, 2026, based on a modeled sweep of 455 feasible agentic AI workflows across two device classes (X100, NVIDIA Jetson T5000), checking spare CPU, GPU, and memory. For more information see: https://www.mimik.com/agentix-compute-benchmarking (REX-019)
- Based on the OpenNav Robotics Workload Benchmark commissioned by AMD, published by Open Navigation LLC on July 23, 2026, as measured on the GMKtec EVO-X2 AI Mini PC AMD Ryzen AI Max+ 395 configured to reflect Ryzen AI Embedded X199 specifications (5.1 GHz CPU, 2.9 GHz GPU, 120W TDP, LPDDR5X-7500), vs. the NVIDIA Jetson AGX Thor Developer Kit: https://opennav.org/news/opennav-robotics-workload-benchmark (REX-016)
- Measurements and projections conducted by AMD as of May 2026 on an AMD Ryzen AI Max+ 395 processor (OPN 100-000002199) configured to reflect Ryzen AI Embedded X199 specifications (AMD Maple CRB, 5.1 GHz CPU, 2.9 GHz GPU, with sustained operation at 45W TDP/sPPT/fPPT, STAPM disabled and 64 GB soldered LPDDR5X-8000), compared to an Intel Core Ultra X7 358H (MSI Prestige 16 Flip AI+ C3MTG MS-2622, 5.1 GHz CPU, 2.5 GHz GPU, PL1 = 45W, PL2 = 64W, MSI BIOS E2622IMS.10E, 32 GB LPDDR5X-8533, measured 30W OEM Sustained Power Limit). Results reflect projected relative CoreMark v1.01 multi-thread performance on Ubuntu 24.04, comparing measured AMD performance at 45W sustained power to 45W Intel performance projected from 30W sustained power measurements, using scaling factors derived from public benchmark data for the 358H. System manufacturers may vary configurations, yielding different results. Results may vary based on configuration, settings, usage, and other factors. (REX-007)
- Measurements and projections conducted by AMD as of May 2026 on an AMD Ryzen AI Max+ 395 processor (OPN 100-000002199) configured to reflect Ryzen AI Embedded X199 specifications (AMD Maple CRB, 5.1 GHz CPU, 2.9 GHz GPU, with sustained operation at 45W TDP/sPPT/fPPT, STAPM disabled and 64 GB soldered LPDDR5X-8000), compared to an Intel Core Ultra X7 358H (MSI Prestige 16 Flip AI+ C3MTG MS-2622, 5.1 GHz CPU, 2.5 GHz GPU, PL1 = 45W, PL2 = 64W, MSI BIOS E2622IMS.10E, 32 GB LPDDR5X-8533, measured 30W OEM Sustained Power Limit). Results reflect projected relative PassMark 11.0.1002 multi-thread performance on Ubuntu 24.04, comparing measured AMD performance at 45W sustained power to 45W Intel performance projected from 30W sustained power measurements, using scaling factors derived from public benchmark data for the 358H. System manufacturers may vary configurations, yielding different results. Results may vary based on configurations, settings, usage, and other factors. (REX-006)
- Measurements and projections conducted by AMD as of May 2026 on an AMD Ryzen AI Max+ 395 processor (OPN 100-000002199) configured to reflect Ryzen AI Embedded X199 specifications (AMD Maple CRB, 5.1 GHz CPU, 2.9 GHz GPU, with sustained operation at 45W TDP/sPPT/fPPT, STAPM disabled and 64 GB soldered LPDDR5X-8000), compared to an Intel Core Ultra X7 358H (MSI Prestige 16 Flip AI+ C3MTG MS-2622, 5.1 GHz CPU, 2.5 GHz GPU, PL1 = 45W, PL2 = 64W, MSI BIOS E2622IMS.10E, 32 GB LPDDR5X-8533, measured 30W OEM Sustained Power Limit). Results reflect projected relative Geekbench 6.1.0 multi-thread performance on Ubuntu 24.04, comparing measured AMD performance at 45W sustained power to 45W Intel performance projected from 30W sustained power measurements, using scaling factors derived from public benchmark data for the 358H. System manufacturers may vary configurations, yielding different results. Results may vary based on configuration, settings, usage, and other factors. (REX-005)
- Measurements and projections conducted by AMD as of May 2026 on an AMD Ryzen AI Max+ 395 processor (OPN 100-000002199) configured to reflect Ryzen AI Embedded X199 specifications (AMD Maple CRB, 5.1 GHz CPU, 2.9 GHz GPU, with sustained operation at 45W TDP/sPPT/fPPT, STAPM disabled and 64 GB soldered LPDDR5X-8000), compared to an Intel Core Ultra X7 358H (MSI Prestige 16 Flip AI+ C3MTG MS-2622, 5.1 GHz CPU, 2.5 GHz GPU, PL1 = 45W, PL2 = 64W, MSI BIOS E2622IMS.10E, 32 GB LPDDR5X-8533, measured 30W OEM Sustained Power Limit). Results reflect the relative geomean across Phoronix Test Suite 10.8.6 benchmarks (GravityMark, Unigine Heaven/Valley, Unvanquished, GLmark2, ParaView, and GpuTest) on Ubuntu 24.04, comparing measured AMD performance at 45W sustained power to 45W Intel performance projected from 30W sustained power measurements, using scaling factors derived from public benchmark data for the 358H. System manufacturers may vary configurations, yielding different results. Results may vary based on configuration, settings, usage, and other factors. (REX-008)
- Measurements and projections conducted by AMD as of May 2026 on an AMD Ryzen AI Max+ 395 processor (OPN 100-000002199) configured to reflect Ryzen AI Embedded X199 specifications (AMD Maple CRB, 5.1 GHz CPU, 2.9 GHz GPU, with sustained operation at 45W TDP/sPPT/fPPT, STAPM disabled and 64 GB soldered LPDDR5X-8000), compared to an Intel Core Ultra X7 358H (MSI Prestige 16 Flip AI+ C3MTG MS-2622, 5.1 GHz CPU, 2.5 GHz GPU, PL1 = 45W, PL2 = 64W, MSI BIOS E2622IMS.10E, 32 GB LPDDR5X-8533, measured 30W OEM Sustained Power Limit). Results reflect the relative geomean of offscreen OpenGL frame rates from GFXBench 5.0.0 on Ubuntu 24.04, comparing measured AMD performance at 45W sustained power to 45W Intel performance projected from 30W sustained power measurements, using scaling factors derived from public benchmark data for the 358H. System manufacturers may vary configurations, yielding different results. Results may vary based on configuration, settings, usage, and other factors. (REX-003)
- Measurements and projections conducted by AMD as of May 2026 on an AMD Ryzen AI Max+ 395 processor (OPN 100-000002199) configured to reflect Ryzen AI Embedded X199 specifications (AMD Maple CRB, 5.1 GHz CPU, 2.9 GHz GPU, with sustained operation at 45W TDP/sPPT/fPPT, STAPM disabled and 64 GB soldered LPDDR5X-8000), compared to an Intel Core Ultra X7 358H (MSI Prestige 16 Flip AI+ C3MTG MS-2622, 5.1 GHz CPU, 2.5 GHz GPU, PL1 = 45W, PL2 = 64W, MSI BIOS E2622IMS.10E, 32 GB LPDDR5X-8533, measured 30W OEM Sustained Power Limit). Results reflect projected relative performance running the Unigine Heaven 4.0 Pro “Extreme” preset on Ubuntu 24.04, comparing measured AMD performance at 45W sustained power to 45W Intel performance projected from 30W sustained power measurements, using scaling factors derived from public benchmark data for the 358H. System manufacturers may vary configuration, yielding different results. Results may vary based on configuration, settings, usage, and other factors. (REX-004)
- Based on the OpenNav Robotics Workload Benchmark commissioned by AMD, published by Open Navigation LLC on July 23, 2026, as measured on the GMKtec EVO-X2 AI Mini PC AMD Ryzen AI Max+ 395 configured to reflect Ryzen AI Embedded X199 specifications (5.1 GHz CPU, 2.9 GHz GPU, 120W TDP, LPDDR5X-7500), vs. the NVIDIA Jetson AGX Thor Developer Kit: https://opennav.org/news/opennav-robotics-workload-benchmark (REX-018)
© 2026 Advanced Micro Devices, Inc. All rights reserved. AMD, the AMD Arrow logo, EPYC, Ryzen, and combinations thereof are trademarks of Advanced Micro Devices, Inc. CUDA, NVIDIA, Orin and Thor are trademarks of NVIDIA Corporation. Core and Intel are trademarks of Intel Corporation. CoreMark is a registered trademark of the Embedded Microprocessor Benchmark Consortium (EEMBC). Linux is the registered trademark of Linus Torvalds in the U.S. and other countries. OpenGL® and the oval logo are trademarks or registered trademarks of Hewlett Packard Enterprise in the United States and/or other countries worldwide. PassMark is a registered trademark of PassMark Software Pty Ltd. SPEC CPU and SPECrate are trademarks or registered trademarks of Standard Performance Evaluation Corporation (SPEC). Ubuntu and the Ubuntu logo are registered trademarks of Canonical Ltd. Vulkan is a trademark of the Khronos Group. Other product names used herein are for identification purposes only and may be trademarks of their respective owners.