Jon Peddie

California

President at Jon Peddie’s Tech Watch

Chases pixels - finds gems

Articles

Modern GPUs are an incredible bargain

1 week ago | jonpeddie.com | Jon Peddie

Nvidia launched the first GPU in 1999 with the GeForce 256, introducing hardware T&L to off-load CPU workloads. The 2001 GeForce 3 added programmable shaders, starting a new era in graphics. Shader counts soared from dozens to tens of thousands by 2025, evolving through unified shader models and AI acceleration. Nvidia’s 2017 Tensor Cores marked a pivot to AI, now central in gaming and data centers. Despite massive performance leaps, GPU prices have risen modestly.
AMD’s Series 350 GPUs

1 week ago | jonpeddie.com | Jon Peddie

At its Advancing AI event, AMD launched the MI350 series GPUs, built on the new CDNA 4 architecture. These accelerators challenge Nvidia’s B200 on cost-performance for small-to-medium AI model inference but are not a rack-scale competitor to the GB200 NVL72 for large-scale training. The architecture uses eight compute chiplets and 288GB of HBM3E memory. Its key innovation is support for new, reduced-precision data types like FP4, dramatically increasing AI throughput and efficiency.
Advancing AI at AMD

2 weeks ago | jonpeddie.com | Jon Peddie

AMD previewed its next-generation AI GPU, the MI400 series and demonstrated how the MI450 would compare to the current MI350. The MI350 series, including Instinct MI350X and MI355X GPUs and platforms, will be available by the third quarter. AMD also showed its next-generation rack system called Helios. Several partners joined CEO Lisa Su on stage to support the open AI ecosystem. AMD executives stated they can beat Nvidia’s B200 on price, performance, and power efficiency.
ROG Xbox Ally handhelds revealed

2 weeks ago | jonpeddie.com | Jon Peddie

Microsoft and Asus are releasing two handheld gaming devices, the ROG Xbox Ally and Ally X, during holiday 2025. These Windows 11-based devices aim to combine the Xbox experience with PC gaming freedom, allowing access to games from Xbox, Battle.net, and other storefronts via an aggregated library. Both feature a 7-inch FHD 120 Hz display and AMD Ryzen processors.
Khronos announces Vulkan Video decode VP9 extension

2 weeks ago | jonpeddie.com | Jon Peddie

Khronos announced Vulkan Video extensions, providing vendor-independent access to GPU video decoding/encoding. The latest Vulkan specification adds VP9 decoding, a widely used, royalty-free codec. This completes the planned decode extensions, allowing developers to build platform-independent accelerated decoding pipelines for all prominent contemporary codecs. The VP9 extension, similar to H.264/H.265/AV1, allows querying capabilities and specifying parameters.
VeriSilicon’s AI GPU IP for automotive and edge

2 weeks ago | jonpeddie.com | Jon Peddie

VeriSilicon developed a GPGPU-AI computing IP core for AI computational work, scalable and suitable for automotive electronics and edge servers. This design combines programmable parallel computation with a dedicated AI accelerator for intensive tasks like LLM inference and real-time decision-making, while managing thermal output and power use. Based on re-engineered GPU designs, it features an integrated AI accelerator, sparsity awareness, and mixed-precision computing.
How does FSR 4 compare to FSR 3 and DLSS 4?

3 weeks ago | jonpeddie.com | Jon Peddie

AMD’s FSR 4 marks a major leap over FSR 3, shifting to an AI-based upscaling method that improves image quality and reduces artifacts. While FSR 3 offered broad GPU support, FSR 4 is limited to AMD’s RDNA 4 GPUs. Nvidia’s DLSS 4 enhances its AI-powered features with transformer models and Multi Frame Generation, offering top-tier visuals but requiring RTX 50 GPUs for full capabilities.
Could an RTX 5060 or 5090 be used in a data center?

3 weeks ago | jonpeddie.com | Jon Peddie

Nvidia’s Blackwell GPUs for AI and gaming share a core architecture but differ significantly in design and configuration. These changes optimize each version for its specific purpose—AI workloads or real-time gaming. As a result, they are not identical chips but distinct implementations of the same blueprint, featuring unique memory systems, power delivery, and core layouts.
HBM competitor from Intel JV?

3 weeks ago | jonpeddie.com | Jon Peddie

SoftBank and Intel are collaborating through newly established Saimemory to develop an AI-optimized memory chip that consumes 50% less power than existing technologies. The innovative stacked DRAM design will differ from current high-bandwidth memory solutions, incorporating Intel’s technology and Japanese university patents, including from University of Tokyo.
Game over?

3 weeks ago | jonpeddie.com | Jon Peddie

Two decades ago, AMD fell behind Nvidia in the graphics business. AMD clawed its way back from the brink, but even so, was unable to keep up with Nvidia in the GPU market. Then, AI came along, and Nvidia left AMD in its shadow. What would happen if they exited that market? Jon Peddie digs out his crystal ball….In 2005, AMD bought ATI to prop up sagging Bulldozer CPU sales. It was a good move at the time and helped AMD get some badly needed revenue.