Anyone who wanted to attend both Nvidia’s AI Summit and AMD’s Advancing AI 2024 event this week had to travel coast to coast, from Nvidia’s event in Washington, D.C., during the early part of the week to AMD’s in San Francisco during the latter part. At least, the abundance of news announcements from both companies–and the abundance of frequent flyer miles–were worth the trip.
After Nvidia earlier this week announced its NIM Agent Blueprints and other efforts aimed at accelerating agentic AI for enterprise customers, AMD announced shipping plans for its next-generation GPUs, unveiled a new EPYC data center CPU, and shared a peek at new Ryzen AI PC chips.
All of these announcements landed amid AMD’s ever-growing forecast for its AI market opportunity in the years to come. During a keynote speech at the San Francisco event, AMD chief Lisa Su said the company’s total addressable market for data center AI acceleration “will grow at more than 60% annually to $500 billion in 2028,” driven by AI demand that has been growing beyond expectations.
To target that opportunity in the early term, AMD has been ramping up shipments of its Instinct MI300X GPU, with the expectation that it will ship 500,000 of the AI accelerator chips during this calendar year, according to Su.
But, there are more AMD GPUs on the way. Su announced AMD Instinct MI325X accelerators, built on the company’s CDNA 3 architecture, are on track to begin shipping during the current quarter, and should be available from companies such as Dell Technologies, Eviden, Gigabyte, Hewlett Packard Enterprise, Lenovo, Supermicro, and others in the first quarter of next year.
With the MI325X, AMD is clearly targeting Nvidia’s H200 GPU. The MI325X features 256GB of HBM3E memory supporting 6.0 Tbps, offering 1.8X more capacity and 1.3x more bandwidth than the H200, AMD claimed, based on its own lab tests. The AMD Instinct MI325X also offers 1.3X greater peak theoretical FP16 and FP8 compute performance compared to H200, and can provide up to 1.3X the inference performance on Mistral 7B at FP162, 1.2X the inference performance on Llama 3.1 70B at FP83, and 1.4X the inference performance on Mixtral 8x7B at FP16 versus the H200, AMD claimed.
While discussing the incoming MI325X, Su also took the opportunity to provide a preview of the next GPU in line. As the first built on the company’s CDNA 4 architecture, Instinct MI350 series accelerators are on track to be available during the second half of next year. They will aim to deliver a 35x improvement in inference performance compared to AMD CDNA 3-based accelerators, and up to 288GB of HBM3E memory per accelerator.
New EPYC and Ryzen devices, too
Among other announcements from the AMD event, the company announced the availability of its fifth EPYC data center CPU, which AMD previously referred to as Turin. It uses the company’s Zen 5 core architecture, which provides up to 17% better instructions per clock (IPC) for enterprise and cloud workloads, and up to 37% higher IPC in AI and high performance computing (HPC) compared to the company’s Zen 4 architecture.
Offering core counts ranging from eight to 192, AMD said its EPYC 9005 Series processors in the 192-core CPU flavor deliver up to 2.7X the performance compared to competitors’ CPUs. New to the series is the 64-core AMD EPYC 9575F, which the company described as “tailor-made for GPU-powered AI solutions that need the ultimate in host CPU capabilities.”
Also leveraging the Zen 5 core architecture are AMD’s latest chips for AI PCs. The new Ryzen AI PRO 300 Series processors arrive as Intel also has unveiled new AI PC chips in a market still very much up for grabs.
AMD’s third-generation commercial AI mobile processors aim to ramp up AI-enhanced business productivity capabilities with Copilot+ features that include live captioning and language translation in conference calls and advanced AI image generators. The Ryzen AI PRO 300 offers up to three times the AI performance than the previous generation of Ryzen, along with greater security and manageability features, the company said.
Laptop PCs equipped with the top-of-stack Ryzen AI 9 HX PRO 375 support up to 40% higher performance, and up to 14% faster productivity performance than Intel’s Core Ultra 7 165U, AMD claimed. With the addition of XDNA 2 architecture powering the integrated NPU, the PRO 300 Series also offers 50+ NPU TOPS of AI processing power.
AMD’s stock price had been climbing in recent weeks with rumors of big announcements targeting both Nvidia and Intel expected to come out of this week’s event. Su delivered the goods, but AMD’s share price was sinking Thursday afternoon, down more than $7 to $163.86 within an hour after the keynote was finished.
RELATED: Intel pushes AI PC desktop silicon with focus on gamers