AMDLimited deployment
Software that moves cold memory pages from DRAM to flash and uses a CPU model to bring predicted pages back before an application asks for them. AMD's October 2026 white paper describes it as software-only, with no GPU required.
Latest update Nov 14, 2026
01
Smerdon told The Register on Nov. 11, 2026 that 145 systems are deployed and estimated about $5 million of avoided DRAM cost. His PrimeTime figures were 95 percent at half DRAM and 80 percent at one-quarter DRAM. The October 2026 white paper reports about 87 percent on a different 1.5 TB to 768 GB PrimeTime test dated Oct. 1, 2025. License sales are a later phase, and HPE is named only as an OEM in the first readiness phase.
02
AMD's first rack-scale AI system: 72 Instinct MI455X GPUs in 18 Open Rack Wide compute trays, 18 6th Gen EPYC Venice CPUs, Pensando Vulcano 800 AI NICs, UALink-over-Ethernet scale-up and Ethernet scale-out, running ROCm. A rack is specified at up to 2.9 exaflops FP4, 1.4 exaflops FP8, 31 TB of HBM4 and 1.7 PB/s of memory bandwidth. AMD claims up to 30% more inference tokens per dollar, 15% more peak FP4, and 50% more HBM than Nvidia's Vera Rubin NVL72.
AMD's data-center GPU line. MI350-series (CDNA 4), led by MI355X (288 GB HBM3E, 8 TB/s, up to 10.1 PFLOPS MXFP4, 1,400 W TBP), is the generation shipping into 2026 and the volume driver of Instinct revenue. MI400-series (CDNA 5), launched July 23, 2026, splits into MI455X for frontier AI and AI factories (432 GB HBM4, 23.3 TB/s, up to 40 PFLOPS FP4, 320 billion transistors) and MI430X for HPC and sovereign AI (up to 288 TFLOPS FP64). ROCm and ROCm.ai are the software stack.
6th Generation EPYC 9006-series server CPUs on Zen 6, first shown at Advancing AI on July 23, 2026. AMD said on Nov. 10, 2026 that the SP7 processors are generally available and that systems are shipping from leading server vendors. The SP7 family reaches up to 256 cores and 512 threads, and up to 5.3 GHz on 96-core parts. SP8, 9006X SP7, and 9006 LP variants are still planned. Microsoft said Azure HDv2 is in preview. Meta is validating the platforms for deployment at scale. Comparisons with Nvidia Vera, including about 3.3 times the agents per CPU watt, are AMD estimates.
AMD's client CPU and APU brand, from desktop AM5 chips through Ryzen AI notebooks and Ryzen AI Halo agentic PCs. Client revenue was a record $10.6 billion in 2025 and $3.1 billion in Q2 2026. Recent additions include Ryzen AI PRO 400 Series, Ryzen AI Halo developer systems, 3D V-Cache on Ryzen PRO workstations, and a commitment to support Socket AM5 through 2029.
RDNA 4 desktop GPUs launched in 2025 as RX 9070 XT ($599 SEP) and RX 9070 ($549 SEP), later joined by RX 9070 GRE and RX 9060 XT. FY2025 gaming revenue was $3.9 billion, but Q2 2026 gaming fell to $779 million (-31% YoY) on lower semi-custom console SoC sales and higher board prices that Lisa Su said weighed on demand.
Open rack blueprint for agentic AI, separate from Helios. The product page lists a 48OU liquid configuration at 36,864 Zen 6 cores and a 44OU liquid configuration at 32,768 Zen 6 cores, both on 256-core EPYC 9006 SP7 processors. An enterprise configuration can add up to 16 Instinct MI350P GPUs as two independent eight-GPU nodes, with up to 102 Tbps of networking. The same page's specification line still says up to 32,768 Zen 6 cores. On Nov. 10, 2026 AMD said some configurations reach up to 73,728 threads. That equals the 36,864-core 48OU line at two threads per core, the SMT ratio AMD states for these EPYC processors. AMD said in October that partner configurations were planned for November 2026. The Nov. 10 release describes the racks and does not say a customer system has shipped.
Helios GPU introduced by AMD on Nov. 12, 2026 with 288 GB of HBM4. AMD says it has the same core compute, memory bandwidth, and scale-up and scale-out connectivity as the MI455X, which the company says offers more memory for larger models, longer contexts, or higher concurrency. The product page lists CDNA 5, a TSMC 2 nm and 3 nm process, 23.3 TB/s peak memory bandwidth, 40.3 PFLOPs peak OCP MXFP4, 3.6 TB/s UALoE scale-up, 600 GB/s UALink scale-out, and direct liquid cooling.
03
SpaceX
Partially reusable two-stage medium-lift launch vehicle. The world's most frequently launched orbital rocket.
SpaceX
Heavy-lift launch vehicle consisting of a strengthened Falcon 9 core with two additional boosters.
SpaceX
Reusable spacecraft for cargo and crew transport to the International Space Station.
SpaceX
Fully reusable super-heavy-lift launch system designed for missions to the Moon, Mars, and beyond. V3 (Block 3) variant with Raptor 3 engines is the production version.