{"slug": "cerebras-unveils-low-latency-ai-workload-speed-demon-rack-scale-platform", "title": "Cerebras unveils low latency AI workload speed demon rack-scale platform", "summary": "Cerebras Systems unveiled its rack-scale CS-4 platform powered by three WSE-3 Turbo chips, claiming performance up to 30 times faster than GPU-based rivals. The CS-4 delivers 750 petaflops of sparse FP16 AI compute and 7.2 terabits per second of I/O bandwidth, with first shipments starting this quarter. CEO Andrew Feldman said the platform fundamentally reshapes product experiences by enabling industry-leading speeds on the largest frontier models.", "body_md": "Cerebras has unveiled its own rack-scale platform powered by next-gen wafer-scale AI chips, claiming performance up to 30-times faster than graphics processing unit (GPU)-based rivals.\n\nThe move sees Cerebras shift away from a standalone appliance approach to a full-scale multi-wafer rack architecture. Designed to house its new WSE-3 Turbos and future generations of Cerebras chips, the rack-scale platform provides double the I/O (input/output) bandwidth to keep AI workloads running at high speeds. It also features a “wafer-scale backpack.” Mounted at the rear, it’s attached vertically to the power array, providing direct liquid cooling around the chips.\n\nNexus is the architecture name, and its first outing is dubbed CS-4. Under the hood, it features three WSE-3 Turbos – each of which packs four trillion transistors and 900,000 AI-optimized cores – and sees improved wafer-to-wafer communication compared to previous generations. Cerebras claims interconnection latency between Turbos is as low as just two milliseconds. With such low latency, the chip firm said its hardware can support more than 1,000 tokens per second for AI models exceeding 10 trillion parameters.\n\nOn the networking side, Cerebras makes use of RoCE v2 (Remote Direct Memory Access, RDMA over Converged Ethernet version 2) to bridge its Wafer-Scale Engines. The standards-based protocol provides an open option to communicate directly with heterogeneous host systems and central processing unit (CPU) storage arrays. The standards approach also opens the doors for users to employ alternative vendor hardware from the likes of Nvidia or AMD to act as a “prefill engine” – where the rival accelerator handles incoming prompts and readies the underlying model, before handing it off to the Cerebras system to perform ultra-low-latency decoding and generate the response. That ties in with deals the chip firm has made with the likes of [AMD](https://www.sdxcentral.com/news/amd-cerebras-partner-on-joint-helios-rack-scale-ai-inference-platform/), so that operators could pair its Wafer-Scale Engines with systems inside platforms like [Helios](https://www.sdxcentral.com/news/amd-cerebras-partner-on-joint-helios-rack-scale-ai-inference-platform/).\n\nCS-4 also makes use of a switch-free interconnect known as “Direct Wafer Links” to connect the new Turbo chips. Little technical information was divulged on the proprietary tech, though Cerebras said that the wafer chip’s I/O module can support open ecosystem connectivity.\n\nAt a system level, CS-4 provides 750 petaflops of sparse floating-point 16 (FP16) of AI compute and 7.2 terabits per second of I/O bandwidth. Engineers won’t have long to wait to get their hands on this speed machine either, with the first shipments of CS-4 set to commence this quarter.\n\n“In AI, speed is productivity,” said Andrew Feldman, CEO and co-founder of Cerebras. “Historically, fast inference meant using smaller and less capable models. Cerebras CS-4 delivers industry-leading speeds on the largest frontier models, fundamentally changing the paradigm. Every aspect of the design has been optimized to deliver the highest speeds with massive throughput. With the CS-4, AI is so fast that it fundamentally reshapes product experiences.”", "url": "https://wpnews.pro/news/cerebras-unveils-low-latency-ai-workload-speed-demon-rack-scale-platform", "canonical_source": "https://www.sdxcentral.com/news/cerebras-unveils-low-latency-ai-workload-speed-demon-rack-scale-platform/", "published_at": "2026-08-19 10:44:29+00:00", "updated_at": "2026-08-19 11:10:40.418720+00:00", "lang": "en", "topics": ["ai-infrastructure", "ai-chips", "artificial-intelligence"], "entities": ["Cerebras Systems", "WSE-3 Turbo", "CS-4", "Andrew Feldman", "Nvidia", "AMD", "Helios"], "alternates": {"html": "https://wpnews.pro/news/cerebras-unveils-low-latency-ai-workload-speed-demon-rack-scale-platform", "markdown": "https://wpnews.pro/news/cerebras-unveils-low-latency-ai-workload-speed-demon-rack-scale-platform.md", "text": "https://wpnews.pro/news/cerebras-unveils-low-latency-ai-workload-speed-demon-rack-scale-platform.txt", "jsonld": "https://wpnews.pro/news/cerebras-unveils-low-latency-ai-workload-speed-demon-rack-scale-platform.jsonld"}}