Free tools Windows power users keep installed
One-click scans. No signup required.
At COMPUTEX 2023, NVIDIA CEO Jensen Huang argued that generative AI would drive a new wave of accelerated-computing infrastructure: large-memory AI systems, faster networking and more adaptable server designs. The announcements framed an opportunity for NVIDIA and its partners, but the keynote figures were company claims reported at the time—not independent benchmarks or proof that announced systems were then shipping.
What was Huang’s central argument?
Huang presented accelerated computing and artificial intelligence as a shift in how computing would be built and used. Generative AI, large language models and recommender systems, he argued, would create demand for data centers designed around accelerated workloads rather than conventional CPU-only computing. As quoted in Nitin Dahad’s EE Times event coverage, Huang said, “Accelerated computing and AI mark a reinvention of computing.”
The business case was that AI capacity depends on more than a single processor: compute, memory, interconnects and networking have to work together. NVIDIA’s announcements addressed those layers through the DGX GH200 supercomputer, Spectrum-X networking and MGX modular server architecture. Dahad’s report also identified digital factories and autonomous mobile robotics among the keynote’s broader themes, but did not provide enough detail to assess specific announcements in those areas.
EE Times reported about 3,500 in-person attendees. Its May 30, 2023 coverage is a trade-publication account of the keynote, not an official transcript, independent benchmark report or current product catalog. The specifications, performance claims and future plans below should be read in that context.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
#1 Best Overall
- AI Performance: 767 AI TOPS
- OC mode: 2632 MHz (OC mode)/ 2602 MHz (Default mode)
- Powered by the NVIDIA Blackwell architecture and DLSS 4
- Axial-tech fan design features a smaller fan hub that facilitates longer blades and a barrier ring that increases downward air pressure
- A 2.5-slot design maximizes compatibility and cooling efficiency for superior performance in small chassis
What did NVIDIA announce for AI data centers?
| Announcement | Role in NVIDIA’s argument | What the event coverage reported |
|---|---|---|
| DGX GH200 | Connect many accelerators into a large-memory AI system | A supercomputer design combining Grace Hopper superchips through NVLink switching |
| Spectrum-X | Build Ethernet networking for AI workloads and multi-tenant clouds | A platform pairing Spectrum-4 Ethernet switches with BlueField-3 DPUs |
| MGX | Make it easier for manufacturers to create different server configurations | A modular reference architecture for AI, high-performance computing and Omniverse workloads |
DGX GH200: linking Grace Hopper systems
EE Times reported NVIDIA’s announcement of a DGX GH200 configuration combining 256 GH200 superchips using NVLink switch technology, so the GPUs could work together as a single system. NVIDIA’s reported headline specifications were 1 exaflop of performance and 144 terabytes of shared memory. The coverage described each Grace Hopper superchip as combining an Arm-based Grace CPU with a Hopper H100 GPU, connected using NVLink-C2C.
Those figures describe the announced system as reported in 2023; they are not independent measurements or a guarantee of performance for a particular workload. Huang said Google Cloud, Meta and Microsoft would be first to gain access to explore generative-AI workloads. NVIDIA also intended to provide the design as a blueprint to cloud providers and hyperscalers. “Access to explore” and an intended blueprint do not establish deployment, present availability or current customer use.
Rank #2
- Powered by the NVIDIA Blackwell architecture and DLSS 4
- Powered by GeForce RTX 5070 Ti
- Integrated with 16GB GDDR7 256bit memory interface
- PCIe 5.0
- WINDFORCE cooling system
Spectrum-X: networking for AI clouds
NVIDIA described Spectrum-X as an accelerated networking platform for Ethernet-based AI clouds, combining Spectrum-4 Ethernet switches and BlueField-3 DPUs. EE Times reported that the platform was designed to offer multi-tenant performance isolation, visibility into bottlenecks and fabric validation. Dell Technologies, Lenovo and Supermicro were named as companies using Spectrum-X; the report does not establish that every related product was shipping at the time or remains available now.
The coverage gave a scale of 256 200Gb/s ports on one switch, or 16,000 ports in a two-tier leaf-spine topology. NVIDIA claimed 1.7× better overall AI performance and power efficiency. That is the company’s reported claim, not a general result established by an independent comparison. The report did not supply a methodology that would let readers apply the multiplier across different networks, workloads or configurations.
Rank #3
- Powered by the NVIDIA Blackwell architecture and DLSS 4. System Requirements: Minimum 850W PSU with 16-pin 12V-2x6 (12VHPWR) connector required. Verify before purchasing.
- Military-grade components deliver rock-solid power and longer lifespan for ultimate durability. Compatibility: 348mm (13.7") length, 3.6 slots, 4.3 lbs. Confirm case clearance and slot spacing. GPU bracket included.
- Protective PCB coating helps protect against short circuits caused by moisture, dust, or debris
- 3.6-slot design with massive fin array optimized for airflow from three Axial-tech fans
- Phase-change GPU thermal pad helps ensure optimal thermal performance and longevity, outlasting traditional thermal paste for graphics cards under heavy loads
MGX: a modular route to different servers
MGX was presented as a reference architecture that let manufacturers choose GPU, DPU and CPU elements around a base server design. NVIDIA said it could support more than 100 server variations for AI, high-performance computing and Omniverse workloads. The named expected adopters were ASRock Rack, ASUS, GIGABYTE, Pegatron, QCT and Supermicro.
NVIDIA claimed MGX could cut development costs by up to three-quarters and development time by two-thirds, to six months. These were projected benefits, not independently verified outcomes. EE Times described QCT and Supermicro as first to market, with designs expected in August 2023. It also named two COMPUTEX announcements: Supermicro’s ARS-221GL-NR with a Grace CPU superchip, and QCT’s S74G-2U with the GH200 Grace Hopper superchip.
Rank #4
- Powered by the NVIDIA Blackwell architecture and DLSS 4
- Powered by GeForce RTX 5060
- Integrated with 8GB GDDR7 128bit memory interface
- PCIe 5.0
- WINDFORCE cooling system
SoftBank planned to deploy MGX in hyperscale data centers in Japan and dynamically allocate GPU resources between generative AI and 5G applications. That was a plan reported in 2023, not confirmation of a completed or current deployment.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.How did Huang compare GPU and CPU server economics?
To argue that data centers should optimize for dense computing, Huang offered a keynote illustration comparing what he said the same $10 million could buy:
Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Fix the driver behind crashes, sound loss and screen glitches3Clear out junk files and repair common Windows errorsBest Value
- Powered by the NVIDIA Blackwell architecture and DLSS 4 OC mode: 2640MHz/Default mode: 2610MHz (Boost Clock)
- Military-grade components deliver rock-solid power and longer lifespan for ultimate durability
- Protective PCB coating helps protect against short circuits caused by moisture, dust, or debris
- 3.125-slot design with massive fin array optimized for airflow from three Axial-tech fans
- Phase-change GPU thermal pad helps ensure optimal thermal performance and longevity, outlasting traditional thermal paste for graphics cards under heavy loads
| Keynote scenario | Servers | Energy figure cited | AI output cited |
|---|---|---|---|
| GPU servers | 48 | 3.2 GWh | 44 LLMs |
| CPU servers | 960 | 11 GWh | 1 LLM |
These are Huang’s figures as relayed by EE Times, not a neutral, independently verified benchmark. The coverage does not specify the workload definitions, measurement period, hardware configurations, utilization assumptions or what “deliver” means in this comparison. It therefore cannot establish a universal cost, energy or throughput advantage for GPU servers. The example shows how NVIDIA framed the investment case, but a real infrastructure decision requires workload-specific performance, acquisition and operating costs, power consumption, network behavior and service requirements.
What did the keynote establish—and what did it leave open?
The keynote laid out NVIDIA’s 2023 view of the AI infrastructure opportunity: connect large pools of accelerated compute, keep them supplied by high-speed networking and give system makers a reusable way to assemble servers for different demands. The announcements and named partners indicate the ecosystem NVIDIA was targeting at that moment.
They do not, on their own, show which announced systems reached general availability, how they performed in independent testing, whether the cited efficiency improvements held across workloads, or what products and deployments exist today. For readers assessing the claims, the key distinction is between architecture and intent, on one hand, and measured results or confirmed deployment, on the other.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
