The SC23 event gathered a record-breaking 14,295 attendees and 438 exhibitors, which reflects the world’s expanding interest in HPC. 89 journalists and analysts were registered, covered the event, and delivered news about the conference to key audiences.

“I Am HPC” was the SC 23 theme, which served to remind everyone that they are an equal and welcome contributor to the thriving HPC (High Performance Computing) community.

We registered a very good quality of visitors at Tachyum booth, met our key partners like Eviden, Chenbro or CGG and created new contacts with potential partners and suppliers like Malico, Micas Networks, Picotest, Nvent, Xscape Photonics, CPC or Omni Services.

62^nd edition of TOP500

During the conference, the 62^nd edition of the TOP500 list of the world’s fastest supercomputers was announced. TOP500 serves as a “Who’s Who” in the field of High Performance Computing. It started as a list of the most powerful supercomputers in the world and has evolved to a major source of information about trends in HPC. The list is compiled using the Linpack benchmark. LINPACK measures a system’s floating-point computing power by solving a random, dense system of linear equations to determine performance and accuracy.

A detailed analysis of the TOP500 was presented by Dr. Erich Strohmaier, cofounder of the TOP500 list, at BoF (Birds of Feather) session.

The Frontier system at the Oak Ridge National Laboratory, Tennessee, USA remains the No. 1 system on the TOP500 and is still the only system reported with an HPL (High-Performance LINPACK) performance exceeding one Exaflop/s. No. 2, The Aurora system at the Argonne Leadership Computing Facility, Illinois, USA is currently being commissioned and will at full scale exceed Frontier with a peak performance of 2 Exaflop/s. It was submitted with a measurement on half of the final system achieving 585 Petaflop/s.

Newly listed as No. 3 The Eagle system installed in the Microsoft Azure cloud in the USA achieved an HPL score of 561 Petaflop/s. The fastest European supercomputer the LUMI system at EuroHPC/ CSC in Finland has been further upgraded and is now listed as No. 5 with HPL score 380 Petaflop/s.

Dr. Strohmaier presented also a list of the systems based on the HPL-MxP (formally HPL-AI) benchmark, which measures performance for mixed-precision calculation. No. 1 Frontier demonstrated 9.95 Exaflop/s.

Historical data on performance increases and core-counts illustrated the progression of the processor performance plateau. In the end of the session, Dr. Strohmaier said that it is “unlikely to achieve 10 Exascale by the end of the decade unless we fundamentally change ‘business’”. It confirms that new technology, like the world’s first universal processor Prodigy, is essential for the development of further progress.

We are looking forward to SC24 which will take place in Atlanta at the Georgia World Conference Center. The SC24 conference theme is “HPC Creates” and will focus on the enormous generative power that supercomputing affords the modern world.

Tachyum released a white paper that details how its Prodigy ATX Platform will democratize AI for those who may not normally have access to sophisticated AI models. The Prodigy ATX Platform allows everyone to run cutting-edge AI models for as low as $5,000.

Built from the ground up to provide leading-edge AI features that address the emerging demand for AI across a wide range of applications and workloads, Prodigy’s AI subsystem incorporates innovative features that deliver the high performance and efficiency required of AI environments. The white paper shows how a single Prodigy system with 1 Terabyte (TB) of memory can run a ChatGPT4 model with 1.7 trillion parameters, whereas it requires 52 NVIDIA H100 GPUs to run the same model at significantly higher cost and power consumption.

“Generative AI will be widely used far faster than anyone originally anticipated. In a year or two, AI will be a required component on websites, chatbots and other critical productivity components to ensure a good user experience. Prodigy’s powerful AI capabilities enable LLMs to run much easier and more cost-effectively than existing CPU+GPGPU-based systems, empowering organizations of all sizes to compete in AI initiatives that otherwise would be dominated by the largest players in their industry.”

Dr. Radoslav Danilák

Tachyum released a white paper detailing how to use 4-bit Tachyum AI (TAI) and 2-bit effective per weight (TAI2) formats in Large Language Models (LLM) quantization without accuracy degradation. Tachyum hardware also enables workable LLM with 1-bit per weight with higher degradation than TAI2 and its AI scientists are continuing to further improve performance to reduce degradation as Tachyum looks to bring it to mainstream.

Tachyum addresses massive LLM models with capabilities that have dramatically increased by more than a thousand times over the past few years. Examples of these increases include ChatGPT-3.5 LLM with 175 billion parameters, the PALM LLM with 530 billion dense parameters and the Switch Transformer with 1.6 trillion sparse parameters.

“By combining TAI 4-bit and effective 2-bit weights with FP8 per activation, we are capable of quantizing LLMs without much accuracy degradation. Our techniques avoid expensive multiplication while simultaneously reducing the size of the model by 4x to 8x, enabling generative AI models that can be applied in use cases from complex language modelling tasks, text generation, drug and chip design, few-shot learning and reasoning to protein sequence modelling. Whole new avenues of calculations can be opened with Tachyum AI.”

Dr. Radoslav Danilák

Tachyum announced that the FreeBSD operating system is now running on the Prodigy FPGA prototype as part of efforts to ensure that customer requirements are met prior to tape-out of the world’s first universal processor. This announcement follows Tachyum successfully running the Linux OS as part of its emulation system.

Tachyum has completed vector-based High-Performance LINPACK (HPL) testing using 1kb vectors on the Prodigy FPGA.